Fundamentals of Process Control by Top ControlFundamentals of process control TABLE OF CONTENTS 0.0... 1. INTRODUCTION.. 1 1.1, WHATIS A PROCESS? 3 11d. Base Elements of a Process. ante eae vee 1.2. CoMPoNENTS OF 4 ContROL Loop 3 6 9 9 1.2.1. Control Objectives 1.3. CONTROLLER Basics, 1.3.1, Human controller. 1.3.2, Automatic controller eae tana 10 1.3.3, Examples of simple processes... ae ul 1.3.4, Closed Loop - Feedback Control... eee aaa 2 Open Loop - Feedforward ContrOl.swiu.o-s. ee 4 1.4. TUNING PID CoNTROLLERS ~ A COMPROMISE : i : sn AS 1.5, THE NEED FOR PROCESS CONTROL. ae : 16 1.6, DEFINITIONS AND FORMULAS : eri sn 16 17 EXERCISES 7 18 Notes, 18 ©TOP Control 1999 iFundamentals of process control 2. PROCESS DYNAMICS (OPEN LOOP). 2.1, SIMPLE PROCESS ~ SELF REGULATING PROCESS. 2.11 First order systems. 2.2. ZIEGLER-NICHOLS MODEL. 2.21. What is a model?, 2.2 The equivalent model 2.3. Natural frequency (closed loop). 2.3, INTEGRATING PROCESS. 2.4, RUNAWAY PROCESS, 2.5. MovELs, 2.5.1. Other characteristics. 2.6. DEFINITIONS AND FORMULAS 2.7, EXERCISES 2.8, COMPUTER EXERCISES 2.9. NoTEs TOP Control 1999 a 23 32 32 33 34 35 38. 39 40 wal 42 45 47Fundamentals of process control 3. PID CONTROLLER. 3.1, INTRODUCTION. 3.2. ON/OFF conTROL. : 3.3, PROPORTIONAL CONTROL ACTION 54 3.3.1. Introduction... : 54 3.3.2. The Watt regulator 55 3.3.3.4 simple level controller. a 56 3.3.4, Proportional band. 58 3.3.5, Proportional band and gain. 59 3.3.6. Controller action... : sn a 3.3.7. Block diagram. : iss vo a 3.4, INTEGRAL CONTROL ACTION (RESET). 63 3.4.1. Integral control a 4 3.4.2, Integral control respOmse.rs.n.nn 6 3.4.3, Pl control. o 66 3.4.4, Integral time, integral gait... 7 3.5, DERIVATIVE CONTROL ACTION (RATE, PRE-ACT ) 68. 3.5.1. Derivative control... a 68 3.5.2. Derivative control response, a i 70 3.5.3, Derivative time aes a 3.3.4, Effects of each control mode. 72 3.5.5. Comparison i a 73 3.6, PID CONTROL wenn oi . "4 3.6.1. Introduction iti 74 3.62. Controller structures (ldeal, Series, Parallel) 74 3.6.3. Comparison peaettiat . 78 3.6.4. Pl controller. nn aaah 79 3.6.5. Response of an ideal controller... . aie 80 3.7, FILTERING... seven : eaeiSt 3.8, OTHER FEATURES IN APID CONTROLLER 83 3.8.1. Limitation of derivative gain. a seen, 8B 3.8.2. Derivative action and setpoint changes. a a 84 3.8.3. Integrator wind-up (anti reset wind-up algorithm) a 83 3.8.4. Manual/Automatic transfer 86 3.8.5. Full PID controller. aie a 86 3.9. DEFINITIONS AND FORMULAS. 87 3.10. EXERCISES 89, 3.11. COMPUTER EXERCISES. : i renee: 9B 3.12, Notes. oe ia 94 ©TOP Control 1999 iitFundamentals of process control 4. TUNING PID CONTROLLERS. 4.1. TUNING A CONTROLLER... 4.1.1. Basic concepts. 41.2. The closed loop 4.1.3. Eliminating the error using the error 4.2. Goopcowrro.. 4.2.1. What is good control? 4.2.2, Performance criteria 4.3. OPEN LOOP AND CLOSED LOOP RESPONSE 43.1. Load change 4.3.2 Setpoint change 43,3, Observations 4.4, DRIVING AN AUTOMOBILE AN ANALOGY. 441. Dead time 4.5, TUNING: A COMPROMISE. 4.6 TUNING METHODS 46.1. Tuning method 4.6.2, Ultimate eyeling method (automatic mode) 4.6.3, Trial and error method (automatic mode) 4.6.4. Open loop method (manual mode) 4.6.5. Using a software package, 4.7. NOISE, DISTURBANCES AND FILTERING. 4.7.1. Noise and disturbance. 47.2. Choosing the proper filter. 4.8. ULTIMATE CYCLING METHOD. 4.8.1. Time needed. 4.8.2, Method. 4.8.3. Example 4.9, TRIAL AND ERROR METHOD, 49.1, Time needed. 4.9.2, Method, 4.9.3, Example 4.10. OPEN LooP METHOD 4.10.1. Time needed. 4.10.2. Method. 4.10.3. Formulas... 4.11, AUTOTUNE, PRETUNE, SELFTUNE, 4.12. TUNING WITH SOFTWARE SUPPORT. 4.13. SuMMaRy. 4.13.1. Comparison of tuning criterion. 4.14, DEFINITIONS AND FORMULAS. 4.15, EXERCISES 4.16. COMPUTER EXERCISES. 4.17. Notes. 100 101 103 103 103 103, 104 105 107 108, 108 108 108 108 109. 110 110 un 113 113 13 15 vol 6 16 16 18. 121 121 121 121 125 126 127 127 132 135 140 144 ©TOP Control 1999Fundamentals of process control 5. PROCESS PROBLEMS, 3.1. PROCESS VARIATION 145 M47 5.2. VARYING TIME CONSTANT. be 5.2.1. Variation with process. pA 5.2.2. Asymmetrical Processes _ 3. VARYING DEAD TiMe. ia 54. INVERSE RESPONSE. iy 3.5. SATURATION tie 153 5.6. HYSTERESIS, BACKLASH AND DEADBAND, 135 5.7. Sniction 5.8. NON-LINEAR PROCESS GAIN, 38.1 Control Valves characteristics and problems. : i 159 582. Solutions for non-linear process gain 162 3.8.3. Applying a characterizer 163 5.9. Noise. fl eat 165 10. RECOMMENDED TEST AND SEQUENCE. 167 5.10.1. Prior to connecting to the signal... 167 3.102 Testing the equipment and current tuning parameters 167 5.103. Testing the Process ae 168 5.104. Tuning and testing the results 169 5.10.5. Recommended test summary. 170 S11. SUMMARY ae 11 211. How to recognize problems in automatic v1 5.11.2. Acceptable limits. o 173 5.12. DEFINITIONS AND FORMULAS. 174 5.13. EXERCISES : 176 5.14, ComPuTER EXERCISES 180 5.15, Nores 2 185 ‘©TOP Control 1995 v \Fundamentals of process control 6. ADVANCED CONTROL STRATEGIE: 6.1. INTRODUCTION ~ A 100P IS RARELY ALONE... 6.2, Mu.mioo?...... 6.2.1. Interacting Loops. 6.2.2. Synchronising Loops 6.3, CascabE Loops 6.4, ADDING FEEDFORWARD 6.5, SPLITRANGE. 6.6. Coarst-Fine ContRot 6.7, ADVANCED CONTROL, 62.1. Special Algorithms ~ P'ID, PED, and others 6.7.2. Sith Predictor su. 6.7.3. Internal Model Controller (Model Predictive) 6.7.4. Fuzzy Logie 6.2.5. Newral Nenworks 6.7.6, Multivariable Controllers 6.7.7. Comparison of advanced controllers 6.8, DEFINITIONS AND FORMULAS 6.9. EXERCISES 6.10, ComPUTER EXERCISES 6.11. Nores ©TOP Control 1999 viFundamentals of process control 7. PID CONTROLLER IN PLCS 7.1, INTRODUCTION. f : 7.2. THE IMPORTANCE OF SAMPLING TIME ON THE PID CALCULATION. 7.2.1. Analog and diaital. : 7 7.2.2. Sampling at constant rate 7.2.3. The PLC scan cycle. 7.3, KNOWN ISSUES WITH PLC's, 7.3.1, Misleading manuals. 7.3.2. PID sampling time or update time 7.3.3, Bump-less transfer. 7.3.4, Engineering units. 7.3.5, Derivative 2.3.6. FIMEPS oo 7.3.7. Scaling / Output Limiting. 2.3.8. Multipliers, integer numbers. 7.3.9. Algorithm, 7.3.10. Anti-reset windup (ARW) 7.4, TERMINOLOGY 7.5, NoTEs, ©TOP Control 1999 viiFundamentals of process control Appendix A - ISA symbols.. ot A. Diagrams cee = eee : 231 AQ. ISA standard. : ae se 231 A2L General 231 A22. Example : Level control... : 232 A3._ Functional identification...” : : 233 AB.L The first letter. 233 A32. The other letters 234 AA. Location : : : 235 A4.1 Discrete instruments seen B33 A4.2. Distributed control and shared display “ 236 A43. Multiple functions... 236 A44. Summary. : 237 AS. Line symbols i sone 238 AG, Graphic symbols... : 239 AT. Graphic symbols for process equipments 240 AS SAMA 241 A9. Example : son 242 A9.1. Pictorial diagram. : ‘ vn 242 A92, PAID... as sens 24D A93. SAMA toa A9A. Wiring diagram, Instrument loop diagram 244 AO. Exercises : 245 Appendix B : Definitions and formulas.. 47 Appendix C : Summaries. 259 Appendix D : Articles ‘TOP Control 1999 ViliFundamentals of process control Learning objectives : * define and understand the terms used in process control, understand the basic concepts of feedback and feedforward control choose the proper control strategy based on performance criteria, > understand how different processes behave identify and explain process gain, dead time and time constant, > understand the effects of each of the above, > define and understand the 3 control actions: P, I, D, * understand the differences between the 3 most common industrial controller structures, * convert units of P, I, D, tune a control loop by different methods, > qualify the design of control loop and understand the process limits, > properly choose filter time constants > leam methods for tuning interactive loops > leam loop troubleshooting techniques ©TOP Control 1999 ixFundamentals of process control 1. Introduction OTOP Control 1999POSS SSSCSCSUSTSCSCSSTSESSSESSESSESSESCSSVSESESESSEHCSCSCETCECEETECECEECEESESE Fundamentals of process control 1.1. What is a process? ‘A process is any operation or sequence of operations involving a change in a product or substance Examples: > a change of dimension as in ore grinding, > a change of composition by a chemical reaction or by mixing different materials, > a change of energy state, such as from cold to hot, or from water to steam. A process may be elaborate or complex, Usually, a complex process is a sum of simple processes. >| Paper mill Water treatment Ly Pastry Figure 1 Industrial processes ©TOP Control 1999 3Fundamentals of process control 1.1.1. Base Elements of a Process A complex process is the sum of many base processes. For example, the Water Treatment process (Figure 2) is broken down into: pumping, sedimentation (decanting) and filtration, decontamination (ozonation or chlorination), pumping and distribution Each of these processes can be categorized as one or more of the following Base Processes Level Pressure Turbidity Flow Speed Consistency Temperature pH Charge Etc. Flow control Level control ss eeq contrat Pressure control Flow control Pressure control Level control Flow control Turbidity control Speed contro! Speed control pH control Level control Pressure control Figure 2 Water treatment TOP Control 1999 4Fundamentals of process control 1.2. Components of a Control Loop ‘A contol loop consists ofthe following elements The base Process being controlled Examples Level of a tank Pressure of a vessel ‘Temperature of a liquid Flow from one system to another Etc ‘The Process Measuring Element Examples: Level Transmitter Pressure Transmitter ‘Temperature Transmitter Flow Transmitter pH Transmitter Etc, syes and ears” for the controller Controller Examples Switch PID Fuzzy Logic Human controt ‘The “brains” of the control loop Final Control Element Example: Heater Valve Pump Damper ‘The “musele” for the controller ©TOP Control 1999 3Fundamentals of process control 1.2.1. Control Objectives The goal of a control system is to maintain the process variable equal to the setpoint in the presence of load changes, 2s of setpoint changes, Tuning a controller means adjusting the parameters (P, I, D) to achieve "Good ‘Good Control” is performance criteria that depends on the application Is the trace from the flow loop in figure 3 an example of “Good Control”? The measurement is being maintained at setpoint but isthe PID controller properly tuned? 55. Pv w Pee ba 2 40 60 80 100 120 140 160 co [21 panerennpennyrsnerrnrnrenttn MRA NAAR AAA Mm AAyON. 35. 2 406080 100 120 140 160 Time (seconds) Figure 3 A flow loop in run, ‘The truth is we do not know. We must induce a change in the setpoint or the load to see the dynamic loop behaviour. loop that looks good in steady state may not behave as desired when changes occur. Figure 4 shows a different picture. ‘The response probably does not meet our objectives. 55, pv co) 40. 60. co 2 0 60 80 100 120 140 160 (3) 3. 2 40 60 80 100 120 140 160 Time (seconds) Figure 4 The same flow loop from figure 3, response to a setpoint change. Se To determine if a loop is properly tuned, one must induce a change. This is normally done by creating a setpoint change or a load change. “4 SP change is more drastic than a load change since with a SP change the error changes instantly. A load change must go through the process and is more gradual. ‘©TOP Control 1999 6Fundamentals of process control What is good control? Isit > Fast? » Without error? > Without overshoot? > Without oscillations? A ‘Maximum error Process value Setpoint ee detail i residual error + tronsientresponss = sa Permanent state > Figure 5 Setpoint response. ‘Tuning objectives will normally vary from loop to loop. Very fast control with slight continuous oscillation may be fine for one loop but dangerous for another loop. ©TOP Control 1999 7Fundamentals of process control For example, the controller of a robot arm moving a boile filled with nitroglycerin will be tuned without overshoot and without oscillation. The response will be slow and sluggish. 70. Py wa 50, 70. @ hie 2 4060 SCOSCOSCOSSA CT Time (seconds) Figure 6 Sluggish Response. Conversely, a flow controlter in a loop filling milk bottles will be tuned tightly and aggressively to ensure the maximum speed of the production line with minimum error, In such a system, the process tolerates oscillations but must minimize the error. 70. pv wf 50. 2 4060 80100120140 ‘160 70. co a ws 50. 240 «6S 18D Time (seconds) Figure 7 Aggressive Response, @ Tuning a controller is not only making the process operational, itis also making the process meet the performance criterion. ‘TOP Control 1999 8Fundamentals of process control 1.3. Controller Basics 4.3.1, Human controller Humans can make rational decisions to control a process. The control cycle goes as follows: > observe the process by way of the Measuring Element > based on knowledge of the Setpoint, make a decision repeat the cycle to eliminate residual errors v ‘manipulate the Final Control Element as required to impact the process > wait for the process to respond Please maintain this process 7.2, Process Measuring Final Element Control Element Figure 8 The Human Controller. ‘TOP Control 1999 9Fundamentals of process control 4.3.2. Automatic controller ‘When automating processes, the PID controller is the most common choice in industry. This controller makes simple calculations based on the current error. The result is sent as a Controller Output (CO) to manipulate the final control element, Final Control Element Disturbances Process | Disturbances Figure 9 Closed loop (control and instrumentation) Each balloon represents a function: > LT Level Transmitter > LC Level Controller > LY Level Valve > LY Level _relaY (converter, mathematical function, ...). Each type of line represents: > Piping, links > Electrical signal > Pneumatic signal 77 See Appendix A ~ “ISA Symbols” for more detailed descriptions of symbols found on P&ID"s. ‘TOP Control 1999 10ee eee ee ee eet ears > ) ) > ) > ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) , ) ) , Fundamentals of process control 1.3.3. Examples of simple processes |____ Figure 10 Level process, free outflow. | ol % |__| 6% Figure 12 Level process, constant outflow. Figure 13 Level process, pressurized tank, Figure 14 Pressure process. bb Figure 15 Flow process. TOP Control 1999 WWFundamentals of process control 1.3.4. Closed Loop - Feedback Control APID controller uses Closed Loop control. A Closed Loop system is any system in which a change in the measured ‘Variable will cause a change in the Controller Output (CO). For example, when the temperature ina room with an on-off temperature controller rises, the heater is tuned off. The term “Feedback” is also used for this arrangement as the current room temperature is fed back to the controller, Ina PID controller, the CO is computed as a function of the Error between Setpoint (SP) (desired value) and Process Variable (PV) (actual value). Since the CO depends on the Error, each time a disturbance occurs, an Error appears before the controller reacts. The controller uses this Error to eliminate the Error! Disturbances Figure 16 Feedback Control. > Cheap Feedback Control > Easy to understand > Low performance (an error is needed before controller acts) ‘TOP Control 1999 12Fundamentals of process control ‘The PID controller is the most popular controller in industry. This controller uses only the Error to compute the CO. Errors occurring in automatic mode are normal. These errors are used to correct the position of the control element (usually a valve). After a certain period of time, the Error will go to zero and the system stabilizes, Setpoint Process variable Figure 17 PID controller Hence, a transient response is always present when a disturbance occurs. Ifthe controller is a Pl or PID model, the PY will return to the SP after the transient response. When the transient response ends, the system is stable (equilibrium is reached). This state is called the permanent state, With PID control... ‘The error is used to eliminate the error. SS A straight line on the trend is out of question. ‘For the controller to move the valve, an error must occur. TOP Control 1999 1BFundamentals of process control Open Loop - Feedforward Control A controller is in Open Loop Mode if changes in the PV do not affect the CO. For example > When a loop is in manual mode, if disturbance occurs the controller will not adjust the CO. > Likewise a valve without a positioner can be said to be an open loop system. An air signal is sent to ‘the valve to open it. However, ifthe valve is damaged and does not respond there is no way of ‘knowing unless we physically watch the valve for movement, For applications that need a higher level of performance, open loop control using feedforward signals can be used, This type of control does not work off the error, but rather the Disturbances. Each of the major disturbances that affect the process is measured before they can affect the process. They are then sent fed forward) to the controler. The controller uses the disturbances that were “fed forward!” to perform calculations that determine the correct CO. ‘When properly configured, this type of control will maintain the process at Setpoint regardless of disturbances Disturbances Figure 18 Feedforward Control. © Expensive Feedforward Control > Complex equations > High performance (errors are eliminated before they can occur) {In most applications, pure Open loop control is not used as any error in setting up the controller calculations will cause the PV to eventually deviate from Setpoint. Without having the PV “‘ed back” to the controller, the controller can not correct for this, Hence most Feedforward systems also Feedback the PV to the controller. ©TOP Control 1999 14Fundamentals of process control 1.4. Tuning PID Controllers — A Compromise ‘When tuning a PID controller a proper filter is chosen and the proportional, integral and derivative gains are adjusted, The amount the gains can be adjusted varies from one make of controller to another. For example a stand- alone PID controller allows the user to enter a proportional gain, Kp, from 0.01 to_1000 Consider the fictitious case of Bob, the technician, installing this controller on a typical wastewater flow loop. He sets the integral and derivative gains to zero and adjusts only the proportional gain, He finds that the loop is unstable for proportional gains greater than 1.6. When he reduces the gain to 0,3 or less the controller no longer corrects large offsets. Adjusting the gain from 1.6 to 0.3 takes the controller response from instable (aggressive) to slow (sluggish or unresponsive). This relatively small adjustable range is typical for most loops. Only a small range of the potential values will work: and the difference from instability to unresponsive is normally around a factor of 4 ‘When tuning the integral and derivative gains in a controller, the same small acceptable range is also found. The tricky part is finding where the acceptable range is within the relatively large range of values the manufacturer allows. Fast, aggressive tuning is near the frontier of instability. Adjusting the controller gain a slight bit higher will cause the loop to oscillate continuously Slow, sluggish tuning allows large errors to persist for a long period of time, The loop seems unresponsive to deviations from setpoint. We say that a loop with this type of tuning is stable, as it does not have the potential to oscillate due to controller tuning. Ina loop with integral and derivative controller gains set to zero, if we decrease the controller gain, K,, enough we will eventually reach this stat, Range of acceptable values = k, minimum allowed K, for K, for sluggish tuning aggressive tuning Figure 19 Controller Gain Range. SS Normally, adjusting the controller gain, Ky, by a factor of about 4 will take the process disturbance response from slow to fast. ‘There are two prevalent tuning methods in practice today. They have different aims, One aims for quick disturbance response even if it causes cycling, This method is called Zeigler-Nichols Tuning (ZN method). The second common method is Lambda tuning, ‘The aim of Lambda tuning is to move slowly but gracefully ~ never cycling after a setpoint change or disturbance. These two methods are the opposite ends of the scale discussed above ~ aggressive tuning (Zeigler-Nichols) versus sluggish tuning (Lambda). ‘When we tune a loop we have to determine which style of tuning we want to use: aggressive, sluggish or somewhere in between, When doing this we need to keep in mind that with aggressive tuning, the loop is always on the verge of starting to oscillate because the controller is working so hard. GTOP Control 1999 15Fundamentals of process control ‘On the other end of the scale, we have to realize that with sluggish tuning, when a disturbance occurs, the controller ‘will allow the process variable to deviate far from setpoint and stay there for quite some time before correcting ‘Tuning a loop is a compromise between speed and stability Figure 20 Tuning - A Compromise Between Speed and Stability. 1.5. The need for process control quality > safety > efficiency > reliability labor cost v > operability 1.6. Definitions and Formulas Pv co SP JAE PID ISA Kp vvvyyvy 2 Process Variable Controller Output Setpoint minimum Integral of Absolute Error Proportional Integral Derivative Controller Instrument Society of America Proportional Gain ‘TOP Control 1999 16Fundamentals of process control 1.7. Exercises 1. Why is it normal to observe an error in a closed loop system? What are the adjustments in a PID controller? Give 3 criteria of | 4. What defines a loop as being Open or Closed? Name 4 base processes, ©TOP Control 1999 7Fundamentals of process control 1.8. Notes ©TOP Control 1999 18Fundamentals of process control 2. Process dynamics (open loop) GTOP Control 1999 19Fundamentals of process control TOP Control 1999 20Fundamentals of process control 2.1. Simple Process — Self Regulating Process ‘The figure below shows a typical level process. Figure 1 Level process In the aim of better understand the process dynamic, we can simplify the above graphic to what is shown below: & ou Figure 2 Simplified Level process ‘The behavior of the above process isto stabilize by itself. If the level increases, the pressure at the bottom of the tank will also increase, Since the outflow (Qou) is proportional to pressure, it will increase as well. For the level to be constant, the outflow (Qout) must equal the inflow (Qi). Such a process is called self-regulating since without a controller, the evel tends to stabilize by itself ‘The complex behavior of this process uses a simple model, the most common being Ziegler-Nichols. ‘©TOP Control 1999 2Fundamentals of process control Each piece of equipment in the system can be considered to have its own process model (transmitters, valves, L/P converters, tank, pipes, pumps, etc...). To simplify our analysis, we will begin by considering the valve and the tank together to be one process, Excitation Signal 3:15 PSI = 0-100% co tt tomo Fixed Opening Figure 3 Process model ‘When the valve is opened the level will rise and stabilize at a new level. The observed response is as displayed below. We will assign a mode! to this response that fits the actual response as closely as possible. Valve Opened by ACO Value Process Response Model ofResponse Sy Level Rises by APV% Time Figure 4 Process response ‘OTOP Control 1999 22Fundamentals of process control 2.1.1. First order systems ‘The model that we fit to the response of our process matches that of a First Order System. The system is characterized by three parameters: Process Gain Gp ‘Time Constant Dead Time fg AAs the goal of this section is to gain a practical understanding of process dynamics, detailed mathematical explanations of this process response will not be given. See Appendix B~ Mathematical Analysis of First Order Lag Systems for the math behind the model Process Gi Process Gain, also called "static process gain", refers to the sensitivity of the level to an output (valve signal) change. A high process gain means the level will rise greatly for a small increase in valve opening, ‘+ Ifthe level is constant, the outflow equals the inflow. + Ifthe inflow increases, the level increases. Therefore, pressure at the bottom of the tank increases causing the ‘outflow to increase, When the outflow equals the inflow, the level stabilizes. ALevel [%]___AProcess value [%] _ APV [% ‘The (sti Gp is the ratic G —_i 4. Press gain (Satie) Gpisthe rato: GP = 7 Fatve %] - ABignal final element |] ~ ACO [6] Figure 5 Tank model Valve Opened by ACO Value APV (%) ACO (%) ACO Process Response Model of Response a Level Rises by APV% Time Figure 6 Level after increasing the inflow For ease of control, the process gain should be: 0.5 < Gp <2. Ifthe process gain is under 1, the valve is too small. If the process gain is over 2 to 3, the valve is too big, ‘TOP Control 1999 2Fundamentals of process control ‘The magnitude of the process gain depends on the density of the liquid and the sizing of the valves. For example, if the valve size increases, the process gain becomes larger. If the process gain is larger, the control will be more difficult since a small change in valve opening can lead to large swings in the level. | Peace rece ee eee SECS Valve “10% Level - 15% Time Figure 7 Increasing the valve size increases the process gain, ‘TOP Control 1999 24Fundamentals of process control Time constant t ‘The behavior of level in the tank is rclated to the difference hetween inflow and outflow. Ifthe inflow increases, the level increases. Inversely, if the outflow increases, level decreases. The rate at which the level increases is proportional to the difference between the two flows nN Level 100 80 632 —e0 40 20 time Figure 8 Step response of a first-order lag system. A first-order lag system consists of a single capacitive (storage) element where the outflow of material is Proportional to the amount of material stored in the capacitive element. [Ifthe system were to continue to change at its maximum response rate (it occurs atthe beginning of a step change), it would reach its final value in one time constant (fime constant = tau =), ‘The rate of response is at maximum in the beginning of a step change and decreases continuously The time constant is the time required for the process value to reach 63.2 % of the total change. “Ff the system were to continue to change at its maximum response rate (it occurs at the beginning of a step change), it would reach its final value in one time constant. ©TOP Control 1999 25Fundamentals of process control A special mathematical relation ‘The response of first-order system to a step is of special interest: The instantaneous rate of change depends on the difference between the final value and the actual value. Therefore, what was true for the initial response is true for all amplitudes, Figure 9 Exponential curve. ‘F The response of a first-order system to a step is an exponential curve. The slope of this curve depends on the difference between the final value and the actual value. The time constant is the time it would take to reach the final value with the actual rate of change. OTOP Control 1999 26Fundamentals of process control Time constant value ‘The magnitude of the time constant depends on the diameter of the tank, the inlet pressure and the valves. For example, if the tank diameter is increased, the time constant increases. The process gain will not change as the pressure at the bottom of the tank is not affected by the diameter. If the ime constant is larger, the control will be easier since a large time constant smoothes out disturbances, Figure 10 Increasing the tank diameter increases tau. TOP Control 1999 27Fundamentals of process control First-order systems examples : time p time : Pp | time a i pea time Figure 11 First-order processes. 8Fundamentals of process control Dead time Systems with transportation time are frequent. In such systems, after a change in the signal to the final control clement, a real delay is present before the process reacts. A real time delay or dead time deteriorates the contro! since the controller is not aware of any change in the process during this dead time. hot water dead time ty aasssseeeeeesecy coldweter warm water igure 12 Mixing of hot and cold water. — deadtimety = — Figure 13 Conveyer. ‘S The dead time is the amount of time needed to detect a change in the measured value after a new signal to the final control element. The dead time is catastrophic since the controller cannot see the process response during this period of time. TOP Control 1999 29Fundamentals of process control N* order system If the process contains two first-order systems in series, the response of the second system to a change in valve opening is much slower as the level in the first tank needs to rise slightly before the second tank sees an increased flow. If we look at both tanks together as a process, the response of this process is delayed, a H2 Figure 14 Two tanks. GTOP Control 1999 30Fundamentals of process control If more first-order systems are in series, each subsequent response is delayed for a longer time period (see figure below). a Tt If you look carefully, you will observe very small 0 changes in the value. The derivative action ofthe | FID controler will use the slope ofthis change to anticipate the measurement, Lt ‘When we try to observe our process, we are looking through many first order systems in series. We have delays created by the transmitter, the process, the converter and the input of the controller. The end result is that we (or the controller) will only see a change occur in the measurement sometime after it has occurred. Likewise when we ask to move the valve we have to wait for the signal to transmit, andthe actuator and positioner to work. Figure 15 N tanks. ‘These delayed responses will make the control more difficult. The derivative action of the controller can sometimes ‘be used to detect the beginning of the change in the process value. ‘©TOP Control 1999 31Fundamentals of process control 2.2. Ziegler-Nichols model 2.2.1, What is a model? ‘A model is a simple representation (mathematical or graphic) ofthe behavior ofa real process. A mode! is not perfect, itis an approximation ‘The Ziegler-Nichols model (Laplat e equation) consists ofthe following equation: Adu _ eo ——" = Gpx Aln leo ‘This equation will not be analyzed, instead the graphical model (next page) will be studied. ‘A real industrial process is a series of first-order systems and time-delay systems. To determine the equivalent model, a step change is applied to the valve. The Ziegler-Nichols equivalent model corresponds to the curve that matches most closely the curve of the process. Control valve dgnal LP ep gers WP ita | I Valve 5,14 Process value ute" Reservoir T . — ‘Transmitter 5, ta Control valve signal process value fa Model % Figure 16 Model ‘When only at the control valve signal and the transmitter output are analyzed, we do not see each individual time constant. Since the transmitter signal includes all ofthe time constants itis a complex signal. Since the analysis of ‘a complex signal is complex, a best-fit model is matched to the real data GTOP Control 1999 32See eee Fundamentals of process control le fs SHG acer Roationer ey ea Valve ee Level Transmitter Output Model Figure 17 Components of transmitter signal An equivalent model is an approximation of the real process. Generally, the equivalent dead time corresponds to the sum of many small time constants (and dead times). It is essential to minimize the number of small time constants and the magnitude of each. 2.2.2. The equivalent model Many formulas and methods exist to determine the best process model. Some methods are sraphic, others are Miglemtical and some use elaborate frequency transformations, The purpose af these methods iso obaing model ‘where the response is almost identical to the real process. Process value meh Figure 18 Model and real process response. 33Fundamentals of process control 2.2.3, Natural frequency (closed loop) “The natural frequency is the maximum frequency reached by a process. “This frequency is obtained by gradually increasing the proportional gain. Ifthe proportional gain is high, the process «vill oscillate, The gain that permits the sustained oscillations is called the ultimate gain. If the proportional gain is maximum (on/off), the process will oscillate atthe same frequency, ‘This limit frequency is called the "natural frequency” of the process. The natural period t,(period=I/frequency) is of the order of ‘magnitude of the dead time: 2xta Response Valve Opened by ACO% Value Process Response ‘Model of Response SE <_< Integrating Time, Tye Figure 21 Integrating process response ‘The Dead Time, t, is as was defined in section 2.1 ‘The Integrating Time, Ti, is defined as the time it takes for the PV to increase by as much asthe CO increased (the PV and CO must be compared in %), TOP Control 1999 36Fundamentals of process control Another example ofan integrating process is tension contro loop. When the speed is adjusted in the loop below, the tension response will be an integrating response: Constant Adjustable Speed Speed Tension Process Feedback Response Drive speed up by ACO% Value Process Response Model of Response > Deadtime Integrating Time, Ti, Time Figure 22 Second Integrating process response TOP Control 1999 37Fundamentals of process control 2.4, Runaway process Some processes, after a new opening of the valve, do not stabilize, nor do they increase at a constant speed. Instead the measured value increases at an increasing speed (accelerating). A batch reactor where a chemical reaction takes place is a good example of such a process, Coolant Li iquid Figure 23 Runaway process. A coolant liquid flows through a jacket surrounding the reactor. The coolant liquid extracts the excess of heat generated by the chemical reaction. If an equilibrium is reached and the coolant flow is reduced: > > > the temperature in the reactor increases, the chemical reaction is stronger, more heat is generated, the temperature in the reactor increases, <— ‘These processes are difficult to operate and proper tuning is acquired through experience, sophisticated tools and strong knowledge of the process. The process model for this system consists of only the dead time. Process Model of R Deadtime | Response Time Figure 24 Runaway process response GTOP Control 1999 38Fundamentals of process control 2.5. Models An equivalent model represents the three different processes. Each of them has a dead time. The self-regulating ‘model has been studied; the other two are more complex and used less frequently integrating selfregulating time << & ee 7 —_ Figure 25 Processes models. TOP Control 1999 39Fundamentals of process control 2.5.1, Other characteristics Real processes differ from ideal models. The following points demonstrate the major differences between the two. > A dead time or a time constant which varies according to the process value (or to the output value). For example, in a cylindrical horizontal tank the time constant varies with level; > When measuring the temperature of a product flowing in a pipe, the dead time will increase as the flow rate decreases; > The process gain is not constant, usually due to a poor choice of valve characteristic; > The process does not respond equally to increasing and decreasing signals; this phenomenon is usually caused by a valve opening quickly but closing slowly (or the opposite); heat exchangers often present this problem since heat is more easily added than extracted. ‘The valve has hysteresis, dead band, backlash, stiction, friction, a > The process has limits; for example an overflow in a tank, Tip: To overcome these problems, obtain the process model for many situations and select the model that corresponds to the worst case when determining tuning parameters. > Large dead time, ta Small time constant, : > ‘Worst Cases!!! High process gain. G p v v ‘Tip: When the “worst case” process model has been found, the process model parameters can be used to determine controller parameters. Detailed formulas are given in chapter 4, but a good approximation for controller gain, K,, is as follows: _ Helpfal ?* Harmfull ~ 1,*G, GTOP Control 1999 40» > ) ) ) ) ) ) ) ) ) Fundamentals of process control 2.6. Definitions and Formulas Qn Flow into a system Qouw Flow out of a system Gp Process Gain. The sensitivity of the process, Measured as the change in measured variable (PV) divided by the change in controller output (CO). This must be calculated as %PV change divided by %CO change. + Time constant or tau, Is a measure of how fast the PV reaches its final value once the PV change has started ta dead time. The amount of time needed to detect a change in the PV after a change in CO. A ‘mathematical or graphical approximation of the behavior ofa real process. model A mathematical or graphical approximation of the behavior of a real process. Ziegler-Nichols model___A mathematical approximation of response of a first order self-regulating process te Natural Period. The natural period is the maximum frequency reached by a process, ultimate gain ‘The controller gain which permits sustained oscillations in the process self regulating process A process that will stabilize at a new PV after a change in CO. integrating process A process where the PV will increase or decrease at a constant rate after ‘a change in CO. runaway process A process where the PV will increase or decrease at an accelerating rate after a change in CO, ‘The process gain (static) Gp is the ratio: ALevel |% AValve[%] AProcessvalue |%| APV |% ASignal final element| 4] ACO lo 4] ‘The time constant isthe time from the start of the process response until the process reaches 63.2% ofits final value eietian te3.2% fa trv change observed The dead time is the time it takes to detect a change in PV after a change in CO: ta tey change observed _~ tco chan: ged ‘The natural period of a loop will be between 2 and 4 times the dead time: Xty digital, > electronic, > pneumatic. Itcan also be stand-alone, > part of a DCS (distributed control system), > programmed in a PLC program, > part of PLC system ‘Many models exist and each of them uses a particular structure > standard or ISA, > series, > parallel, ‘Also, many of units are used for different parameters > Proportional: * gain, © proportional band; > Integral: * seconds, * minutes, Derivative: > Filter: repeats per second (r.p.s.), repeats per minute (r.p.m.); seconds, minutes; seconds, minutes, He, rad/s. ‘Some manufacturers use their own structures and units! ©TOP Control 1999 51Fundamentals of process control 3.2. ON/OFF control ‘This type of control is old and simple. The domestic hot water heater and most home-heating systems have two-position control. This type of control will induce cycling at ,, the natural period in the loop as was explained in section 2.2.3 process variable | 100 time Figure 1 Two-position controller. Usually, this controller uses a simple relay. A relay normally has inherent hysterisis or a dead band, which decreases the cycling but increases the error of the process variable since the controller is less sensitive, process variable | | : vA setpoint T I |7 time ! | 100 % autput Figure 2 Two-position controller with dead zone. ‘TOP Control 1999 32Fundamentals of process control Another method used to reduce cycling is three-position control, This decreases cycling, but the error (@eviation from Setpoint) increases. process variable time 100 output time Figure 3 Three-position controller. ‘The two-position control (or more positions) is widely used but is limited to processes where the cycling that is produced is not detrimental, This is usually in processes that have a long cycling period (processes with large time constants) To decrease cycling, the number of steps can be increased which causes the error to increase. The process variable must deviate further from the setpoint to greatly change the output. F To decrease cycling, an “N" position controller can be used, however, as the number of steps are increased, the magnitude of the error increases. The process variable must deviate further from the setpoint to move the output. ©TOP Control 1999 33 setpointFundamentals of process control 3.3. Proportional control action 3.3.1. Introduction ‘The figure below represents proportional control as multi-position control where the number of positions is infinite. IE 7 L a 2 positions 4 positions n positions Proportional Figure 4 Proportional control. With proportional control, for each process value, there isa corresponding output value. The slope (sensitivity) is adjustable. ‘The proportional gain Kp is the ratio Kp = Doutput__ Aout _ output Aprocess. MPV Aerror ‘The more the output moves as a result of a change in process value, the higher the proportional gain, With proportional control, the response of the output is smooth, but the process value must move far enough away from the setpoint to move the output. The user must choose between cycling or deviation from setpoint. GTOP Control 1999 34Fundamentals of process control 3.3.2. The Watt regulator ‘The Watt regulator was the frst industrial proportional controller (the Romans used level controllers in their aqueduct). ‘This controller was designed by James Watt in 1788 to control the speed of a steam turbine, As the speed increased, the spheres moved upward by centrifugal force. This movement closed the valve and the steam flow was reduced. low speed high spe + 5 wars sor Opens valve | Closes valve | Figure 5 Watt’s regulator ) To open the valve, the speed must be under the setpoint. To maintain the valve open at this position, the speed must remain under the setpoint. Although the original error was used to reduce the error, a residual ‘error, or offset, must remain ‘The control is not perfect. If process load increases, the speed of the turbine decreases. Therefore, the ‘output increases but an error is needed to obtain a new output ed ©TOP Control 1999 35Fundamentals of process control 3.3.3. A simple level controller ‘The following figure presents a simple level controller. This controller is mechanical; the position of the float varies the inflow, Figure 6 Simple level controller. If the level is constant, the inflow must equal the outflow: If the level increases, the float moves upward and the valve plug moves downward causing the inflow to decrease until the level stabilizes (inflow = outflow). Conversely, Ifthe level decreases, the float moves downward and the valve plug moves upward, causing the inflow to increase until the level stabilizes, For example: ‘© initially the level is at Setpoint and inflow equals outflow ‘+ _ ifthe output restriction is enlarged, the outflow increases and the level decreases ‘© asthe level decreases, the float moves downward forcing the valve plug upward, thus increasing inflow when equilibrium is reached, inflow equals outflow. + At-stability inflow must equal outflow, however, we reached the new inflow rate by increasing the valve opening, For the valve to be at an increased opening, the float must be at a lower value... ‘A proportional controller will always have a residual offset error after eliminating a disturbance It is possible to reduce the residual offset by increasing the proportional gain; however, increasing the gain will generate more loop cycling. Ifthe fulcrum is moved to the right in the diagram above, the controller proportional gain becomes larger. ‘TOP Control 1999 36> » > > > > > > > > > , > > > > > > > > > > > > > > > > > > ? > > > » » ? > > > , > > ) > ) ) y > ) > ) Fundamentals of process control Disturbance Response with Controller in Manual SLAIN A a Ww e v w wr w ‘w w Se Proportional Control— Low Gain | Oaow increases = Setpoint — Pv} Tine (Gee Proportional Control — High Gain Setpoint — ott py! 7 Time (see)) Figure 7 Proportional Control - No Controller and High versus Low Controller Gain. ‘4 proportional controller uses the error to modify the output to the valve. A permanent error is present and this residual error changes the output. F skis possible to reduce the offset by increasing the proportional gain but increasing the gain will generate more loop cycling. ©TOP Control 1999 37Fundamentals of process control 3.3.4, Proportional band In the simple level controller, by moving the fulcrum point, you change the sensitivity ofthe controller. If the fulcrum is moved to the right, the valve opens and closes with a smaller change in level; hence the controller has a higher gain and the proportional band is narrower. ‘The proportional band is the magnitude (expressed in %) of the level needed to obtain a full travel of the valve, Using a large proportional band (small gain) decreases the cycling after a disturbance but increases the error needed to move the valve. Ifthe proportional band is zero, the controller is On/Off. ‘The maximum offSet corresponds to the proportional band. The smaller the proportional band, the smaller the offset. However, cycling will increase as the proportional band gets smaller. n /~ setpoint i Ft Offset % © eee * } : process variable valve signal h Figure 8 Proportional controller introduces an offset. ‘©TOP Control 1999 38> > » J ) ) ) ) ) ) ) ) ) ) Fundamentals of process control 3.3.5. Proportional band and gain The proportional band is the magnitude (expressed in %) ofthe level needed to obtain a full travel of the valve. ‘The proportional gain is AOutput(%) Kp = SQuiput(’6) P= "AError(%) %Error/ PB = A00% _ Abrror(%) 100% AOuiput(%) ————— [: raagaoms = © kp Referring to the figure below, Gain = Kp = 100% _ 100% 10cm 100% PB, =1OGM= 00M 999, _ 100% = 0% 094 - 0994 100% —0% 100% Valve opening 10 80 6 “ 20 o 2 4 6 2 w Leven) om @ 1m Leal 09 3 0 5 Enor(om) Figure 9 Proportional controller : output/input ee Proportional band is the inverse of gain. The proportional band ratio is expressed in percentage (%). ©TOP Control 1999 39)Fundamentals of process control ‘The proportional band is narrowed (and the gain increased) by moving the fulcrum to increase the ‘movernent on the plug valve opening (%) 25 change ee peer oor i say ope 1uj%becal ++ 1» Figure 10 Proportional controller, gain ‘The proportional band is widened (and the gain decreased) by moving the fulcrum to decrease the movement on the plug. valve opening (%) ‘uly (P| valve to ge fom fly open to filly colosed. ore | ‘eve woud ned to charge by 40% fo foes + oe 6 8 © devel (om) Figure 11 Proportional controller, gain = 0,25. GTOP Control 1999 60SEE EEE eee eee ee ee ee ee Fundamentals of process control 3.3.6. Controller action Direct action: the output increases if the PV increases. Gon = S0.8F = 20% valve opening (%) Gon = 20 aR = sox 100 Gon = 10 BP = 10 % 20 on = 05 BP = 200 % 60 Gain = 02 BP. = S00 x 40 0 0 2 4 6 © © level (cm) 0 0 ww oo Figure 12 Proportional controller, direct action, Inverse action: the output decreases if the PV increases, valve opening (%) Goin = 10 BP. = 100 2 = 20 BP soy 100 5 = 50 BP. = 20% eo = 02 BP. = 500 x 6 as BP, =e00 0 20 Die ettateteie thy level (cm) o 2 0 6 a 10 o> Figure 13 Proportional controller, inverse action. ©TOP Control 1999 61Fundamentals of process control 3.3.7. Block diagram ‘The block diagram of a proportional controller is: setpoict ouput Figure 14 Block diagram of a proportional controller. ‘The manual reset ensures an output when the error is zero. Usually the manufacturer fixes the manual reset at 30%, but on some controllers, the manual reset is adjustable. The user varies the manual reset to obtain zero error at the normal load, SF 16 the error is zero, the manual reset is the output value. ‘The simple level controller has a proportional band, setpoint and manual reset that can be adjusted. independently. The manual reset is adjusted as required to eliminate offsets at Adan SPA Figure 15 Level proportional controller. ‘OTOP Control 1999 62Fundamentals of process control 3.4, Integral control action (reset) The “advanced version” of the mechanical level controller has a mode that is very similar to integral action. It uses a slow moving DC motor to eliminate any offset from setpoint. Since this replaces the need to ‘manually reset the system, itis called an automatic eset. Anytime the float linkage is not level, the motor is tumed on (in the positive or negative direction). Thus the motor will keep moving at this slow speed until the offset is removed. ‘This method of eliminating offset is similar tothe integral control mode, If we wanted to implement integral control with a motor, the only change we would have to do is to make the motor turn faster for large offsets and slower for small offsets. YZ ZN ° e Figure 16 Simple level controller with automatic reset. ©TOP Control 1999 63Fundamentals of process control 3.4.1. Integral control Integral, or reset action, is the integration of the error. Reset is usually combined with the proportional Control action and is called PI control ‘To explain integral control action, it will be analyzed alone. ‘The output of the integral mode is always moving if an error is present. Ifthe error is zero, the output remains stationary, The rate of change of the output is proportional to the error, Automatic Reset Input ——+ Ki +> Output Figure 17 Block di yram of an integral controller. Integrator output speed = Kix Error ad(output,) at output, = Ki Error dt = Kix Error output, «© magnitude of error x time - If an error is present, the output of the integrator is changing (increasing or decreasing). OTOP Control 1999 64» > J ? > ) > > > » > > > ? > > > > , > > > > > > > > > > > > > > 4 > > > > > > > > > > > > > > > > > > > > Fundamentals of process control 3.4.2. Integral control response To slowly remove the offset... Ifan error is positive, the output increases at a rate proportional to the error. + = aece Ifan error is negative, the output decreases at a rate proportional to the error. 0 seca If.an error is zet0, the output remains stationary. I 1 I I I I I I i T 1 ina va iene | | I I I I 1 | ' ' \ ' ! ' output 1 ' 1 ! ! time Figure 18 Integral controller response. From 0 to A, the error is zero, the output is stationary; from A to B, the error is positive and constant, therefore the output increases at constant speed; from B to C, the error is greater, therefore the rate of change of the output is increased, from C to D, the error is negative, therefore the output decreases at constant speed; from D to E, the error is zero and the output remains stationary; from E to F, the error increases (negative), therefore the speed of the output increases. ©TOP Control 1999 65Fundamentals of process control 3.4.3. Pl control ‘When the integral control mode is combined to proportional control, itis impossible to have a permanent error since the output of the integral mode moves until the error reaches zero. The integral mode eliminates the offset but increases instability Exsor setae Output Pr Figure 19 PI controller. ‘The integral gain is adjusted to obtain a slow response to gradually eliminate an offset. Ifthe gain is too high, the controller overreacts and cycling occurs. Ifthe integral gain is too low, the offset is eliminated very slowly. & 4 Pleontoolleris like a P controller except that the reset is automatically adjusted until the error reaches zero. ‘TOP Control 1999 66Fundamentals of process control 3.4.4, Integral time, integral gain ‘The integral gain is the number of repeats of the input (the error or the proportional output) by unit of time. If the input is repeated 4 times per minute, then the integral gain is 4 p.m, ‘The integral time is the time required by the output to match the input A Input Input, output Figure 20 Integral time. If the integral gain is high, then the integral time is low and the output varies with a high rate. Ki= 7 (repeats per second or per minute) 7 fi ie Ti= +. (minutes or seconds) Ki & ‘The integral time is the time needed for the integral block output to repeat the input. The lower the integral time, the higher the integral gain. TOP Control 1999 67Fundamentals of process control 3.5. Derivative control action (rate, pre-act ) ‘The mechanical level controller never had a derivative mode, However, if it did it would have been a ‘mechanism that looked at how quickly the level was rising or falling and moved the valve accordingly. If the level started to rise quickly, it would close the valve by a value, say 5%, and then wait and see ifthe action helped. Ifthe level starts to drop slowly, the reset mechanism would open the valve 1% and then ‘wait to see if the movement helped. Derivative action does not measure how far the level is from setpoint, it only looks at just the rate of change of the level Automat Reset, SP Adjustment Derivative action — open valve if level is dropping, close valve if level is rising (irregardless of what the setpoint is) Figure 21 Simple level controller with automatic reset and derivative action. 3.5.1, Derivative control ‘The derivative control action is never used alone; itis always combined with P or PI control. ‘To explain rate control it will be analyzed alone. ‘This control action reacts to the rate of change of the error only. The output is independent of the magnitude or the polarity ofthe error. This control action generates an output that is proportional to the speed of the error to anticipate the behavior ofthe process. Anticipating that the present determines the future OTOP Control 1999 8Pe eee ee Fundamentals of process control Derivative Input —__} Kg Output Figure 22 Block diagram of the derivative controller, Output, = speed,,., x Kd Output, % rate of change of error - If the error is stationary, the output of the derivative block is zero. ©TOP Control 1999 9Fundamentals of process control 3.5.2. Derivative control response To anticipate errors... If the rate of change of the eror is positive, the output has a positive value 7 proportional to this rate of change. Sec goee Ifthe rate of change of the error is negative, the output has a negative value @ @ @ @ —| proportional to this rate of change S806 If the error is stationary, the output is zero, : error output time Figure 23 Derivative response. From 0 to A, the error is stationary and the output is zero. At A, the error increases abruptly, the speed is infinite and the output is maximum positive, From A to B, the error is stationary and the output is ero, ‘ACB, the error decreases abruptly, the speed is infinite and the output is maximum negative, From Co D, the rate of change of the error is constant, therefore the output is constant From D to E, the rate of change of the error is larger constant and the output is larger constant. fier E, the error is stationary and the output is zero. ‘TOP Control 1999 70> > > , ) ) ) ) Fundamentals of process control 3.5.3. Derivative time ‘The derivative gain is also the derivative time. The derivative time is the time when the output is before the ‘input, Within this period of ime, the ouput is higher than the input: the output leads the input Enor [_ Output D a Outpt D Poe eeeee a —5 Output D 0 oe Po time Ouaput n+p Figure 24 PD controller response, petit derivative time is the time that the derivative block output will be ahead of the blocks input (error or output of the proportional block). The longer the time, the stronger the derivative action, TOP Control 1999 71Fundamentals of process control 3.5.4. Effects of each control mode In proportional contro! mode, The controller reacts instantly to the error. In integral control mode, it reacts slowly to the permanent error. The derivative control mode reacts instantly to a change in the error (the speed of error) Ieee Kp x error Eros ct ci) 7 t yeu ue, Figure 25 Control modes response. ‘The first to react to a change is the P and D, followed by the I. ‘SA properly tuned controller is a controller where each control mode reacts in the proper amount. The main correction must be done by P. The I and D supports P by improving the correction. ‘The Tis essential to eliminate the offset. Since it moves slowly, the I tends to lag the response and introduces a "delay" which generates cycling and overshooting. ‘The D improves the speed of response and reduces overshooting and cycling. The D decreases cycling and improves stability. If excessive noise is present, the D will react violently and will introduce large fluctuations to the terminal element. In those cases, itis recommended to filter or avoid using derivative control action. OTOP Control 1999 2Fundamentals of process control 3.5.5. Comparison Magnitude and Duration of error Advantage Eliminate offset (Output proportional to Disadvantage JOvershoot, cycling To increase elements effect Tit, orKit Objective c Eliminate offset Phase Frequencies Low [Deals with Slow error Rate of change of error Speed of response Sensitive to noise Td, Kat Anticipate LEAD High Fast errors SP, Error eee co P PV D Figure 26 The PID Controller. TOP Control 1999 BFundamentals of process control 3.6. PID control 3.6.1. Introduction Industrial controllers are usually the PID type. ‘The proportional control mode is the most important mode. The integral and the derivative control modes are complementary control modes. AD cine sous temasasa Pp > Proportional: © to correct the process value by using the error to move the valve > Integral: * tocliminate the residual error, or the offset, by integrating the error until it disappears > Derivative © toimprove the stability of the loop. If properly tuned, it permits an increase to proportional gain and integral gain. 3.6.2. Controller structures (Ideal, Series, Parallel) The three control modes are combined to build a PID controller. The exact structure varies from one ‘manufacturer to another, ‘We will look at the three predominant structures: > Ide also know as standard, ISA, non-interacting; > Series: also know as interacting; > Parallel: also known as non-interacting, ‘©TOP Control 1999 4Fundamentals of process control Ideal Model (also known as ‘standard, ISA, non-interacting) Figure 27 PID controller, non-interacting. ‘The equation is Controller Output = Error x(P + PI + PD) Controller Output = Error x P x(1+1-+D) Controller Outpui («» xE + KpKix J Edt + KpKa x Comrolter Output = Kp x (e + 2 j where: E=PV —SP = Process Value -Setpoint 100 Kp =Gain= 7. 1 Ti= Integral Time = — fegral Time = ‘Td = Derivative Time = Kd ©TOP Control 1999 5Fundamentals of process control Series Model (also known as interacting) — Outpus P+PI4PD+PID [ ‘ PID Ouipusa a : a 2 Ol; + re ten Deane Figure 28 PID controller, interacting. Output = P+P1+ PD + PID. x(1+1)x(1+D) +PI+PD+PID (P + PID)+PI+PD =Pf(i+1D)+1+D] Output = The equation is eH Td\. 1p, dE Output = ta(v+ i ‘JE+ = J ‘Edt +r) + Series Format P[(1+ID)++D] dE ; Ideal F¢ PtH Onpu = Kpx( 144 Es rrayl Bay uur l uel 4a* Yi By writing the equation in the same format asthe ideal equation, we see that the proportional contribution to the output has been increased while the integral and derivative gains are slightly smaller. To obtain the same results as an ideal controller you must reduced P gain and increase the I and D gains: Td ) = equivalent ideal ein, Ti= Mi =T'i+T'd = equivalent ideal integral time, Td =~ = equivalent ideal derivative time. ‘OTOP Control 1999 76Fundamentals of process control Parallel Model (also known as non-interacting) Ermer Outputp - >, Output — f Output; Outputs Figure 29 Parallel controller. Output= —— P+T4D, ‘The equation is: 1 dE, Output = KpE + — | Edt + Td x= pu RPE, J a ‘The proportional gain no longer affects the I and D. The three gains are said to be independent The ‘equivalent 1 and D must be increased by comparison with the ideal structure. To obtain the same response as an ideal controller with a parallel algorithm, the P gain remains unchanged, but the I and D gains are increased: Kp" = kp = equivalent parallel gain, xi" = Kixkp = equivalent parallel integral gain, Te" = TaxKp = equivalent parallel derivative time. Kp = Kp = equivalent ideal gain, T= = Tix Kp" = equivalent ideal integral time, Ki Ta = it = equivalent ideal derivative time. kp Kp’, 1", Ki", Ta" values for parallel controller, Kp, Ti, Ki, Td values for ideal controller. ©TOP Control 1999 7Fundamentals of process control 3.6.3. Comparison Is possible to obtain equivalent parameters from one structure to another and obtain the same response from the controller. Each manufacturer has its own structure corresponding tothe three just discussed, The series and ideal ccontrolters are most common. The parallel controller is found only in some PLC and some DCS systems and controllers. The parallel controller is dificult to tune and itis not easy to interpret the results Ideal Series Parallel An example ‘The following values of Kp, Ki and Xd will give the same control (same equivalent P, I and D gains) Ideal Series Parallel ‘TOP Control 1999 8Fundamentals of process control 3.6.4. PI controller For a Pl controller, the ideal and series are the same, Oupet papi Ep > > > > > > > Exror| Outputr PO PHPL ee FD ie > PY : Ll p=0 ie Outpusi pulp Figure 30 PI controller, ideal and interacting, A Pl ideal controller is equivalent to a PI series controller. TOP Control 1999 79 ee ee eee ee ee eee ee eee eee et ee ee ee eet ee eee eee ee eeFundamentals of process control 3.6.5. Response of an ideal controller Enror Enor Exror Error ol LW] ee P P P P = LU] piel mmanudtrese L_ 1 | Fine fe lisa 7 oC D D D oe Figure 32 Filtering. , ) . Output Vr Ta Figure 33 Ideal PID with a filter. GTOP Control 1999 81Fundamentals of process control ‘The following figure illustrates the response of a closed loop with and without filter. 0 % 7 without filter 66 0 6 ‘0 00 S&F When a filter is added to a PID controller, the time constant must be small to ensure that the filter does not add a dead time to the process. The time constant must be large enough to reduce the amplitude of the noise. ‘OTOP Control 1999 82.Fundamentals of process control 3.8. Other features in a PID controller 3.8.1. Limitation of derivative gain In the presence of high frequency noise, using derivative action may be a bad idea, The amplitude of the derivative output will follow the noise and be excessive. To avoid ths, limit the high frequency gain with a {ter, A filter block i inserted before the derivative block to attenuate the signal, the objective being to limit the high frequency gain toa ratio of Ts Ta Maximum Derivative Gain =Gain,,.. where: ‘Ty is the derivative time, ‘ais the time constant of the supplementary filter. Figure 35 Ideal PID controller with limitation of D. Some controllers (usually in PLC's) do not have this supplementary filter and in these cass, the derivative action should not be used. F To reduce the derivative gain at high frequencies, a filter is added before the derivative block. The gain is limited to Ta/7j. In most Popular controllers, the gain is limited from 5 to 10 by default. ©TOP Control 1999 83Fundamentals of process control 3.8.2. Derivative action and setpoint changes If the operator changes the setpoint quickly, the derivative action will react violently. To avoid excessive derivative action, most manufacturers offer the user the choice of where the derivative action should take place : on the error or on the process variable ‘The following figure illustrates a PID controller where the derivative action senses the process variable only. To ensure the correct value, derivative time is multiplied by the proportional gain. = 2 1 Fitter Pv | Figure 36 Ideal PID controller, D on PV only. ‘TOP Control 1999 84Fundamentals of process control 3.8.3, Integrator wind-up (anti reset wind-up algorithm) ‘The following problems could occur in a closed loop’ > avalve is fully open but the PV cant reach the SP: > a pump stops temporarily; > during start-up, the conditions are abnormal Inall of the above situations, output for integral control action will continue to increase. If the controller ‘output is saturated and integral action continues to increase, integral output will increase. When conditions return to normal, integral output will return slowly to normal values. To overcome this wind-up, manufacturers of PID controllers add a special algorithm, This algorithm verifies if itis useful to continue to integrate. When the outpot of the controller saturates, the algorithm verifies if continued integral output will increase the saturation (same direction) or decrease the saturation. ‘This will depend on the direction of the error. If the integral action will only increase saturation, the integral output is clamped. ‘Some manufacturers add an option to adjust the values (minimum and maximum) where this algorithm, checks the usefulness of the integral action. %0 75 4 without wind-up algorithm xo | a PV) es | oo | © with wind-up elgorithan «J «4 36 ° 100 200 300 “0 ao time Figure 37 Closed loop with integral wind-up TOP Control 1999 85Fundamentals of process control 3.8.4. Manual/Automatic transfer ‘When the controller is in manual mode and the operator transfers the control to automatic mode, itis important fo ensure a bumpless transition. Usually, in manual mode, the SP is forced to follow the PV and the manual output value is put in the integral control mode. Itis possible to compute the value of the integral output that will ensure a smooth transfer from manual to automatic if the PV is different from the SP. ‘When the controller isin automatic mode, the manual output is forced to follow the automatic output 3.8.5. Full PID controller 7 SP Figure 38 Ideal controller. ‘This figure illustrates a complete ideal PID controller: filter, derivative gain limiter, wind-up algorithm, ‘manual/automatic transfer. TOP Control 1999 86Fundamentals of process control 3.9. Definitions and Formulas PB Proportional Band. Magnitude of the PV (expressed in %) needed to obtain fll travel of the CO. Kp Controller proportional gain. Manual reset Ina proportional controller, the CO selected when at setpoint with no error (similar to a bias). Ki Controller integral gain. Expressed in repeats per second (RSP) or repeats per minute (RPM) Tr Controller integral time. Expressed in seconds per repeat (secirep) or minutes er repeats (min/rep). Ka Controller derivative gain, Expressed in seconds or minutes Ta Controller derivative time. Expressed in seconds or minutes Ideal structure Recommended controller structure: CO = P(L-+1+D) (also called ISA, Standard or non-interacting structure) Series structure Controller structure with increased P contribution when derivative is added: CO =P{(LHD) +1+D} {also called interacting structure) Parallel structure — Controller structure that does not follow rules of thumb for tuning as gains are independent: CO =P +1+D (also called non-interacting structure) % Filer time constant. Normally expressed in seconds, minutes, Hz or rad/sec ta Derivative filter time constant, Normally expressed in seconds, minutes, Hz, rad/sec or as a fraction of derivative time. Anti-reset wind-up A function that limits the integral output once the controller output isin saturation Bumpless Transfer A function that forces the controller PID output to be equal to the last manual value when transferring to automatic Controller Proportional gain is the ratio of how much the controller output moves for a given eror. output _ Aout _ Aoutput_ Aprocess APV error Proportional band is the inverse of proportional gain expressed in percentage. 1 1 PB.=—x100% © —— x100%| * Kp=55% Integral gain is the inverse of integral time, Ki x (repeats per second or per minute) 7 ri=£ (minutes or seconds) Ki ©TOP Control 1999 87Fundamentals of process control Derivative gain equals derivative time. 4 =T,__ (minutes or seconds) The derivative filter limits maximum Derivative gain, |Maximum Derivative Gain = Gain. Ts Formulas for converting ideal parameters to series architecture: FE Kp = K'px| dea \aaite Ti= Vg,=T%+T'd = equivalent ideal integral time, Y= sinter ain Td “—- = equivalent ideal derivative time. Kip, Ti, T'd = values for interacting controller Kp, Ti, Td = values for ideal controller Formulas for converting ideal parameters to parallel architecture: [ker = Ep = equivalent parallel gain, Kit Kix Kp = equivalent parallel integral gain, Tae = TdxKp = equivalent parallel derivative time. Kp" Ti", Ta" values for parallel controller, | Kp. Ti, Td values for ideal controller. Formulas for converting parallel parameters to ideal architecture: Kp Ke ‘equivalent ideal gain, T= UKi = Ti"x Kp" = equivalent ideal integral time, Td = Ta'1Kp" = equivalent ideal derivative time. Kp", 1", Td" values for parallel controller, Kp. Ti. Td ‘values for ideal controller. ae OTOP Control 1999 887 » » . > > 4 > 2 ° ° ° > > > > > > > > > > J > » > > > ) ) ) ) ) > ) ) > ) ) ) ) ) ) ) ) Fundamentals of process control 3.10. Exercises 1-Proportional controllers control closed loops. Each line corresponds to a different controller. Complete the table below (iffa value is impossible, put an X) output Gain PB. Manval] action Pv SP reset | Controtter#1 | 30% T 100% 30% Direct 23% 5% [Controller #2) 50% 2 40% Direct 60% Controller #3 | 40% 25% Inverse 75% 80% Controller #4 | 65% 100% G% B% 30% Controller #5 150% 30% Tnverse 10% 00% ‘Controller #6 25 30% Direct 85% 3% 2-Explain why a proportional controller produces offset ©TOP Control 1999 89)Fundamentals of process control 3+ Find the equivalent units: Gain Proportional Band Proportional: 1 % 2 % 0.25 % 3.33 % ee 66% ae 500% Gain Time Integral: 6RPM. 10 see 0.167 min OIRPS see min 3RPM. atti pee min & RPM. 12sec min a e RPS. 0.5 sec min e e RPM caaiiiees 2 min e e e ¢€ @ e é e e e é ‘TOP Control 1999 90 é 64-Find equivalent tuning parameters Fundamentals of process control ryvyyyyyvyvyyy?yY ? 1 ©TOP Control 1999Fundamentals of process control 5-Sketch the responses for the following (assume parallel controller structure): > Kpis05, > This 60s, > Tdis 10s lp 1 division = 10 seconds Error ¥ 2 ep 4 med pate + * BRR ¥ A DEH + ¥ “ pee ee PID He PE ‘TOP Control 1999 92,HERE EEE EERE eee eee eee eee ee CeCe ene ee ee eee oe eee ee eee ier ee ase ae me aer-/et fame ne Sata Fundamentals of process control 3.11. Computer Exercises 1 - Observing offset error with a P controller @ Open up the simulated process “Simulation - DDE Server” — START PROGRAMS EXPERTUNE Simulation - DDE Server b. ‘Make sure that the flow process is selected ©. ‘Open up the data acquisition system (trending package) “DDE PID Tuner”— START PROGRAMS EXPERTUNE DDE PID Tuner 4 Go Online with the process — From the menu select “simulate.tun” and press the “FACEPLATE” button or ‘if you are inthe set-up screen, just press the “FACEPLATE” bution e ‘Wait until the flow is stable at setpoint and change the ‘tuning parameters to a “P” controller — Set I toa large time (ie. - 99999 seconds) Set D to 0 seconds £ oa setpoint change and observe the response. Is there an offset? g Add a small amount of integral (30 seconds), wait for the process to stabilize at SP then return the integral time to its previous value (99999 seconds). h ‘Change the P value (increase or decrease) and do a setpoint change, Is there an offset? Is it larger of smaller that in the last step? i Repeat steps g and h until you find a proportional gain that gives no offset on a SP change. What is the proportional gain found? Is the loop stable with this setting? TOP Control 1999. 93Fundamentals of process control 2-- Seeing the effects of filtering a. Open up the data acquisition system (trending package) “DDE PID Tuner” b. Select a datafile with a manual bump test in it (or archive a bump if there is not one in archives) At the bottom of the trending package window select “Tune from archived dat: ‘Select the proper file and open it using the “Tune...” button Note: Ignore any error messages associated with the tuning Observe the effects of filtering the Process Variable ~ From the window displaying the archived time data, select “Edit” Select “Filter” “Test” a filter with the following time constants: 0.5, 1, 2, 4, 8, and 16 seconds Does the filter add dead time to the process response as seen by the controller? d. Select and test a 0.4 second “First Order”, “Second Order”, “Second Order Butterworth” and “Moving Average” filer. Which does the best job? Which does the worst? Which adds the most dead time? 3.12. Notes TOP Control 1999 94Fundamentals of process control SECC ETT: 4. Tuning PID controllers ©TOP Control 1999 95Fundamentals of process control 4.1. Tuning a controller 4.1.1, Basic concepts Tuning a controller means determining the optimum value for each parameter of the PID controller : Xp, Ki, Kd and filter time constant t, As discussed in the first chapter, in a closed loop system, it is normal to observe errors, These errors are used to climinate the errors caused by disturbances or setpoint changes. 4.1.2. The closed loop A loop is closed when the process variable affects the manipulated variable. A simple way to automate a process is to react to the changes in the measured variable. ‘The controller observes the Measured Variable and adjusts the Final Control Element to maintain the Measured Variable near the setpoint "PID" controller ert patient, cautious, experienced and reliable, Setpoint 4 \ : Process Final Control Element Disturb ances Measured and controlled variable Figure 1 Closed loop. ‘This isa simple way to automate a process since 1, The designer does need to know in advance the disturbances involved 2. The controller needed is independent of the type of process 3. _A standard PID controller can be used 99% of the time. The output is computed using the error between the setpoint and the measured variable. ©TOP Control 1999 oTFundamentals of process control Final Control Element isturh ance Disturbances Process Measured and controlled variab ke /-—{T)- oe Disturbances PID cortotler Figure 2 Closed loop (control and instrumentation). OTOP Control 1999 98Fundamentals of process control 4.1.3. Eliminating the error using the error The automatic controller computes its output based on the error between setpoint (desired value) and process variable (actual value). Since the output depends on the error, each time a disturbance occurs, an error appears before the controller reacts, Setpoint Cortrolles Valve Process Process variable Error Ouput leasuremert Figure 3 Block diagram of a control loop. ‘The PID controller is the most popular controller in industry. The user must choose the amount of each control mode and the right direction for the correction, Setpoint Output Process variable Figure 4 PID controller. TOP Control 1999 99Fundamentals of process control 4.2. Good control 4.2.1. What is good control? > Fast? > Without error? > Without overshoot? > Without oscillations? A Meximum error Process value Setpoint j——_tensientresponse = ere Permanent state Figure 5 Setpoint response. Most of the time we desire perfect control: no residual error, no overshoot, no settling time and no oscillatic ‘ith a PID controller we have to expecta compromise. ‘The residual error is zero if the controller is a PI or PID type. Hence, the compromise is between settling ‘maximum error (and oscillations) The goal of a control system is usually to maintain the process variable equal to the setpoint in the prese: changes and disturbances. The system uses the error to correct the error. Tuning a controller corresponds to adjusting the parameters (P, I, D) to achieve a "good control". The pei» criteria depends on the application, For example, some processes can tolerate oscillations but need and have minimum error as their goal whi!> o*1-r processes cannot tolerate any oscillation at al. ©TOP Control 1999 100Fundamentals of process control 4.2.2. Performance criteria TT Critical damping : maximum speed. without overshoot a ae Process Variabe Figure 7 Overshoot less than.. T 7 Figure 6 Critical damping. 2 Tine (os) TTT ‘Procers Variable 20 Po oe ae 2 Tine (es) — Figure 8 Quarter amplitude decay. eee ee ee L ee a Figure 9 Integral of absolute error. — 380 “Tine (ee) TOP Control 1999 101Fundamentals of process control ‘The least aggressive tuning displayed on the previous page is called "critical damping’. It gives the quickest possible response without overshoot. If we try to go even a slight bit faster than critical damping we will have overshoot, for this reason processes that can tolerate no overshoot will normally use even less aggressive tuning just to be safe, There is a method that has been developed for this purpose. It is called "Lambda tuning" and is heavily used in the paper industry. ‘The second response has an overshoot H, It is normally expressed as a percentage of how big it is in reference to the steady state value (Percent Overshoot = H/C x 100%), Common tuning goals are 5 or 10% overshoot. This tuning sives a fast response but you should expect some oscillations. ‘The third and fourth responses are “Quarter Amplitude Decay” and “Minimum Integral of Absolute Error” or “LAE. These two have a very similar shape and give very quick response with evident oscillations. (Quarter amplitude decay specifies a dampened oscillation in which each successive positive peak is one-fourth of the preceding positive peaks magnitude. This criterion is popular because it is easy to understand and easy to apply in the field. The difference between it and IAF is that IAE will be slightly quicker and will have slightly less error (PV compared to SP). A faster response than TAE is not practical as it will entail a greater total error, more overshoot and will take longer to stabilize, Loops tuned to have quarter amplitude decay or LAE response should be re-tuned more frequently as any error in tuning or changes in the process will give a highly oscillatory response that could become unstable. A good compromise is tuning with 10% overshoot. This response is quick with very litte oscillations. This tuning also gives peace of mind, If the process dynamic changes slightly, the response may become more aggressive, but it will not become unstable, as may occur with IAE or Quarter Amplitude Decay. ‘There is a relationship between these responses that is very useful when tuning loops on the fly. Reducing the controller gain, K,, by a factor of two will take the response from aggressive to moderate. Cutting the gain in half again takes the response from moderate to slow. It is important to note that this is only the case with Ideal and Series type controllers. Ifa controller with a Parallel structure controls the loop, the integral and derivative gains must also be cut "Quarter amplitude decay" is easy to understand and easy to apply in the field; however, the tuning is very aggressive and itis best to reduce the proportional gain to avoid excessive oscillations. Se "Critical damping" is attractive because it has no overshoot and no oscillations butt is often not quick enough when disturbances occur. Increasing the proportional gain will add a slight overshoot but improve performance a lot. ‘©TOP Control 1999 102Fundamentals of process control ‘A closed loop is ess stable than an open loop but a closed loop reacts 4.3.1. Load change more quickly 4.3. Open loop and closed loop response py 7d (a) os | Figure 10 Closed and open loop after a disturbance. 4.3.2. Setpoint change PV es | (%) 60 OT Figure 11 Closed loop response after a setpoint change, 4.3.3. Observations > When a disturbance occurs, the derivative reacts instantly to the speed of the error, the proportional corrects the error and the integral gradually eliminates the error. PID is the fastest Plis the slowest v v PI produces the highest overshoot PID produces the smallest overshoot v A SP change generates more oscillations than a load change since the error is seen immediately by the controller TOP Control 1999 103Fundamentals of process control 4.4. Driving an automobile - an analogy. Anew driver ‘A teenager driving a car for the first time is like a proportional controller improperly tuned. The driver reacts vigorously to each disturbance, and the car follows a strange pattern on the road. After some hours of practice, the driver leas to react skilfully to the disturbances. The proportional gain is now properly adjusted. As the driver gains experience, he observes the road ahead and ensures thatthe car is properly aligned on the road by imagining a point in front of him (setpoint). He now gradually corrects his driving when the car tends to stay on a side of the lane, The integral part is now properly tuned. The driver will adapt his driving to his personality. This corresponds to choosing a performance criterion. An experienced driver An experienced driver will sense a disturbance (wind, hill, ice, .. and he or she will react before the disturbance affects his or her driving, The derivative action is now properly tuned. ‘The driver's skills have gradually improved. The proportional, the integral and finally the derivative actions have been used by the driver. When a controller is tuned, itis important to adjust each of the parameters in order. The proportional is always the first parameter to be adjusted. TOP Control 1999 104Fundamentals of process control 4.4.1. Dead time Dead time is the process characteristic that impacts control loop performance the most ‘As dead time is increased, control will deteriorate and the control loop will be less efficient. Example 1: ‘To understand the importance of the dead time, we will ask an experienced driver to drive a car on a racetrack (the driver will be alone and the objective is to drive ‘Properly without leaving the track). First, the driver drives normally to learn the trac, ae caver windshield with cardboard and installa wide-angle camera onthe hood. A special screen i installed Fre at the tack is displayed on the screen forthe driver. With the proper equipmen, the drives see he cn ‘mage of the track but in our camera we add a hal-second delay. We ask the driver to drive on the same tack What will the driver do? He is more careful, he drives slower, his movements must be slow and smooth. IPhe becomes more cautious, the car will stay on the road but the errors are greater... but he can drive! “ater, we add more delay to the camera, This time, the delay is 10 seconds. What will the driver do? He is more and more careful He drives vey slowly (i another car appears suddenly the driver is on his way toa Spee aula crash), his movements are calculated, slow and smooth. The errors are large the ear goes fon one ‘ide to the other but stays on the road, The driving is not perfect. but he can drive! Conclusions: > ina loop, if the dead time is increased, > the period of the oscillations increases; © the proportional action must be decreased :Kp is decreased, P-B. is increased; ® the integral action must be decreased :Ki is decreased, Tis increased: ® the derivative action must be increased : Kd(T) is increased: > the integral of the absolute error is increased; > the overshoot is increased; > globally, the performances are decreased ee Ina control loop, the overshoot, the integral of the absolute error and the Period are proportional to the dead time. The P and I are reduced if the dead time { is increased, but the D is increased. TOP Control 1999 105Fundamentals of process control Example 2: Here is an experiment that you can try at home to see the effects of dead time. Day 1: Experiment: Day 1: Time how long it takes you to get to the perfect water temperature in your morning shower. Day 2: Add a 50-foot garden hose attachment before your standard showerhead Time how long it takes you to get to the perfect water temperature in your moming shower. ‘What will you notice? * Ata flow rate of 10 feet/second, the hose attachment has added approximately 5 seconds of dead time to your “temperature” loop. Getting the temperature in the ball-park - “Proportional action” + You make a guess for the original valve settings, ‘+ Once you realize the temperature is too cold, you must suffer through cold water temperature for 5 seconds after making a big adjustment. ‘+ Ifyou make too much of a correction and the water temperature becomes too hot, you must suffer through hot water temperature for 5 seconds after making a second adjustment. Getting the temperature perfect ~ “Integral action” ‘+ You will make a series of small corrections to get to the final water temperature, waiting at least 5 seconds between corrections. 5 seconds waiting for the dead time, another second or two to feel the effect of the water on your skin (measurement delay), Anticipating temperature drops - “Derivative action” + Ifsomeone flushes a toilet in another part of the house, the water temperature in the shower stars to rise, In the first few seconds, the temperature may still be acceptable, but you anticipate that the temperature will continue to rise and you make a correction that will counter the “predicted” total rise in temperature Overall Performance: ‘+ Ittakes much longer to regulate the water temperature with the hose attachment, ‘+ Atincreased dead time, a larger temperature deviation occurs and it takes longer to get back to setpoint when a disturbance arises. ‘TOP Control 1999 106SE eee ee eee ee ee ee ee ee ee ee ee, oe ee ee ee a | ae ee Fundamentals of process control 4.5. Tuning: a compromise ‘Tuning a controller consists of making compromises between speed and stability. The more the loop is stable, the more it reacts slowly, and the larger are the errors following a disturbance. In industry, control loops are non-linear (the process mode! is not constant) and itis essential to tune the loop for the worst case, ‘The more the loop is tightly and aggressively tuned, the more it will react quickly but the pice to pay is instability cycling and overshooting, Before tuning a loop, you must ask > ‘is the process slow or fast? is the process self-regulating? is the process symmetrical? is the process linear, is the right valve characteristic used? is the process gain near one, is the valve properly sized? is the transmitter properly installed and is the transmitter working properly? does the valve have hysterisis? is dead time dominant? is noise excessive? vVvYVVYVvVVV Vv will the tests I do represent the worst cases? Afier tuning, the tuning parameters will be valid until a change is made to the process or to a component in the control loop. Stabilit) ~ Fast Response Time ~ Slower Response Time ~ Oscillations present ~ No Oscillations ~ Less Enror ~ More Error Figure 12 Tuning ~ A compromise, Speed or Stability ©TOP Control 1999 107Fundamentals of process control 4.6. Tuning methods 4.6.1. Tuning method When tuning a controller you should have clearly defined performance goals for the loop, ‘The process is tested and the results will be used to select tuning parameters that will give the desired response when the controller i placed in automatic. The testing itself can be done many ways: > Is the test done in manual (open loop) or automatic (closed loop) test? 7 Is the method based on mathematical formulas or by trial and error? v Will the tuning be done with software or by hand? Will we model the process to determine parameters? Will it be a time domain or frequency domain model? Pretune? vvvy Autotune? Selftune...? v 4.6.2. Ultimate cycling method (automatic mode) The ultimate cycling method (also called the closed loop method) is used for fast processes since oscillations are used to determine a model of the process, leading to parameters for the controller. To initiate cycling, a load change or a setpoint change is made. Since the controller is part of the tet, its behavior is included in the model. Some controllers react only with the integral action on a setpoint change (the P and D react only to PV change); this makes this method impossible with such algorithms. 4.6.3. Trial and error method (automatic mode) ‘The most common tuning method is trial and error. This is another closed loop method (done in automatic). With this ‘method, it is not essential to find the process model. By gradually adjusting the parameters the acceptable performance is reached. This method is time consuming, creates a lot of disturbances and rarely permits optimum performances. 4.6.4. Open loop method (manual mode) ‘The open loop tuning method is generally useful for stow processes. The controller output is changed in manual (process is bumped) and the process model is computed. Formulas based on the performance criterion selected are used to determine the controller parameters ‘The step change must be representative of the usual disturbances. If you know the process model, many formulas are available to obtain the parameters corresponding to a performance criterion. Usually these formulas are used for the ISA (ideal, non-interacting) controller. To use them with a series or a parallel controller, it is necessary to modify the formulas. With an integrating or a runaway process, the formulas are Complex and elaborate ‘TOP Control 1999 108Fundamentals of process control 4.6.5. Using a software package Many software packages are available to tune orto assist in the tuning of industrial controllers ‘The software should as a minimum: > nun under Windows, % accept ASCII data and/or provide a data acquisition system and/or permit direct connection to the DCS or PLC through DDE v compute the tuning parameters from data obtained in manual or automatic mode, v compute the tuning parameters based on user selectable performance criteria, v analyse the data to determine © process model, © hysteresis, * noise, © power spectrum, © Bode plot, * robustness plot, quality of the tests, © quality of data, > know the structure and characteristics of the industrial controller used (the user specifies the model and manufacturer only), > inform the user of any disadvantages of the specific controller, > have an excellent support and help file. ‘Many software packages on the market do not mect all of these criteria. Some suggest strange values and use non- coherent methods. ‘Most software packages use the frequency domain to analyze and find the tuning parameters, ©TOP Control 1999 109Fundamentals of process control 4.7. Noise, disturbances and filtering 4.7.1. Noise and disturbance Noise is high frequency “garbage” induced in the measurement of a signal. A variety of reasons could ‘cause the noise: the measuring device itself, improper routing of cables, improper grounding, ete Before tuning a controller, itis important to properly adjust the filter to eliminate the high frequency noise ‘on the PV signal. The basic idea is to choose the time constant so it will not increase the dead time but it © Figure 13 Noise A disturbance is a process upset that causes real variation in the measured signal. For example, a valve that strokes from open to closed every 2 minutes will cause a disturbance in the line pressure that the pressure controller will try and eliminate, If the period of a disturbance is too fast, the controller will not be able to return the process to SP between disturbances. When fast disturbances occur, we have to eliminate them through mechanical fixes or tune the controller sluggishly so that it doesn’t fight the disturbance. + The controller will never be able to eliminate disturbances if the period of the disturbance is less than the natural period of the loop, The disturbances appear to be “noise” to the controler. © I the controtter is tuned aggressively it will be able to eliminate disturbances that occur with a period > ty, «Ifthe controller is tuned sluggishly, it will take a long time to correct disturbances. It will only be able to eliminate disturbances that occur with a period >>> to ( E10 percent” igure 14 Disturbances ‘OTOP Control 1999 110Fundamentals of process control Sometimes process upsets occur very quickly. For example, an imbalanced pump will induce cycling in the line pressure with a moderately high frequency. ‘The cycling is real, but it happens too quickly to be ¢liminated by a pressure control valve further down the line. Because the disturbance is too fast for the controller, it can be considered as “noise” ‘fa disturbance occurs that is too fast for the controller to remove, its considered noise. ‘F Before adding a filter, its time constant should be much less than the dead time. 4.7.2, Choosing the proper filter Choosing a filter by trial and error method ‘To adjust the filter time constant, the controller is put in manual mode and the filter time constant is increased until the noise on the PV has been decreased to an acceptable level ©TOP Control 1999 ulFundamentals of process control Computing the proper filter By evaluating the apparent period of the noise, it is easy to determine an order of magnitude for the time constant. To evaluate the period, you must determine the number of times the signal crosses the average PV; the noise period isthe average time needed to obtain two crossings (2 crossings for each cycle), “NAY + @s —— Ras PORE PArrr| 40s 16 crosses “2 Figure 16 Evaluating the apparent period of noise. noise period 'S Thus choose a filter time constant of approximately 5 seconds. Using software to determine the proper filter ‘When using software, itis possible to visualize the effects of adding a filter to the actual data. The time constant is increased until the PV signal is acceptable and the behavior of the process is not modified too much, Also, if the Bode plot is available, the time constant can be chosen ten times less than the natural period expressed in rad/s, T jitter © Noise period T ater << ta t T per © 5 BERR TOP Control 1999 112 REBEL AR.Fundamentals of process control 4.8. Ultimate cycling method In this method we determine what Proportional Gain will induce sustained oscillations in the process. We measure the period of the oscillations and apply formulas based on the period found and the proportional gain that caused the cycling ‘The formulas given are for 4:1 amplitude decay tuning During the tests, the process is always under control, in automatic mode, It is possible to make small changes, so the amplitude of the oscillations will be small 4.8.1. Time needed If the steps are executed in a logical sequence, the time needed to tune the loop willbe from 10 to 30 periods of oscillations which corresponds to approximately 30 to 100 dead times. ‘The time to tune a loop is approximately * flow loops, pressure loops, 1 to 3 minutes, + level loops, 3 to 20 minutes ‘© analysis loops, minutes to hours © temperature loops, hours 4.8.2. Method Note: Each time a parameter is modified, itis essential to follow a geometrical progression : multiply or divide by two until the parameter is too far from ideal; then use the halfway value 1, The Land D actions are completely eliminated (or their gains are reduced to minimum). K;= 0, T.=: very large, Ty =Ky=0. 2, ‘The proportional gain is increased until cycling is sustained. If the gain is properly chosen, the oscillation will not increase nor decrease in amplitude. The period of the oscillations is the natural period, ty, The proportional gain that caused sustained cycling is called the ultimate gain, Kpo, Ifworking with Proportional Band instead of Proportional Gain, it must be decreased to induce cycling, ‘The setpoint changes done to find ultimate cycling should be in the order of magnitude that the process normally sees, Insert the values found for natural period, t., and ultimate gain, Kpg, into the formulas. The formulas will give 4:1 amplitude decay response. For less aggressive tuning, divide the final Proportional gain found by 2 or 4 ©TOP Control 1999 113Fundamentals of process control Formulas: Controller Formulas only TK, x 100% wT; | PI ve Tip: For peace of a mind, decrease Kp by a factor of 2 ‘©TOP Control 1999 4Fundamentals of process control 4.8.3. Example > Tuning a flow loop, controller in automatic, > Integral and derivative gains are set to 0, > Set point changes are made, increasing the Proportional gain until sustained oscillations occur, > Parameters are calculated for a PI controller and SP changes are made to test response. ‘A geometrical progression is used to determine the gain, for example: 1, 2,4 .. until the gain is too big. Reduce the gain by using the halfway value between the actual gain and the preceding. Repeat until the ultimate gain is reached 54 52 50 48 46 44 | 42 40 — . LHPD LPH LSI SSH HS POCO PLL PPP MPP Time (See) ws) Percentage Figure 17 Ultimate eycling. Time (s) 0 4 oo | 117 | 150 | 200 | 216 SP (%) 44 48 50 51 51 54 46 Kp 1 1 2 4 3. | wa | 136 Ti (sec) @ © © © wo | NA | 125 Ta (sec) 0 0 0 0 o | wa | o If 4:1 amplitude decay tuning is too aggressive, the proportional gain is reduced by a factor of 2, 3 or 4, The elapsed time is less than 5 minutes. ©TOP Control 1999 15Fundamentals of process control 4.9. Trial and error method ‘This method also called "tuning without mathematics" will give good performance if properly used. ‘This method is performed in automatic and is similar to the ultimate cycling method in that it increases the proportional gain until cycling is induced, However, instead of looking for ultimate cycling, 4:1 amplitude decay is sought out. Since this response is harder to find, the parameters are “fine-tuned” instead of just plugging into formulas. 4.9.1. Time needed If the steps are executed in a logical sequence, the time needed to tune the loop will be from 30 to 100 times the period of Joop oscillation, which corresponds to approximately 100 to 400 times the loop dead time. ‘The time to tune a loop is approximately: * flow loops, pressure loops, 2 to 5 minutes, * level loops, 10 minutes to hours * analysis loops, minutes to hours ‘© temperature loops, hours to days. 4.9.2. Method Note: Each time a parameter is modified, it is essential to follow a geometrical progression : multiply or divide by two until the parameter is too far from ideal; then use the halfway value. ‘The I and D actions are completely eliminated (or their gains are reduced to minimum). K;= 0, T; large, T:= Ky = 0. sor very ‘The proportional gain is increased until a 4:1 amplitude decay response on a setpoint change is observed, The period of the quarter amplitude cycling is the damped period, ty. Record the period and the gain that caused this response. See Figure 8 “4:1 Amplitude Decay” for a picture of this response. If working with Proportional Band instead of Proportional Gain, it must be decreased to induce cycling. ‘The setpoint changes should be in the order of magnitude that the process normally sees. Insert the values found for damped period, t,., and proportional gain found into the formulas given below and then adjust Integral and Derivative times to find best response. The formulas will give 4:1 amplitude decay response. For less aggressive tuning, divide the final Proportional gain found by 2 or 4. GTOP Control 1999 116Fundamentals of process control Formulas: Controller Formulas Poa Wp Gan Fmd Gea [Remembers PB= TK, x 100% oe Kj= 7, Pr Kp=(GatFaumd)x09 (gan Ti=tye Gecondt) Start with these values and then increase I (by decreasing Ti or increasing Ki) by using a ‘geometrical progression until itis impossible to increase it without losing the 4:1 amplitude response, Kp=(Gain Found)x13 (gain) PID sin ifnd for PI MPP eccontt) Tae! — Gecont) 0 ‘Add derivative, start with one tenth ofthe damped period, Do a setpoint change or watch the disturbance response and sce if the performance has improved over PI control. Also watch the CO, is the output moving a lot more than it was previously? If you see increased overshoot or a lot more CO variation, do not use derivative action. But if the response is better then increase P and I gains by 30% (increase K, and K;, decrease PB and T), Increase D by using a geometrical _Progression until itis impossible to increase it without losing the criterion chosen. ©TOP Control 1999 117Fundamentals of process control 4.9.3. Example Set Integral and Derivative Gains to zero, Determine P In auto, the proportional gain is increased until the performance criterion is reached; usually the person tries to obtain a 4 to 1 decay ratio. If the proportional gain is actually unknown, begin with 1. Make a setpoint or load change and watch the response, Use a geometrical progression. For example, 1, 2, 4,8, . until the gain is too big. Reduce the gain by using the halfway value between the actual gain andthe preceding. Repeat until you have found approximately 4:1 decay Note the period of the cycling that is produced at the gain (natural period ty). Example: P value 1, setpoint change — response is too small, 2 setpoint change — response is too small, 4, setpoint change — response is too big, 3, setpoint change ~ response is too small, 38 setpoint change ~response is too small, 3.75, setpoint change ~ response is too big, 3.67, setpoint change — response is OK. Add Start with the integral time (Zi = 1/Ki) equal to the damped period, ty. Increase I (by decreasing Ti or increasing Xi) by using a geometrical progression until itis impossible to increase it without losing the criterion chosen. Add D 1 Few persons use derivative action. This is historical since with old controllers (pneumatic and electronic) using D action was catastrophic when noise was present. With modem controllers, adding D action will increase performance from 20 to 100% if apparent dead time is present. Add derivative, stat with one tenth of the damped period ty/10. Do a setpoint change or watch the disturbance response and see if the performance has improved over PI control. Also watch the CO, is the output moving 2 Jot more than it was previously? If you see increased overshoot or a lot more CO variation, do not use deriv. action. But if the response is better, increase P and I gains by 30% (increase K, and K,, decrease PB and T)) Increase D by using a geometrical progression until itis impossible to increase it without losing the criterion chosen. Increase P until the criterion chosen is reached. When tuning this loop, SP changes are made. £ ‘end, a change is made in each direction, ‘OTOP Control 1999 118: : x 7° = > 2) = 0 o D > ay ° > 2 2 ad o FSOSSSSSSSTTSSSOOSSSETS Fundamentals of process control Finding ideal P 58 56 34 52 ¥ 50 Bae 5 as | 44 | 42 | 40 38 eee 0 20 40 60 60 100 120 140 160 180 200 220 240 260 Time (See) =a Figure 18 Finding P to obtain 4:1 decay. Time | 0 7 75 [14s [210 | The lst SP change was found be the closest to 4-1 SP Lae se ss as Tay Tt alot 01 decay Ki 1 1 2 [1s [175] x, found = 1.75 Ti(see) | © a wo | «© | w | Tha found=17 sec Ta(secy | 0 0 0 0 0 Adding I E 0 20 40 60 80 100 120 140 160 180 200 220 Time (See) Figure 19 Adding I to obtain 4:1 decay Tee 3 7] 120] 180] The sooond SP change was found tobe the best. SP (%) 40. 50. 40 | 50 Mote eee kp 1375 | 1575 | 17s | 1575 | 1575 315 (gain) 1=85 (0) Ti (see) wv | 7 | ss | 425 | 6375 Ta (see) 0 outta 0 0 ‘GTOP Control 1999 119Fundamentals of process control Adding D 72 88 eae i at with derivative = 80 before derivative 8 56 5 52 é8 | a4 40 36 0 2 «640 «6608100120 140160 180 Time (Sec) rv sp ——c0 Figure 20 Adding D to obtain 4:1 decay. Time) [0 5] 110] 135] When derivative was tested twas found to have a negative impact on the loop and sp@%)_| 40 | 50 | 40 | 50 _|tnuswas not used. Note the increased | 5 ‘movement on the CO when derivative is Ky 1.575 | 1.575 | 1.575 | 1.575 _|mavementon the CO Tiec) | 85 | 85 | 85 | 85 Ta(sec)_| 0 o [a7 [iz If the sequence is followed the tunings will give the criterion chosen: ‘The total time needed is about 15 minutes for this loop, ©TOP Control 1999 120 @ € € € ¢ e oe @ e ° « e oe @ e @ e e e @ e e e e e e e e e e e e e @ @ @ @ @ e @ a ¢ a a @ @ @ « ‘ aFundamentals of process control 4.10. Open loop method ‘This method is also called the bump test method, This method is different since the tests are done in manual mode. The main advantage ofthis method is the possibility o obtaining parameter values for any performance criterion, 4.10.1. Time needed ‘The time needed to tune a loop is approximately the sum of the dead time plus 6 times the time constant. * flow loops, pressure loops, seconds, (difficult to accomplish without a fast recorder or data acquisition system) * level loops, minutes to hours * analysis loops, minutes to hours + temperature loops, minutes to hours 4.10.2, Method ‘The method consists of producing a step on the manual output of the controller and the determining the process model (Gp, td, 2) by the response. 1 -Stabilize the process in manual mode, 2-Make a step change to the output (same order of magnitude the valve normally does), 3-Determine the process model parameters from the response, 4-Using formulas, calculate the controller parameters 4.10.3. Formulas Ziegler-Nichols Many formulas are available for many performance criteria. For example, the Ziegler-Nichols formulas for 4:1 decay: Controller Formulas eee eee eee we Note: For P Kp=—*— (gain) integrating processes, 1,xG. ae. ay replace 7 by Tin in r = a Pal bs PT cca these formulas Ti=3.3xt, (sec) P+ED Kp=1.2x—*— (gain) v= Tip: For peace 14xG, of mind, decrease Kp Ti=2xt, (sec) by a factor of 2. | Td =05xt, (sec) TOP Control 1999 121Fundamentals of process control Ziegler-Nichols Tuning Example ‘The test procedures are as explained in section 4.10.2 Result (After tuning) Figure 21 Open loop - Ziegler-Nichols Tuning. ‘The process model found is: G, 1, ta 10 seconds, 1 42 seconds, Parameter values: 42 seconds Oseconds x 1.1 Ti = 2x10 seconds = 20 seconds = 0.3 min. ‘Td =0.5x10 seconds = $ seconds = 0.083 min, Kp= x1.2=4.6 If the tuning is too aggressive, the proportional gain can be cut by a factor of 2 as shown on the right. If this is still too aggressive, the gain can be cut further, ‘The elapsed time is less than 10 minutes. s Tel) Figure 22 Result after tuning - gain cut in hall. ©TOP Control 1999 122Fundamentals of process control Lambda Tuning Another popular open loop method applies lambda based formulas. Using these formulas, the response will never have ‘an overshoot on a setpoint change. As a result, the performance when a disturbance occurs is quite poor. The PV will deviate further from setpoint and will take longer to return to setpoint when a disturbance occurs. This method yields similar results to setpoint tuning and the two are sometimes used synonymous. Relaxed ( Tex=2 Tox) z G,Qr+t,) pee Very Relaxed ( Tax=3 Tox) G,Gr+t,) + To." open loop time consant, tx closed Toop time constant Lambda Tuning Example ‘The test procedures are as explained in section 4.10.2. Bump Test Result (After tuning) reg py - eae tee Figure 23 Open loop - Lambda Tuning. The process model found is G, 11, Pa 10 seconds, + 42 seconds, Parameter values: 42 seconds Kp = Sons _____ 41 P11@x42seconds +10seconds) Ti = 42 seconds = 0.7 min. ©TOP Control 1999 123,Fundamentals of process control ‘The previous examples used formulas based on Ziegler-Nichols and lambda formulas, Other popular formulas would give the following results for the same process model: Process model td = 10,00 5; % = 42,005; gain= 1,10 [Ziesler-Hichols | ohen-Coon Reliable values | JE* (Load) pe | a2 4.2 16 78 PAL | 3144 0,56 Hal 0,37 Bal 0,78 70 0,49 P+L4D] 9,98 0,33 0,08} 3,40 0,38_ 0,05) 2,17 0,86 0,07) 5,27 0,21 0,03) JIE (¢ sp} LIEl (é Load) | FUEl#t (6 SP YT) Eat Gabel J 2,28 s 1,29 24d Pe | 2.37 0,43 3,58 0,42 1,58 0,54 3,1? 0,39 Pel4D] 3,44 0,78 0,07] 4:89 027 0,02) 299 0.71 0,05] 4/80 0128 0,05) Maxim um speed with an overshoot of : 0% G SPY Ov bead) | 20x sp] a 3] %A0 Lead) 2 t EB o.oy LE 13 ba Bho ib 2,67 2, 0,82 4.29 0, $29 0,71 i 0,70 0,08} 3,63 ui 0,02] 4,63 ae 0,08) 4 Figure 24 Different tunings. TOP Control 1999 124Fundamentals of process control 4.11. Autotune, pretune, sefftune, ‘Manufacturers of industrial controllers (stand-alone, D.C.S. ) claim that their controllers are intelligent, are self tuning and are able to adapt themselves to the process. Reality is often quite different lot of nice names ae used: autotune, slftune,optune, aecutune, ‘There exists two different approaches to controller automatic tuning > pretuning, > adaptive tuning Pretune functions: A program makes a test on the process and computes parameters tuning by analyzing the process response. The most Gomtmion tests that the programs do are ether a bump test in open loop orulimate cycling test in closed loop, A technician can do the same tests and analysis in less time and the results will be better. Frequently, the program needs the intervention of the user and generates error messages Adaptive tuning pybzoeram analyses the behavior ofthe process in automatic mode and computes parameter tuning. The programm is based on heuristic rules or on complex mathematical models and formulas, ‘The firs, using heuristic nules are sometimes able to adapt correctly tothe process but they need the supervision of experienced and skilled persons. Frequently, the programs compute bad tunings but if properly supervised (limits, prune >. these bad tunings will not differ too much from initial tunings. Only two manufacturers use this type of adaptive tuning, the algorithm computes improper tunings. Many industrial processes are non-linear and have hysteresis, so rarely are the users satisfied with these algorithms, TOP Control 1999 125Fundamentals of process control 4.12. Tuning with software support ‘When tuning with software, the principal is the same as with manual tuning » acquire valid and representative data in manual or automatic mode > always start with a stabilized process, make a step change (to the output in manual mode or to the setpoint in automatic mode); wait for the process to stabilize. If necessary, adjust the output or the setpoint to stabilize the process again > analyze the data and compute the tuning parameters > try the new tuning parameters. ‘The main difference between tuning with software and tuning using traditional methods is how the computer treats the data: If you acquire data and have pairs of points (PV, CO), you can use formulas to obtain the Fourier transform, This mathematical conversion will permit you to obtain a Bode plot (or Nyquist or Nichols) describing the behaviour of the process at different frequencies, Converting time series (pairs of PV-CO points) to frequency series (pairs of PV-CO points) opens many Possibilities. The frequency domain corresponds to the speed of reaction of the process, The frequency permits to ‘easily compute the tuning parameters and to make a deep analysis of the process, Today's computers have the capacity to compute these conversions (it would take weeks of work by hand!) in Seconds. The user does not need to be involved with these complex calculations. The computer analyzes, computes, ddoes the hard work and presents nice graphs and plots to the user A novice will use only the results. An experienced user will use the frequency data (graphs, equations, formulas) to earn more about the process. For example, observing the behaviour of the process in specific frequencies permits us to determine the origin of noise and disturbances. Also, the experienced user will modify the suggested tuning to achieve a specific goal orto optimise a criterion Using software fo tune industrial controllers permits us to cut the costs and time of start-ups and allows us to quickly ‘optimise the process. ©TOP Control 1999 126Fundamentals of process control 4.13. Summary Each of the control actions must be properly adjusted: > P, the proportional action deals with medium frequency and is the main control action, > I, the integral action deals with low frequencies, slow disturbance, residual errors, > D, the derivative action deals with high frequencies, fast disturbances, Increasing P (high Kp, small P.B.) will increase overshooting and cycling but the period of the oscillations will be shorter, Increasing I (high Ki, short 7) will quickly return the PV to the SP, but the overshoot will increase and the period of the oscillations will be longer. When adding I, the P must be reduced from 10 %, Increasing D high 7a) will increase stability, overshoot and settling time will be reduced. The period of oscillations will be shorter. When adding D, P and I can be increased of 20 to 50 %, 4.13.1. Comparison of tu 9 criterion ‘The same process is used to compare performance using different criteria. Four tests are made: 1 SP change of 10%, Pl controller, 2. Load change of 10%, PI controller, 3. SPchange of 10%, PID controller, 4. Load change of 10%, PID controller, Four tuning values are used; the formulas are defined for: > 4:1 decay, > LAE, > 20 % overshoot (ASP), > 0 %overshoot (ASP). ‘The resulting responses will be compared for response time, overshoot, and absolute error. ©TOP Control 1999 127Fundamentals of process control PI controller PV (%) wo mom 0 mm ww time (s) Figure 25 Setpoint change PI controller. 4 PV(%) « 0 oo Figure 26 Load change P+I controller. ‘We see more oscillations in the setpoint change. As previously discussed, a setpoint change is instantly perceived by the controller buta load change must go through the process before reaching the controler. ‘TOP Control 1999 128FOO DSTO FOOTE SST OTT IITIETETSESESESETEESSESS : > =~ Fundamentals of process control ‘The following values are found: Performances (P+I controller) ———— ee Cee ercaeleee = 5 TEE Eee nn East ee F = = 2 S & x 8 2 ® ° 2 & a = Setpoint change of 10% fo [a 393 | 050 0 2as | 357 2 | minje) | 265 | 0.43 0 1s9_| 293 3_| Ovr=20% | 2.61 | 0.73 0 180_ [55228 4 | ovr=0% | 153 | 08s 0 238 | 296 Load change of 10% | s 41 3.93 | 050 0 125] 143 [gait 6 Min {E| 2.65 0.43 0 28 adie] 95 la Ovr=20% | 2.61 | 0.73 0 29 | 168 | 148 8 | ov=0% | 153 | 085 0 34_| 230 | 286 + The aggressive tuning gives the best results on load changes but cause eycling on setpoint changes. ©TOP Control 1999 129Fundamentals of process control PID controller Sertassregeeeageee 4“ ~\ Bw ne a oo 180 700 220 aun 280 280m) H 38) a0 8 525 52 2 as @ Figure 27. Setpoint change P+I+D controller wUi~a & 180 200 za M0280 280 NaH a8 Figure 28 Load change P+I+D controller TOP Control 1999 130 DRRAOMASMROKPAPROKRAROCHRRRARARREeRRERAAAAAAaegagaeaeaeagaaaaaFundamentals of process control ‘The following values are found: Performances (P+I#D controller) Test Criterion Kp Ti (min) Td (min) Overshoot (%) Settling time 5% (%) Integral of absolute error Setpoint change of 10% 1 4:1 | 524 | 030 | 007 2 | ming | 385 [081 | 006 3_| Ovr=20% |_4.15 [100 | 0.07 4 | ovw=0% | 261 | 073 | 007 Load change of 10% 5 [4a 524 | 030 | 007 F a 6 | Mini) | 385 | os: | 006 | 21 | 163 | 110 7 | Ovr=20% | 41s | 100 | 007 | 19 | 189 | 122 8 | ov=0% [261 | 073 | 007 | - | im | 147 SA PID controller has the potential to gives far better results than a PI controller: less overshoot, faster settling time, less integral of absolute error. TOP Control 1999 131Fundamentals of process control 4.14. Definitions and Formulas Load Change A disturbance that causes the measured signal to deviate from SP, Once the CO reacts and. stabilises the process, the CO will have stabilized at a new value. Disturbance A process upset or load change that causes the measured signal to deviate from SP. Once the CO reacts and stabilizes the process, the CO will have stabilized at a new value. Noise High frequency “garbage” that is induced in a measured signal. Filter A device that tries to eliminate noise. Tier Filter Time Constant, A filter will attenuate noise with a period faster than the filter time ‘constant. Tuning Choosing PID parameters that will give the desired response goal. When tuning, one must choose between speed and stability Speed Response time Stability Maintaining control during process changes Critical damping Tuning criteria that gives the quickest possible SP change response without overshoot. Percent Overshoot Tuning criteria that allows for an overshoot on a SP change. ‘%Overshoot = H/C x 100% This tuning gives a fast response but you should expect some oscillation. Quarter Amplitude Decay Quarter amplitude decay specifies a dampened oscillation in which each successive positive peak is one-fourth of the preceding positive peak’s magnitude. LAE Minimum Integral of Absolute Error. Specifies a response that has the minimum possible error, Open Loop Method Any tuning method that is done in manual, Normally tuning parameters are calculated based on process response toa CO bump. Closed Loop Method Any tuning method that is done in automatic. Ultimate Cycling Method Closed loop tuning method that induces sustained oscillations in a loop, Tuning parameters are based on the gain that caused the cycling and the period of the cycling. Trial and Error Method Closed loop tuning method that subjectively determines the proper amount of Proportional Gain then Integral and then Derivative. Also called Tuning Without Mathematics Bump Test Method Open loop method that determines tuning parameters based on the process response to a CO bump. Also called Zeigler-Nichols method or open loop method to Natural Period. The natural period is the maximum frequency reached by a process. Kpo Ultimate Gain. The proportional gain that causes sustained oscillations in a process. thia Damped Period. Period of loop cycling observed after a SP change or a disturbance. For aggressive tuning, the period will be short. AutoTune Function in an industrial controller that allows the controller to determine the proper PID tuning parameters. Adaptive Tuning Function in an industrial controller that allows the controller to continually adjust the PID tuning parameters for optimal response. Software Tuning Using a software package to determine proper PID tuning parameters ‘©TOP Control 1999 132 BLLA2M22ADAADAMAADAAMAAAAADAAAAAAAAAAAARADAADAAADADAADADEADADEDFundamentals of process control Filer time constant should be approximately as large as the period of the noise: Crier ~ noise period| Filer time constant should be much smaller than the process dead time ft Uflter SS ty Formulas for Quarter-Amplitude Response using the Ultimate Cycling Method. Controller Formulas PID T=2 Gesonsy ©TOP Control 1999 133Fundamentals of process control Formulas for Quarter-Amplitude Response using the Trial and Error Method Controller Formulas : P only Kp=Gain Pound (gain) | PL Kp=(om Fon)x09 (gan) Timtye oon) Start with these values and then increase I (by decreasing Ti or increasing Ki) by using a {geometrical progression until i is impossible to increase it without losing the 4:1 amplitude response, Kp= (Gein Found)x1.3 (gain) PID Tifewrd Pl econ) 13 =f econ ‘Add derivative, start with one tenth of the damped period. Do a setpoint change or watch the disturbance response and see if the performance has improved over PI control. Also watch the CO, is the output moving a lot more than it was previously? If you see increased overshoot or a lot more CO variation, do not use derivative action, But ifthe response is better then increase P and I gains by 30% (increase Ky and K,, decrease PB and 7), Increase D by using a |_geometrical progression until its impossible to increase it without losing the criterion chosen. Formulas for Quarter-Amplitude Response using the Open Loop Method: Controller Formulas Kp=— P ‘P ix, (gain) Kp=09 PH P x x6, (gain) Ti=33xt, (see) P+HD Kp =1.2x—*— (gain) 1a*G, Ti=2xt, (sec) Td =05x1, (sec) ©TOP Control 1999 1342 3 Fundamentals of process control 4.15. Exercises 1-We stated previously that t th natural period is approximately four times the dead time, ultimate gain is twice the gain for 4:1 decay Using the formulas for the ultimate eycling and the open loop method, prove that fo Axl Kp Ti Ta ©TOP Control 1999 135See Fundamentals of process control : 2-A series controller is tuned with the values; a P Ko = 35 ¢ I Ti = 1.2 minutes ¢ D Ta 0.5 minutes e ¢ What are the equivalent tunings for: « e An ideal controller e P Kp = Pee @e panes @ T Th i @ D Td = e oe A parallel controller e P Kp = ¢ e 1 ee D Td = See e < @ a rf e@ a 3 a OTOP Control 1999 136 z q aFundamentals of process control 3-Find the PHI tunings for 4:1 decay using the trends below. Consider the oscillations as the natural frequency, ‘The proportional gain has been modified as follows: from 0 to 32 seconds, gain from 32 10 232 seconds, gain = 2, from 232 t0 436 seconds, gain = 4, from 436 to 560 seconds, gain from 560 to 772 seconds, gain = 6, from 772 seconds to the end, gain = a mearaccn PY (%) s = : \ : V ° « ° 20 co 0 0 1000 time (s) Figure 29 Trends, problem 4. ‘amplitude decay P Geese eeaeeaet rain % ; : sec __RPM GTOP Control 1999 137Fundamentals of process control 4 - The ultimate cycling method was applied to a flow loop and the following trace recorded: Ultimate Cycling 7 J 53 — [| 46 — - 01234567 8 9 101412131415 16 17 18 19 2021 2223 Time (Sec) ‘The controller settings were as follows: P= 3 gain 99999 sec O sec Auto 50 Determine tuning parameters for % amplitude decay and 10% overshoot: amplitude decay 10% overshoot P P gain gain % % T T sec. | sec D D sec sec F F | sec sec a ©TOP Control 1999 138Fundamentals of process control 5 ~ Determine tuning parameters for Yi amplitude decay using the open loop method Controller Output Percentage 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 Time (See) Process Variable Bea [aay [ [ vo} LOA | Z a Percentage elf | a [PR ° LCE CPP 0 2 4 6 8 10 12 16 18 18 20 22 24 26 28 30 Time (Sec) amplitude decay P gain ©TOP Control 1999 139Fundamentals of process control 4.16. Computer Exercises 1- Determine the PID tuning parameters for the simulated flow loop in ExperTune using the Ultimate Cycling, method! a. The flow loop I and D actions are completely eliminated (or their gains are reduced to minimum). K,= 0, very large, T= Ka= 0. b. __ Imautomatic, the proportional gain is increased until cycling is sustained. Ifthe gain is properly chosen, the oscillation will not increase nor decrease in amplitude. The period of the oscillations is the natural period, t). ‘The proportional gain that caused sustained cycling is called the ultimate gain, Kpo. If working with Proportional Band instead of Proportional Gain, it must be decreased to induce cycling. ‘The setpoint changes done to find ultimate cycling should be in the order of magnitude that the process normally sees. Archive data to get an accurate measure of the natural period. c _ Determine tuning parameters based on the formulas below. ‘The formulas will give approximately 4:1 amplitude decay response. For less aggressive tuning, divide the final Proportional gain found by 2 or 4 4. Test the new parameters by doing a SP change, e Cut in half and then double the proportional gain, doing a SP change after each. What are the effects? Formulas for Quarter-Amplitude Response using the Ultimate Cycling Method: Controller Formulas Ponly Kp A? (gain Kpo PI kp (gain) att Tiz ‘sec) i=75 Gee) - pe sain Kp = (gain) PID nee if (oe Ta= 42 Gee) ‘©TOP Control 1999 140 DBAADVMHRAHAADARDIOHRARAHHOKRARHRRAAARARATRLARARALEFundamentals of process control 2 - Determine the PID tuning parameters for the simulated flow loop in ExperTune using the Trial and Error method: a The flow loop controller I and D actions are completely eliminated (or their gains are reduced to minimum). K; = 0, Ti=:or very large, Ts= Ks=0, b In automatic, the proportional gain is increased until a 4:1 amplitude decay response on a setpoint change is ‘observed. The period of the quarter amplitude cycling is the damped period, ty. Record the period and the gain that caused this response. See Figure 8 “4:1 Amplitude Decay” for a picture of this response. If working with Proportional Band instead of Proportional Gain, it must be decreased to induce cycling. ‘The setpoint changes should be in the order of magnitude that the process normally sees, Archive data to get an accurate measure of the natural period. c. __ Determine tuning parameters based on the formulas below. ‘The formulas will give approximately 4:1 amplitude decay response, For less ageressive tuning, divide the final Proportional gain found by 2 or 4. 4 Test new the parameters by doing a SP change. e Cut in half and then double the proportional gain, doing a SP change after each What are the effects? Controller Formulas P only Kp=Gain Found any PI Kp=(Gain Found)x09 (gain) Ti=tyy (see Start with theses values and then increase I (by decreasing Ti or increasing Ki) by using a geometrical progression until itis impossible to increase it without loosing the 4:1 amplitude response Kp=(Gain Found)x13 (gain) nell fend Ph ie PID St 10 ‘Add derivative, start with one tenth ofthe damped period. Do a setpoint change or watch the disturbance response and see if the performance has improved over PI control. Also watch the CO, isthe output moving a fot more than it was previously? If you see increased overshoot or a Jot more CO variation, do not use derivative action. But if the response is better then increase P and I gains by 30% (increase K, and K,, decrease PB and T}). Increase D by using a geometrical __Drogression until it is impossible to increase it without loosing the criterion chosen. (ee) TOP Control 1999 41Fundamentals of process control 3 - Determine the PID tuning parameters for the simulated flow loop in ExperTune using the Open Loop method, a Determine the process model for the loop as per exercise 1 in section 2.8 b Determine tuning parameters based on the formulas below. c Test the new parameters by doing a SP change. a Cut in half and then double the proportional gain, doing a SP change after each. What are the effects? Formulas for Quarter-Amplitude Response using the Open Loop Method: Controller Formulas P PH Ti=33xt, (seo) PHD Kp=1.2x— G (gain) T=2x1, (00) Td =0.5xt, (sec) : ‘GTOP Control 1999 12 DOOO099000200209290AVOOCOHAODAOOO2200008ARAAAAAAAAADADFundamentals of process control 4 - Determine the PID tuning parameters forthe simulated flow loop using the ExperTune software package a. Archive atest in auto or manual, Make sure o archive data that has “stable - quick change - stable” on both the PV and CO. b. Analyse the data to make sure the results are good, What is the quality of frequency data fit (“PID Grid” button)? What is the process model found (“Analysis button)? Do you recommend using a Filter (“Use Filter” selection box)? Do you recommend using Derivative or not (“Use Derivative” selection box)? © Change the safety factor to 1 to determine aggressive tuning parameters, How do these parameters compare with the other methods tested? 4 Test new the parameters by doing a SP change ©. Test the filter suggested by Expertune and observe the resuts. ©TOP Control 1999 143,Fundamentals of process control 4.17. Notes ©TOP Control 1999 144Fundamentals of process control 5. Process Problems ©TOP Control 1999 145———— eee ee ee oe Fundamentals of process control 5.1. Process Variation The previous chapter showed that tuning a PID loop is not rocket science, It can be fairly easy if you have formulas ora sofware package to help. The tough pat about tuning a loop is determining how “perfect” itis. A perfect control loop will have the following characteristics: > Process gain, G,, near 1 and constant G, > Small dead time, ts and constant t, > Large time constant, + anda constant t > Very little noise If for example you tune a loop and determine the following 10 seconds t= 10 seconds Using Ziegler-Nichols formulas would give the following tuning parameters and response toa SP change eo reieuaeeaesat XG, “T0x1 1=2x1,=2x10= 20sec D=05xt, =0.5x10= 5sec .2xK, =12x Figure 1 Tuning and Response ‘This tuning is very fast, but not very robust. If for some reason the dead time were to double the next day, your process model is now: G=l ta= 20 seconds = 10 seconds Instead of using the gains above, the following tuning should have been used: P=12xK, =12x—2_ 0.6 1,xG, T=2xt,=2x20= 40sec D=0.5x1,=0,5x20= 10sec Robustiess isa measure of stability in the face of change. Figure 2 shows that Ziegler-Nichols tuning is not robust; ‘when the process dead time doubles the response becomes unstable. O10? en deFundamentals of process control Since the incorrect tunings were used, the response to a SP change becomes unstable: P= I= D= 12 20sec Ssec Figure 2 Response becomes unstable due to change in dead time In this chapter reasons for process problems and non-linearity’s will be discussed. Methods for testing the process and equipment will also be discussed. Aggressive Tuning Process Model a G,=1 1 = 10sec ta = 10 see ‘Tuning used P=12 * I D sec 0 see eee eee Benes Process Model G=1 0 sec 10 see ‘Tuning used Robust Tuning “ Dead time changes - ty = 20 sec Tuning same as above: P12 1 =20sec D Figure 3 Aggressive vs. Robust Tuning when the process model changes ‘F When tuning a controller, any possible variation in the process model should be uncovered. Tuning should be based on the worst process model found. When in doubt about a processes linearity, choose more robust tuning. TOP Control 1999 148, EARALELE. oeFundamentals of process control 5.2. Varying Time Constant ‘There are many potential reasons why a process time constant could vary. In gener, time constants can vary throughout the process range or be asymmetrical (differ from increasing PV to decreasing PV) If we have a variable time constant, x, we should base our tuning onthe smallest time constant found. This may Seem counterintuitive at first since you may think that a small process response time (or time constant) would be advantageous. It may be quicker to ill up a coffee cup with a garden hose than a swisnmnng pool, but which ase o.mantan at $096 full? The swimming poo of couse! If someone takes a drink out of the cote cup the level has dropped to 50% full, if someone takes a drink out ofthe pool, the level only drops by a traction of percent 5.2.1, Variation with process {some processes, the time constant can increase or decrease linearly or nonlinearly with increasing PV. In the Sel show below the horizontal tank the time constant is greatest at 50% level. As the level rises or drops the fim constant gets smaller, If te anki sometimes kept at 80% fll and sometimes at 0% fll, the tuning parmiers shouldbe determined when the level i t 80% sine the time constant is smallest there. Howeser, ifthe {RK always kept at 50% full. and disturbances cause the level to drop or rise to 60% then the tuning should be based on the time constant tha is found at 60% level. ©TOP Control 1999 149 aFundamentals of process control 5.2.2. Asymmetrical Processes Some processes have a time constant that varies depending on the direction of the process variable, For example the ‘temperature of a tank of water. If you add heat with steam, the temperature will rise quickly. However, if you cool the system only by removing the steam, it will cool very slowly. Cooling — Large t i 2] iW By Ambient Air (Cool) ‘co Figure 5 Heating vs. Cooling Another common example ofthis behavior isthe opening and closing characteristics of a spring and diaphragm valve. When moving the valve in one direction, you add air ata pressure expressed in PSI, when moving the in the other direction, you evacuate the air using a spring, If the spring force does not exactly equal the fore’ ‘generated by the air pressure, the valve will have different opening and closing times, For a flow or press application, the process dynamic is primarily the dynamic of the valve. ‘Thus the different opening-closing translate into a varying process time constant. @ {fa variable time constant, 1, is present, tuning should be based on the smallest time constant found. PREPARA AALARE ‘©TOP Control 1999 150Fundamentals of process control 5.3, Varying Dead Time ‘There ae several processes in which the dead time can vary. For example, in most analytical measurements such as DH, conductivity, or consistency the measurement is taken in a sampling line. Ifthe measured value is below setpoint, the controller opens or closes a valve to correct for it, but the effect of the correction will not be seen until the additive has traveled through the process and the sampling loop. So the dead time is a function of distance and flow rate. Anytime there isa disturbance in the flow rate ofthe sampling line, the dead time varies F Anytime we have a variable dead time, ts, we should base our tuning on the largest dead time found. Thickening Agent ‘Variable speed drive - Product flow varies from 1-2 f/se, ‘Dead time varies from 75 to 150 sec. Figure 6 Consistency Loop | Thickness { Measurement Adjustable ; Speed i (CO ) ; qagRA0 ~~ bbe | Speed Press Dryer Thickness Winder Control Measurement Figure 7 Dead time as a function of conveyor speed ©TOP Control 1999 151Fundamentals of process control 5.4, Inverse Response ‘The PV ina process with inverse response will initially tend in the opposite direction of where it will end up. For ‘example on a level loop with inverse response, if we open the valve, the level drops initially but eventually rises and stabilizes at a new higher level ‘One example of an inverse response process is found in the steam generation section of a boiler. Water is added to control boiler drum level. The steady state result of adding more water is an increased drum level, however, when first added, the relatively cool water collapses some of the steam bubbles, lowering the tank level. The drum level initially responds in the opposite direction to its steady state level Another example of a process with inverse response is the level control at the bottom of a distillation column, A process with inverse response is also called a “non-minimum phase process” due to the shape of its frequency response. It is also sometimes called a “shrink-swell” process. When inverse response is present in a system, itis imperative that derivative not be used. Derivative anticipates the process response. When the level drops slightly, the derivative anticipates thatthe level will keep dropping which may not be the case. If derivative is used the controller could inadvertently flood the tank. ‘Steam Out Figure 8 Boiler Drum Level TOP Control 1999 152SOSCCOSSSSCESSESTC ESOS oS 2 o io 2 e o o > D> 8 Fundamentals of process control 5.5. Saturation Saturation occurs anytime we want to go faster than a limit. For example if I rey my car engine up to 6000 RPM, it vill handle this without any problems (I hope!). However, if put my foot further down, the engine will not rev faster as there is a cut-off chip in the engine controller. I have not gained anything by driving the peddle into saturation, it just takes longer to remove my foot when I decide to stop revving the engine, To avoid adding dead time to the process, you should limit each PID CO to the range that it affects the process. For example if you have a through por ball valve that has almost no flow until itis 15% open, limit the CO to 15-100% Otherwise the time that the controller takes to open the valve from 0 to 15% is added dead time as it has no affect on the process. Pipe Figure 9 Ball valve opening ‘The first type of limit discussed was a cut off. There are also speed limits that we need to be aware of. For example if a large pulse is sent to a variable speed drive (VSD), the drive internally has many user set-up limits: > Speed Limit Current Limit Voltage Limit Torque Limit Aceeler: ion Limit vyvvyy Deceleration Limit Because of these limits, the response to the pulse will be relatively slow compared to what the drive can actually do. fa small pulse is sent to the drive, the limits may not be reached. As such, the response is much quicker. In effect Wwe have a process that has a large time constant for large errors and a small time constant for small errors. Ifthe technician uses a large bump to tune the drive, he/she will find a large time constant and tune the drive aggressively. ‘The response will look good for a large SP change but will probably oscillate once at SP Avvalve also response differently to large and small bumps, but usually in the opposite way. A large bump causes the valve to respond quickly whereas a small change in position takes alot longer to do. In this case if the valve is tuned with moderate to large bumps, it wil stil be stable (but slower) for small changes. TOP Control 1999 153Fundamentals of process control : X a * Fast Response 2% \ (short time constant) re Slow Response ze ong time constant) ze B13 +d o 19 2 © 300 so 8070 Time (see) Figure 10 VSD — Large vs. Small Changes Figure 10 shows that the VSD responds with a long time constant on a large bump and with a short time constant on a large bump. In this case the loop should be tuned using the second smaller bump. TOP Control 1999 154Fundamentals of process control 5.6. Hysteresis, Backlash and Deadband ‘There is no such thing as a perfect mechanical connection, The control elements used in process control have many interconnected mechanic parts, leading to control problems. One such problem is Hysteresis, We will define hysteresis to encompass any backlash and deadband. Backlash occurs when we reverse the direction of two connected parts. For example when we reverse the direction of two gears, there isa small distance that the drive gear must travel before it engages the driven gear. Even in near perfect gears, there will always be a small amount of backlash 5 Deadband is a zone in which nothing occurs. Its similar to backlash, but e different in that it can be minimized in most couplings. Normally when we buy a coupling there is very little deadband, however with wear it builds up Drive Daven Rod Motion [zee SS+ [aS New Used Rod Motion JES Figure 11 Deadband and Backlash ‘One way to test valves and dampers for hysteresis is to stop the process and bench test them. This gives and accurate reading but is costly. A simpler and cheaper way to test for hysteresis is to measure its effects through the Process. If there is slop in the valve it will show up when we bump the valve in one way and then the other. ‘The caution is that you are not just measuring the hysteresis in the valve or damper, but also the UP, positioner or any ‘other piece of equipment in the 4-20mA loop. ©TOP Control 1999 155Fundamentals of process control ‘The hysteresis test is pictured below. 1, Im manual - the valve is moved in one direction, just enough to see a response in the PV 2. The valve is moved again inthe same direction fo calculate the true process gain, G,. This must be the tric gain since all ofthe deadband and backlash was removed from the valve in step 1 3. The valve is moved in the opposite direction, enough to see the process respond. ‘The difference between how much the valve was asked to move and how much it affected the process i the hysteresis, 4 2 ’HHysterisis = Deadband + Backlash 7 B 4Hyst =Valve Commanded Valve Actual B 24 7 2 %PY moved (step 3) Hyst = % Valve Commanded (step 3) ~ “PY moved (step 3) ‘ieG, (step 2) B 3 _y 35% 50 219-22 % 2.67% 24 15 arteaal 2 al Be on DDH DH DH OM Figure 12 Calculating %Hysteresis Unfortunately a new valve will display some hysteresis. 0.1 to 1% is typical, depending on the type of valve and ‘manufacturer, If there is more than 3% hysteresis in your control element, you will not have a performant control loop: cycling, response time and overshoot will be increased. enna ee —TT Fundamentals of process control 5.7. Stiction Stiction is an invented word that covers two characteristic, stickiness + friction = stction, When a large object is ‘moved, it takes a lage initial force to stat it moving, however, once in motion the object moves more easily, Once set in motion, the friction force between the object and the ground decrease. For example, if someone asks you to ‘move your refrigerator exactly % of an inch, you probably will not be able o. You push on the fridge and it docsn't ‘move, so you push harder until finally it moves, but by the time you stop pushing it has moved 2 inches, When stiction becomes large in a valve its effects can be disastrous. Medium push — No motion Large push ~ Moves too far Figure 13 Stiction in Automatic Ifthe process variable is slightly off of setpoint, the controller integral action will slowly increase to remove the error, However, the valve stiction does not allow it to move until the force builds up. Once in motion the valve has Joved too far and the integral action corrects in the opposite direction. The result is eycling in the process. If the integral gain is decreased the period of the cycling increases, but the magnitude stays the same All valves have stiction, luckily the magnitude and effects are not usually significant. Approximately 4 of the Yates ina plant will have significant enough stiction tobe observed with a good chart recorder. You will probably WoL es ion an operator interface screen. The trademark sign of stiction isa square wave on the PV and a triangular wave on the CO. | 69 178¢eeycing in the process al ee ak at ig 6 0.9% Stiction ! Figure 14 Stiction in Automatic gt ke a —I Tine eo) ©TOP Control 1990 aeFundamentals of process control To test for stiction, we should induce a ramp on the CO and see how much CO change it takes before we see a process response. Since this is not possible with most controllers we simulate the ramp with a series of small ‘bumps. To ensure that we are not measuring the deadband, we do a large bump first. In summary In manual, bump the CO enough to see the PV move. Wait until it the PV stabilizes, Do a very stnall bump (0.1%) in the same direction as step 1 Wait and see if the PV moves. Ifyou do not see movement repeat steps 3 and 4 until you see the PV respond. ‘This method will give you an approximation of how much stiction the valve has, For example if you made five 0.1% bumps before you saw a response, the valve has between 0.4 and 0.5% stiction ] ] PV ie %) “ 7 23 nik D1% 01% bebatie co "4 Time (sec) Figure 15 Calculating %StictionFundamentals of process control 5.8. Non-Linear Process Gain ‘The most common cause of a varying process gain is poor choice of valve. As such, a brief explanation of valve type and characteristics will be given. 5.8.1. Control Valves — characteristics and problems A control valve normally consists of three parts: > Positioner > Actuator > Valve A current to pressure transducer is sometimes the fourth part in the final element loop, but will not be discussed. Positioner The job of the positioner is to ensure that the valve moves as requested by the controller. Ifthe controller sends a 30% signal to the positioner, it will send an increasing signal to the actuator until the valve is at 50%, ‘There are many popular positioner algorithms on the market, some are even P, PI or PID control. As such, the positioner must bbe properly set-up to avoid large overshoot and oscillation, but not be so slow as to increase the process dead time and time constant. ‘Some of the newer positioners will also try and overcome stiction and hysteresis, but as of publication date, none have completely succeeded Actuator ‘The actuator is the muscle used to open and or close the valve. Normally the force required to open and close the valve is calculated during the design phase and an actuator that provides slightly more force than required is selected. The actuator is oversized to account for increased friction over time or errors in the calculation, ‘The rule of thumb in industry is 1.5 times the force required, however, this is often not enough several years later. If the actuator is not strong enough, the valve will take a long time to open and close, causing an increased process dead time and time constant Figure 16 Valve, Actuator and Positioner GTOP Control 1999 159Fundamentals of process control Valve Valves are generally classified by their construction: “Globe” of “Rotary”. Class Globe Valves Rotary Valves Full Ball Valve Segmented Ball Valve (V-Ball) Butterfly Valve High Performance (Eccentric Butterfly) Members of Class Characteristics Expensive Heavy Large actuators required ‘Less force required Used less often More common Low noise More noise Less cavitation More cavitation Large pressure drop Small pressure drop Flow Characteristics Quick opening Linear ‘Available Linear Equal Percentage Equal Percentage Modified (between linear and equal) ‘The selection of the proper control valve for an application is the subject of many other books, From a control ‘standpoint, we are most concerned that the right flow characteristic is chosen. The flow characteristics for constant pressure are pictured below Equal Percentage Valve Opening _| Flow Figure 17 Flow Characteristics for Constant Pressure As is highlighted, the characteristics are for constant pressure. Ifthe pressure drops with valve opening, the installed characteristic will be the sum of the pressure curve and the characteristic curve, ©TOP Control 1999 160 RARAURAARARAARAERREREREREAASD DAHKRRAAARRARRAALRRDRRRERE.Fundamentals of process control Constant Pressure Dropping Pressure 7 (mr) et) —— e Figure 18 Linear Valve - Installed Flow Characteristics ‘As can be seen in the diagrams above, the proper choice for a constant pressure application is a linear valve, The proper choice for a pressure that drops off is an equal percentage valve, If the incorrect valve is chosen, the installed flow characteristic is non-linear, leading to non-linear process gain ‘F The control valve flow characteristic should be chosen such that the installed flow characteristic is linear. This will ensure that the process gain is near linear. ©TOP Control 1999 161Fundamentals of process control 5.8.2. Solutions for non-linear process gain ‘When non-linear gain is present the first question to ask is how non-linear isthe gain over the operating region? The ratio of the maximum gain over the minimum gain will determine the solution, G. pio G pan Tune at point of highest gain, Gpmax G, 7° <3 Tune at point of highest gain, Gyug. Gym <2" > 3. Live with dissimilar response times or Linearize the system [If the ratio of max gain over minimum gain is less than or equal to three over the operating range, we can “de-tune” the loop and still get decent results. De-tuning means to determine aggressive tuning parameters atthe point with the highest process gain, the response will be adequate at this point but will be approximately two to four times slower at the point of minimum gain. Response at high process gain (~ 15 seconds). Py © | [Response at low process pain (~ 45 seconds). % ee Live with dissimilar response times Figure 19 Response After Detuning {If the ratio of max gain over minimum gain is greater than three over the operating range, we can still “de-tune” by ‘tuning at the point of highest process gain. However, the response time will now vary quite a bit. ‘TOP Control 1999 162Fundamentals of process control Linearize the system ‘There are several ways to get constant response time in automatic with a non-linear system. Depending on the type of controller and valve used, not all ofthe solutions will be attainable. 1. Apply different tuning parameters based on operating point ~ This solution may seem easy at first, but will probably take some thought and programming. One important point is ensure thatthe transition point to and from one set of parameters to another is not the same point. For example, switch to P=1, I=10sec when PV becomes greater than 30%, switch to P=0.5, I=10sec when PV becomes less than 25%. This will avoid toggling back and forth. Another way to get the same result isto switch tuning based on the SP and not the PV. 2. Apply a characterizer to the controller output — A characterizer takes an input and calculates an output based ona lookup table, formula or series of linear translations. This isthe best option as it allows for one set of tuning parameters and will give a linear process response in manual as well as automatic (if programmed properly). Applying a characterizer is discussed in section 5.8.3. Replace the positioner with an intelligent positioner — This option will give the same results as applying a characterizer. The difference is that the characterizer function is in the positioner instead of in the controller or program. This will limit the access to the characterizer for fine tuning and will also necessitate a buying a new positioner. 4. Use a cam to characterize the valve positioner This is often the most common solution used in industry; the system is non-linear so just keep trying different cams until it is closer to linear. Although a cam may be found that will linearize the process, the positioner now has a non-linear cam to control the position of. This could cause instability in the positioning of the valve. Replace the valve by one with the proper flow characteris — Expensive but sometimes the only option. 5.8.3. Applying a characterizer ‘A characterizer takes an input and calculates an output based on a lookup table, formula or series of linear translations. By graphing the process response, we can define a characterizer that will take a linear input and output the inverse of the process response, The effect will be a linear process response, ‘This will allow for one set of tuning parameters. The first step to building a characterizer is to determine the process gain at different operating points. You can do this in either auto or manual, Plot 5 to 10 PV (y-axis) versus CO (x-axis) points on a graph. Ifthe points line up in a straight line, the gain is linear. The process gain is the slope of the line(s) that connect the points. ‘The characterizer will have the mirror image shape of the PV versus CO plot; at points where the gain is small, it will output a large value, where the gain is large, its output will be small. ‘The number of segments or curved lines that the characterizer will use will depend on how critical the process linearity is and what is practical to implement on the control platform. ©TOP Control 1999 163,Fundamentals of process control Large Gp Input to characterizer (%) + ae * POE | CO(%) | Output to valve (%) Figure 20 Building a Characterizer ‘©TOP Control 1999 164Fundamentals of process control 5.9. Noise Noise is high frequency “garbage” induced in the measurement ofa signal. A variety of reasons could cause the noise: the measuring device itself, improper routing of cables, improper grounding. ete. Before tuning a controller, itis important to properly adjust the filter to eliminate the high frequency noise on the PV signal. The basic idea is to choose the time constant soit will not increase the dead time but it will reduce the noise sreatly ‘When adding a filter, you may be presented with a number of choices: > Where to add the filter: measuring device, input card, in the program, in the PID block > Add deadband in the measuring device instead of filtering > Create filter through programming > Type of filter to add or create: first order, second order, Butterworth, moving average As far as filter location is concerned, a filter in the field will do as good a job as in the controller. However, the further away the filter is, the easier itis to forget about. The filter time constant should be validated every time the loop is tuned, if the filter is in the instrument, it is often overlooked. Deadband is not really a filter. Adding a lot of deadband will give a noiseless signal, but will also drastically increase the process dead time. As well, any noise ofa magnitude large than the dead band will ill be “noisy ‘The only time that deadband should be used is ifthe valve life is more important that the process being controlled Tre signal Signal after deadband is added Figure 21 Effects of Deadband When adding a filter through programming, moving average is most commonly used since itis the easiest to rogram. However, it gives the worst response to random noise. A good filter should not only attenuate noise, but ‘add very little dead time to the signal. The best choices to meet these criteria are Butterworth, Sccond order. First order, and moving average (in that order). The algorithm for first order and moving average filters are given in figure 16. The response ofthe filters are shown in figure 17, Moving Average First Order py ( sum(last n—1 samples) + Raw Data Sample ) PY = PV + (Raw Data Sample ~ PV) & n - t= sampling time t= filter time constant ty sampling time filter time constant tr Figure 22 Moving Average and First Order Filter Algoritms TOP Control 1999 165gecaeace Fundamentals of process control es [Er § Hl owing ee | e hae | | | ; ee, 1s | 7 / : hie : a ae ya AAI a = @ secede / 7 AW . : Filter 5S 4 : pac Order [~ ne eal | Figure 23 Filters and their Responses It should be noted that when noise of a know frequency is trying to be eliminated, and not random noise, the moving average filter is the best choice, In this case the filler time constant should equal the period of the noise and the ‘sampling time needs to fast enough to allow for atleast 8 samples in this period. Figure 24 Moving average filter response to non-random noise TOP Control 1999 166Fundamentals of process control 5.10, Recommended Test and Sequence The following section is what can be considered as “the full tuning test". This procedure would take too much time to do on every loop in a plant, but the quick and critical loops should follow this method ‘The procedure will be broken down in four sections 1, Prior to connecting to the signal 2. Testing the equipment and current tuning parameters 3. Testing the process 4, Tuning and testing the results 5.10.1. Prior to connecting to the signal efor attempting to tune a loop, you must gather information about the loop. Another name for this section could be, “doing your homework on the loop”. You should be able to answer the following questions before tuning any Joop’ Fs the loop normally left in auto or do the operators control it manually? S@ What oops interact with this one? Will oscillation inthis loop be detrimental to the process? SB” How quickly does this lop need to react to disturbances and setpoint changes? SF” Does this process always need tbe tightly at setpoint? 7 What isthe age, make and model ofall components inthe loop? ‘Fis the contol for this loop pure PID or is there logic assisting the contol? What isthe strcture ofthe conrotter? or Is the controller properly set-up? he Pest way to gather this information is by asking questions to the people who are most femiliar with the process and by going through whatever documentation is available (P&ID diagrams, SAMA"s, lnsisumest loop diagrams, etc). 5.10.2. Testing the equipment and current tuning parameters ©TOP Control 1999 167Fundamentals of process control How does the loop respond to the SP change? Is it fast but oscillatory? Is it too slow? Is it oscillatory and slow? If the loops is not to0 slow, a SP change back to the original SP should be done. This will allow you to see ifthe loop has any of the process problems that you can determine in automatic (discussed in section 5.11.1). During this test ‘atch how much the CO moved to accomplish the SP change, this will give you a rough idea how much you can bump the CO once you switch into manual, Before switching into manual, you should record some data in steady state, You may see visible cycling in the loop or stiction, Next the equipment should be tested, Switch into manual and test for stiction and hysteresis as was explained in sections 5.6 and 5.7. If the equipment has too much stiction or hysteresis, the loop will not be performant even after tuning. As such, the equipment should be repaired before continuing, Automatic Manual ‘SP change, before tuning Steady state “before” data co Stietion (%) test HYysterisie test so ot 7 1 er tr te 7 7 Time (sec) Figure 25 Testing the current tuning parameters and equipment 5.10.3. Testing the Process If the equipment is in good working order, then we can test the process for linearity and asymmetry as detailed in sections 5.8 and 5.2.1, In the ideal case we would be able to test the process over every possible PV value thatthe Joop will ever see. In reality, we will only be able to move the PV by as much as the process and those that it affects will allow (or as much as the operator will allow), ‘©TOP Control 1999 168 REGGRARARRALARRAGRRRERRREEREREAReFundamentals of process control Manual - Linearity Test py (%) co. 0) ‘Time (sec) Figure 26 Testing the Process 5.10.4, Tuning and testing the results ‘Based on what was uncovered when testing the process, there are a few options available IF the process is fairly linear and asymmetrical, then select the bump that gives the worst process model and tune based on this bump, ‘The aggressiveness that you use wil be based on how confident you are that this bump istrly the worst process model. If you did not get to test over the full operating range, it is not prudent to tune aggressively Tf the process is non-linear then a characterizer will be needed for performance, Depending on the controller that is used, this could be an easy or impossible task. Xf the process is highly unsymmetrical, then to get performance, separate tuning parameters will be needed for each Process model identified. Depending on the controller that is used, this could be an easy or impossible task. Once the tuning parameters, sets of tuning parameter, or characterizet has been implemented, it is time to test them. At a minimum you should compare the new tuning tothe old, doing the same SP changes that were done in section 3.10.3. This wll allow you to compare the results ona level playing field. Ifyou do not get the results you Fane aed, retune or modify the proportional gain, To fully tet the new parameters, do a series of SP changes and induce a disturbance in the process if possible. ‘aut should also observe the new tuning parameters in steady state, Is there cycling induced by the new parameters? Are they aggressive enough to eliminate the disturbances that are present? TOP Control 1999 169Fundamentals of process control ; cstnentvanvanene | | ‘ a ] le “Tine es) “Time (ee) Figure 27 Testing and Comparing the Results 5.10.5. Recommended test summary Ona quick loop, the entire testing time can be under 15 minutes. The time to analyze the data could be minutes to hours, depending if you use a software package or not and also how non-linear the proces is. Automatic Manual Automatic 2 change Hystrsis SP change, Pv | aoe test Stiction er" Steady state | ) Steady sate text Lineaiy ne sea | “before” data co cor | Beery ‘Time (sec) Figure 28 Full Test of the Process GTOP Control 1999 170Fundamentals of process control 5.11, Summary ‘This chapter showed that tuning a loop also means: > Choosing the proper performance criteria > Choosing the proper filter > ‘Testing the equipment in the loop > Uncovering non-linearity’s in the process 5.11.1. How to recognize problems in automatic Many of the problems discussed can be identified by analyzing the performance of the loop in automatic, The following are the pictorial examples: An asymmetrical process will have a different response time or shape for an increasing versus decreasing SP change in auto. 4 t Figure 29 Asymmetrical Process . roms ] A non-linear process will have a & I different response time or shape as the | esteem | process is stepped through the operating range in auto, woth —y — + * + Tine Figure 30 Non-Linear Process ©TOP Control 1999 71Fundamentals of process control Proves Vane fon the same scale, the PV moves alot, for a small change in CO, the process has a high process gain ‘Contr Outpat Seale Tee) Figure 31 High Process Gain Ifa final element has stiction, cycling can be observed in the PV and the CO. ‘The shape will be a square wave on the PV and a triangular wave on the CO. TOP Control 1999) 172E : E P o : Fundamentals of process control Hysteresis can be seen (but not quantified) in auto in one of two ways, 1. Fora SP change up and then back down to its original value, the valve should return to about the same position (unless a disturbance ‘occurs at the same time). If it does not return to the same place, the element has hysteresis, Tine (es) Figure 33 Hysteresis in Auto 1. Ifthe loop is tuned aggressively, hysteresis can be observed by the increasing period of oscillation on &Y the PV. ay Tine Figure 34 Hysteresis in Auto with Aggressive Tuning 5.11.2. Acceptable limits ‘There are many specs published for how much deterioration valves and process equipment can have before they need to be repaired or replaced, ‘The specs, although well intended, are often more strict than need be and would a2 = > > ee = oy a 2 od a 7 o . a 2 > 2 > > > > > > > > > > > » > , , . sar)” process ~ Click on *No” when prompted Select “Temp Non-linear)” (from the process type list) c Adjust the faceplate screen as tequired, and archive a “Full tet” as described in section 5.10. The PV can never go outside of the range 10-70 % during the testing, 4 Once done testing, stop archiving and analyze the data, Determine tuning parameters to meet the criterion above, test the results and answer the questions on the following page OTOP Control 1999 183Fundamentals of process control Questions 1. Girele the descriptor(s) that best describe the setpoint change response before tuning Cycling Slow Fast Unstable Stable 2. What is the response time to a disturbance or setpoint change before tuning s 3. What is the valve Hysteresis _%. 4. What is the valve Stiction: HM, 5. What is the process gain, G,’ ay 6. Whatis the process dead time, ts) __s. 7. What is the process time constant, +:__s. 8. Tuning Parameters determined: Pe D= __ sec sec F= __ see 9. Values before and after: Valve travel index(response to noise): Performance increase: % Robustness increase: % 10. Circle the descriptor(s) that best describe the setpoint change response after tuning, Cycling Slow Fast Unstable Stable 11, What isthe response time to a disturbance or setpoint change after tuning 12. Test your new tuning parameters. Compare with predicted (simulation in analysis). ‘TOP Control 1999 184Fundamentals of process control 5.15. Notes ©TOP Control 1999 185Fe ee ee ee ee ee Fundamentals of process control 6. Advanced Control Strategies ©TOP Control 1999 187Fundamentals of process control 6.1. Introduction — A loop is rarely alone APID loop rarely acts alone. Normally it feeds or is fed from another control loop. Tuning loops as if they were independent is one sure way to wreak havoc on plant production. ‘This chapter will > Give principals that determine whether loop response should be tuned to match or differ from the speed of loops around it. > Discuss ways to adapt PID control to increase performance or make the control better suited to production needs. > Discuss and compare “advanced” control strategies. These are strategies that will enable better performance in instances where PID control alone is not adequate. Team HEADER Figure 1 Typical PID - Many Loops Interacting Together TOP Control 1999 188SOC SOC CSC SSS SUS UL 2 = o 2 =D 3 2 > Fundamentals of process control 6.2. Multi loop Unfortunately, when many Toops are inthe same system as in figure 1, there is no magic formula that tells you if one Joop will affect another or not. This knowledge will oly come through knowledge of the process. If one loop directly feeds another, oscillation in the first loop will cause oscillation inthe second and possibly any other downstream loops. If two flow loops are fed from the same pump, oscillation in one loop could cause oscillation in the second. Other times, loops do not interact, but itis imperative tht they respond with the same speed 6.2.1. Interacting Loops Figure 2 shows a solution flow loop. Two liquids are brought together, The top flow is an expensive product and is ‘maintained at 100 USGPM. The lower flow is water; it is adjusted to give the total flow required (between 200 and 400 USGPM), L-~(re2)----- Figure 2 Interacting Loops All three of these loops have the potential to be fast. A response time of less than 30 seconds is attainable on all tree loops, but which loop needs to be fastest? |nthis case, the pressure loop must be faster than flow loop #1; its job isto maintain pressure forthe flow loop regardless of what the flow is. Ifthe pressure loap and flow loop #1 are tuned at the same speed, they may work for while, bt eventually a disturbance will occur that will cause the two loops to oscillate For example 1, A disturbance causes the line pressure to increase 2. If the pump is too slow, the flow will increase 3. The valve closes to compensate, causing the pressure to increase 4. The pump eventually slows down causing the flow to decrease 5. The valve opens to compensate, causing the pressure to decrease 6. The pump speeds up, causing the flow to increase 7. Steps 3, 4, 5 and 6 repeat continuously ~ the two loops will oscillate and potentially resonate ©TOP Control 1999 189Fundamentals of process control Likewise, since flow loop 1 feeds flow loop 2, flow loop 1 must be faster than 2. Otherwise a disturbance in flow I could cause both flow controllers to react — oscillation would result. ‘When loops interact together, we need to select response speeds that differ. The exact same speed could cause oscillation. Speeds that differ but are close have the potential to oscillate. To be safe we sliould choose response Speeds that differ by a factor of 3 105. If a speed difference of less than 3 is selected, the loops may start to oscillate one day. When loops are highly interactive, to de-couple them, a speed difference of up to 10) may be required For the system pictured in figure 2, this would mean response time of the pressure loop determines the response ti of flow loop #1. ‘The response time of flow loop #1 in turn determines the response time of flow loop #2. ‘The Procedure for tuning this system would be to put the downstream loops in manual and tune the first loop. the pressure loop. It should be tuned fairly aggressively because its response time determines the system response time Once tuned, put the pressure loop in auto, its behaviour is now part of the process. Flow loop #1 is tuned next, leaving flow loop #2 in manual. This loop must be tuned for a response time that is atleast 3 times slower than the Pressure loop response time, ideally 5 to 10 times stower. Once tuned, leave the pressure and flow loop #1 in auto and tune flow loop #2. It too must be tuned at least 3 times slower than flow loop #1 (ideally 5 10 10 times slower). @ When tuning interactive loops, select response times that differ by a factor of 3 to 5 (sometimes 10). PC FCI FC2 Pressure Loop | Flow Loop #1 | Flow Loop #2 Te ‘ Step 1 Manual Manual [732 fe Tune fee (for quick do response) de stot) s i a Tine (ee) | Auto Step 2 Manual | 32 are ge Be Tune as (for minimum ® 3 x 30 second & ca a a ce response) Tine Ge) ; ‘Auto ‘Auto Step3 |g= = “| ge | Be Tune | 3a | (forminimum | 22 Ll 3x90 second | FE 0 response) iain reg } ‘©TOP Control 1990 —Fundamentals of process control 6.2.2. Synchronising Loops Batch mixing is one example ofa system that needs tobe synchronised. Figure 3 shows thre ingredients that are Sent into a mix tank. ‘The proper ratio of ingredients must always be maintained ~ even when the syscin stare, speeds up or shuts down, See section 6.3 forthe tuning of the level loop, U Figure 3 Synchronising Loops {at iy manual and bumped ll thre flow logps would probably have different process gins, dead times and time Constants especially with such a large difference in valve size. As such if all three loops were tuned for 10% pe Zshoot the response times would not be the same. When the level loop calls for increased flows, the recipe will be out of balance until all three flows are stable and at setpoint To ensure that all three loops actin harmony, determine which loop is te slowest and match the response time of ‘the others to it, Normally the slowest loop is the one with the largest dead time Procedure for tu ig synchronising loops: Do a bump test on each loop L 2 Determine which is slowest ‘Tune slowest aggressively and measure speed of response when done ‘Determine tuning parameters for other loops that will give approximately the same response time ‘@ When tuning loops that need to act in harmony, select tuning parameters that give similar response times. ©TOP Control 1999 191Fundamentals of process control 6.3. Cascade Loops ‘A system isin cascade ifthe controller output of one controller becomes the setpoint for another (or others). The loop receiving the setpoint is called the inner or slave loop. The loop sending the setpoint is called the outer or ‘master loop. Innee Loop ib Inner Process B Py, Figure 4 Cascade Loop Setup Normally the outer process isthe primary concern and could be controlled by just the outer loop PID, However, adding the inner loop can greatly increase performance if the inner loop has disturbances that can be quickly eliminated by the inner loop, but not the outer. For example, the level controller in figure 5 can control the level on its own, However, its response to a disturbance in the incoming flow is poor. By adding a flow controller in cascade, the performance is increased by the ratio of the outer loop response time over the inner loop response time (normally a factor 10 or greater) 7 Tine oa) OT To Ho Tank Level Pregure co WTemetay 207 Tor Too Figure 5 Performance with and without Cascade Outer Loop Response Time Inner Loop Response Time Performance Increase ©TOP Control 1999 192Fundamentals of process control ‘As cascade loops are interacting, the inner loop needs to be 3 to 5 times faster than the outer loop (3 times faster at an absolute minimum). ‘Tuning cascade loops: Inner Loop Outer Loop Normal Operation Cascade Auto Tuning Step 1 Tune Ignore (Local - Auto ot Manual) Tuning Step 2 Cascade Tune Select response time at least 3t0 5 times slower than inner loop. Cascade systems work best when the inner loop is the temperature loop in figure 6, Fluctuations in the incomin, after a large time delay. ‘The inner loop will detect fluctuations propagate through the tank. can eliminate quick disturbances quickly. Another example of this ig water temperature will affect the tank temperature ‘much quicker as the disturbance does not need to Temp Constant Figure 6 Cascade Temperature Loop faster than the outer loop. ‘When tuning cascade loops, the inner loop needs to be at least 3 to 5 times ©TOP Control 1999 193Fundamentals of process control 6.4. Adding Feedforward Feedforward is a simple but often misunderstood concept. Feedforward is added to “help” a properly tuned PID loop. Without the “help”, the loop will act as any other PID loop, deviating from setpoint when disturbances occur. When feedforward from a specific disturbance is added, the process should not deviate from setpoint when that particular disturbance occurs. The feedback loop was defined in section 1.3 as working from the past... an error is needed to move the controller output. Feedforward is added when a disturbance that causes an error can be measured and its effect on the controller output is repeatable. By predicting the effect of the disturbance and reacting to it before it affects the Drocess, there isa huge gain in performance. For example, if | am driving my car on a cold day and my passenger opens the window, I could act lke a PID controller and wait for the temperature to drop before reacting (tuming up the heat). If react in this manner, after a few minutes the temperature will be back in the comfort zone. The better choice would be to turn up the heat as soon as the window is opened, before the temperature drops, This thinking ahead ~ similar to feedforward will ensure that the cabin temperature hardly even changes. To add feedforward: > first make sure that the loop is properly tuned > in automatic observe the effect of a disturbance on the controller output several times (make sure that there are no other disturbances or changes that occur during the test) > determine the feedforward factor” by determining the relationship between Adisturbance and ACO > test the loop performance with feedforward ” - Feedforward may be added, subtracted or in some cases multiplied Tine) Figure 7 Before adding Feedforward to a level loop For the example above, the feedforward factor is calculated to be 7/10 from the first disturbance and 3,5/5 from the second. Since these two factors are equal, we know that the disturbance effects are repeatable and feedforward will TOP Control 1999 194Fundamentals of process control seb ice bits iS inserted in the feedforward equation so the feedforward will nor ect Ihe COmaT ‘upon insertion Tein should be chosen to equal the disturbance measure at the time of insertion, This bas aden internally in Some controllers and thus it does not always need to be programmed. on ED T + Fesenwar TTT TP Ty ig) L | Lgelco+ eg Figure 8 After adding Feedforward to a level loop poet example ofa feedforward system is given below. In this example the feedforward factor is calculated to be {265 -13. Because ofthe 100-second delay, a lag block may need to Be added so thar de feedforward does not act to0 soon. 9 Oe Figure 9 Adding Feedforward to a temperature loop ©TOP Control 1999 195Fundamentals of process control 6.5. Split Range ‘When a control loop has been designed with more than one final control element, spit range control i used. A split ange controller is a standard PID controller, but its output controls two valves (or two final elements) with the same Controller output signal. The controller output is sent to both valves, however it is normally spit by calibrating the UP or valve positioner to respond only to a certain range of what it receives. Other ways to split the signal include using logic and scaling functions in a DCS or a programmable controller. A common split range application, heating with steam, cooling with water, is shown below. One standard PID controller can be used to control this application but there are some important considerations, § c on ' 1 1 ' eee Q e ' 1 1 ' It 1 ' I ' ' I 1 ' Figure 10 A Common Split Range Application ‘A commen (incorrect) approach to setting up the control for the above loop is to calibrate the /Ps such that the cooling water valve is fully open for 0% CO and closed for 49% CO. The steam valve is fully closed for 31% CO and fully open for 100% CO. This configuration has a deadband; both valves are closed over 2% of the CO span, ‘The implementers hope thatthe process will not need heating or cooling and the CO will stay at 50% during run ‘This is akin to driving with your hands off your steering wheel if your car is going straight... the car may maintain a straight line for a while, but it is safer to always keep your hands on the wheel, ready for action. ‘The correct way to set-up the U/Ps is shown in figure 1. A small overlap is induced instead of a deadband, Calibrate such that the cooling water valve is fully open for 0% CO and closed for 51% CO, The steam valve is fll closed for 49% CO and fully open for 100% CO, Ifthe same lines are used for heating and cooling, then you may be forced to have a deadband as you can't force steam and water into the same line at once. When deadband is added the loop will work, but the performance will not be as good OTOP Control 1999 196Fundamentals of process control ‘Tuning a split range controller is like tuning two loops. You have a heating process to tune and a cooling process to tune. Ifthe processes react quite similarly, then one set of tuning parameters will work. However, ifthe process dynamics (Gp, t, and t) are different when you are heating vs. cooling, then the loop must be de-tuned (tune for the ‘worst case) of two sets of tuning parameters must be used, switching from one set to the other based on CO. One orto sets of tuning | parameters? Figure 11 Split Range Set-up ‘The preferred set-up for this scenario isto use control logic to split the PID controller output into two NON- OVERLAPPING standard 4-20 mA signals to the VP's, This eliminates any special field set-up. Figure 12 Split Range Preferred Set-up ©TOP Control 1999 197Fundamentals of process control Ratio Control Ratio control is commonly used to control lows of two or more ingredients that must maintain a certain mixture. For example to control a fire, the air-fuel mix should always be maintained. Two independent flow controllers could be used as shown in figure 12; however, this is not the best approach. Ifa disturbance were to cause the fuel flow to decrease, the fuel controller will react and fuel flow will return to setpoint in perhaps 30 seconds, During these 30 seconds the air flow would have been at its setpoint, but the mixture is not maintained. The air flow is relatively too high and the flame quality would suffer. The correct implementation would be to use ratio control — the flow of one ingredient times a multiple becomes the setpoint for the other controller. Thus if the fuel flow decreases, the air flow will follow it Ingredient A Ratio Setpoint A= Flow Bx Ratio Ingredient B allowes) (eater) Correct Ratio Control Incorrect Ratio Control Figure 13 Correct and incorrect ratio control ‘When tuning a ratio system, all of the loops that are ratioed together should have the same response speed, The tuning procedure is the same as for a synchronized loop which is explained further in section 62 2. Another consideration that is often neglected witen tuning ratio loops is filtering. As with all PID loops, the PVs should be properly filtered; however, the same filter that was applied to the leader's PV should also be applied to the followers setpoint (they are the same signal), This will ensure thatthe follower’s setpoint will not jump around unnecessarily ‘TOP Control 1999. 198Fundamentals of process control 6.6. Coarse-Fine Control Coarse-fine control addresses the problem of valve rangeability. Rangeabilty is defined asthe maximum PV to be controlled over the minimum, For example a valve that must control flow between 400 and 2000 GPM would hace a rangeability of 5. When the rangeability increases the valve accuracy decreases. To get both accuracy and range, of course the valve price increases, but also in some cases there is no valve that can accurately contol over she ‘ated range. The solution is to pair up two valves in parallel A large valve to give the range and a small valve for the precise control, ‘The proper set-up ofa coarse-fine loop is to have the large valve work to try to maintain the small valve at 50% opening, Thus when a load change occurs, the small valve willbe able to make as much correction in ether direction as possible. After the disturbance the large valve will work to get the smaller back to 30% As cycling may occur due to the interaction between the two valves, a control strategy must be used that will de= couple the valves. Normally the larger valve needs to be slower than the small valve, Ways to ensure that the valves are de-coupled include: ‘+ Tuning the large valve 3-5 times slower than the small valve. + Using an “integral only” control algorithm on the large valve * Using a gap controller on the large valve. This will only allow the large valve to move when the small valve has strayed from 50% opening by a set amount ‘+ Controlling the movement of the large valve using a logic based program. large valve - provides range DK small valve - provides precision ' ' SP = 50% Figure 14 Coarse-Fine Control ©TOP Control 1999 199Fundamentals of process control 6.7. Advanced Control ‘The following section is an introduction to modified PID and non-PID control algorithms. Most loops perform adequately with PID control alone. When this isn’t enough performance can be improved by characterising the controller output, adding feedforward, or by putting loops in cascade, ratio or by using fine-coarse control. When these strategies are properly implemented and still do not give the required performance the system must be redesigned or more controller “horsepower” is needed. The rest ofthis chapter will introduce the base concepts of ‘the most popular advanced controllers. For a more detailed discussion, please reference the texts noted in the bibliography 6.7.1. Special Algorithms — P’ID, PI?D, and others A standard PID controller is normally implemented using one of the three most common structures ~ Ideal, Parallel, or Series. Although the structures differ (section 3.6.2 discusses controller structure), the behaviour of all three is similar, the input to the controller is always the error (PV-SP or SP-PV) and the controller output is composed of a proportional, integral and derivative contribution, CONTROLLER OUTPUT = ERROR X (PROPORTIONAL CONTRIBUTION + INTEGRAL CONTRIBUTION + DERIVATIVE CONTRIBUTION) ‘This standard algorithm is a universal solution. It works on all processes, but it does not always provide the desired Performance. To increase performance, especially on non-linear processes, the standard algorithm can be modified Special algorithms change a standard PID controller and make it non-linear when some facet of control needs improving, A non-linear controller is usually used to compensate for process non-linearities. Fora linear process, standard PID Control is a beter choice than a modified algorithm, however, when the situation warrants a modified algorithm and itis implemented properly, the result is improved performance. Some controller manufacturers make these algorithms available as a choice in the controller setup. If they are not available, you may be able to program their ‘behaviour into a programmable controler. TOP Control 1999 200Fundamentals of process control ‘Algorithm Description Standard The implementation shown here is using an ideal controller structure (the recommended structure) Don PV A seipoint change will be handled by the proportional and integral components only ‘The derivative is only active when the PV moves. This algorithm is recommended any time derivative is used and the algorithm can be implemented. [=] J) Dand P on PV A setpoint change will be handled by the integral component only. As a result there should be no overshoot on a setpoint change. ‘The proportional and derivative are only active when the PV moves. This algorithm allows for ‘more aggressive tuning and is always Tecommended any time it can be implemented, f eI Caution — When tuning a loop that uses this algorithm the setpoint change response is not indicative of the controller parameters. As such the ultimate cycling and tial and error methods can not be used for tuning Setpoint Filtering A setpoint change will not provoke as aggressive a response, Overshoot on a setpoint change will be less and the proportional gain can be increased as a result. This algorithm allows for more aggressive tuning and is recommended on loops that have frequent setpoint changes and disturbances, fre Caution — When tuning a loop that uses this algorithm, the setpoint change response is not indicative of the ES] mg controller parameters. P'ID An error squared controller will act quite (error squared) aggressively when the error is large and quite sluggishly when the error is less than 1%. At 10% error, the input to the PID is 100%. At 0.5% error, the input to the PID is 0.25%. This algorithm can give aggressiveness and steady state stability when property implemented. The ‘most common uses of this algorithm are for level control (averaging and surge) and pH. Care must be taken when using this algorithm, if the tuning is too aggressive, sust [m= itis more common to modify the algorithm by limiting the result of error squared example) or by using a less ageressive function. see Ly tained oscillations will result, In fact limit it to 2 times the real error for PED A less commonly used algorithm is the PPD, (integral squared) Since integral action is normally linked with ddeadtime, it makes sense that integral squared ‘action works best fo compensate for processes with a variable deadtime. Integral action causes overshoot in level loops — with integral squared as the level approaches setpoint, the integral contribution goes to zero. as ©TOP Control 1999 201Fundamentals of process control 6.7.2. Smith Predictor ‘The Smith predictor is used primarily on processes with long deadtime, It was developed, as one might have guessed, by Mr. Smith in the 1950's, ‘Mr. Smith realised that deadtime was the problem in process control. His solution was to model the process and use this model to get process feedback without the deadtime. The effects of any movement on the controller output were seen much quicker and controller gain could be much larger than that of a standard PID, To account for any errors in the model, the difference between the model (including deadtime) and the real PV was taken. This correction was called the modelling error. On a setpoint change this controller can work quite wel; the result i a fast response with no overshoot. However, When a disturbance occurs it does not affect the model, only the real process. As a result the controller is no better than a standard PID during load changes. Note — Since a model with no deadtime is being controlled, there is nothing to anticipate and derivative can not be used. Output P| Controller Process. |------ v Setpoin 1f the Smith predictor is not included in your controller, it can be implemented through programming using the following pseudo-code: Process Variable = Modelling Error + Model Feedback Model Feedback = First order filter of (CO x G,) First order time constant = Process time constant Modelling Error = True process feedback - Deactime look-up table of (model feedback) Deadtime used in lookup table = process deadtime ‘TOP Control 1999 202Ieee EEE Eee Eee eee eee eee EE Ee Dee Ee Ee ee ee eee Cee eee eee eee ee See ae ee ee ae a a a a Fundamentals of process control 6.7.3. Internal Model Controller (Model Predictive) ‘The internal mode! control concept bases controller solutions on the process model, The solutions can be implemented as a unique controller algorithm but are most commonly transformed into numbers that canbe put ina standard PID controller. The Dabiin algorithm isthe most common internal modelling method used in industry. Although it is a subset of internal modelling, the Dahlin controller and the internal model controller are sometimes used synonymously. In the 60's, Dahlin, was trying to apply a simple mathematical relationship to a more complex problem. Algebra tells us that the product of a number multiplied by one over itself gives one as the answer. Dahlin reasoned that if ‘we could tune a controller to react as one over the process, the result would be one (or the setpoint if we would like) 1 u SP Xx—T=1l 2. SPx Process x ——_ x Process ‘The result on a setpoint change would be the same as for a CO bump - a first order response, If we have the controller specified perfectly, we don't even need feedback to see if the process is under control ‘The mathematical model for a first order process is given as: G, xexp(-jar,) 1+ jar Since j equals the square root of negative one, the solution to one over this model is impossible to solve and program into a controller. What Dahlin found was that the solution is easier to find if the deadtime is ignored. When this is the case the solution becomes: Process Controller=> —P -3 -, 1=T,=r(sec) 'p If this is familiar, itis because itis very similar tothe formulas used for Lambda tuning. (discussed in section 4.10). Lambda tuning builds off of Dablins results, using 2, lambda, asthe closed loop time constant and a multiplier of the above formulas. Using these numbers ina PID controller wll give a fist order type response to a setpoint change, but the only way that the controller will know when a disturbance occurs is by way of the feedback loop. The response on a load ‘change or disturbance is much slower than other traditional tuning parameters. Controller « ey oS oy ia point 1 > 2 =P Pics: eet Process Process. ‘T | variable Variable 1 1 ' Figure 16 Internal model controller ©TOP Control 1999 203)Fundamentals of process control 6.7.4, Fuzzy Logic Fuzzy logic isa very interesting topic as its approach solving problems differs from conventional methods. Whether fuzzy logic is used for process control orto fly a helicopter, we do not need to know anything about the ‘mathematical dynamics of the element being controlled. For example, when a baseball player chases a baseball, he or she does not need to know Newton's laws of motion to catch it The fuzzy logic concept was brought forward by Lofti Zadeh in 1965. As the head of the Electrical Engineering, department at Berkley and author of several articles and textbooks, he was well versed in traditional controls problems, He realized that as the degree of complexity of a system increases, the solution becomes tougher for us to ‘work out “mathematically”. He proposed a solution that would skip all of the number crunching and implement solution based on “fuzzy rules”. Since the concept of fuzzy logic was stil a little fuzzy at inception (partion the pun), the reaction was less than enthusiastic. However, less than 10 years later, in the U.K.. a fuzzy based controller had been implemented to control a laboratory steam engine A good example of a mathematically complex system is the equations that govern the balancing ofthe “inverted pendulum” (picture a stick balanced on your finger). Whether it be in two or three axes of motion, the solutions to the equations that govern the motion are the subject of many mathematical doctorate theses; yet a quick tour of the school yard will find many children that can control “the process”. Fuzzy logic proposed a new set of logic that ‘would yield the control solution without having to solve the mathematical model The logic part of fuzzy logic is similar to human reasoning For example if we classify people by height, traditional logic would have finite states —if a person is 167 to 178 cm tall they are “average”; 178 cm and above is “tall”. This, ‘would lead to classifying someone who is 177 cm tall as being “average”. Fuzzy logic would classify this person as being 40% tall and 60% average. Instead of black and white or “I” and “O” we have shades of grey. With very litle information, tnzmans can make good judgements ~ the baseball is kind of far away, but it’s speed is pretty fast and it looks like itis dropping alittle ~ this means I should run fast towards the infield to catch it, Similar a fuzzy logic ‘controller takes one or several measurements of what a process is doing and applies a set of rules to determine the best response to maintain setpoint. traditional logic merbeship fuzzy logic =001 Tall = 010 4 Sto Average q Tall 167 178 Height (em) Height (om) 177 em ~ Average I77em= 60% Average and 40% Tall, ’ ‘ Figure 17 Fuzzy Logic vs. Traditional Logic Fuzzy logic is easily understood as the variables are classified linguistically (increasing, decreasing, flat, hot, very hot, luke-warm, etc.). If you were to ask the operator of a heating furnace what he would do if the furnace temperature was very hot and increasing, his answer might be “I would decrease the setpoint by 20 degrees until the temperature dropped”. This operator's knowledge could be transferred to a look-up table, defining the rules for the fuzzy controller. OTOP Control 1999 204Fundamentals of process control Furnace Temp.Trend Dec Flat Inc. Possible reactions: ZE = Zero change to setpoint (0%) Hot | ze | sor} uw SP = SmallPosiive changeto setpoint (194) ead MP= Matin Positive change to setpoint (5%) ‘urna LP = Large Positive change to setpoint (15%) sp | zx Temp. Norm.| s® | 28 | sw SN = Saal Negavecbagewoseom (lag MN = Medium Negative change to setpoint (39) Cold | 1? | mp | ze IN = Large Negative change to setpoint 15%) Example ofa rule - If the fumace temperature is cold and is not changing (trend is fat), then give the temperature setpoint a Medium Positive increase Figure 18 Fuzzy Rules Remember that fuzzy logic would not define the temperature as being “Hot” or “Normal” or “Cold; they would be Isonly developed once and then > Users often prefer to have more is mass produced control > Saves money by reducing the ‘umber of dials and switches > Does a better job than most users In transport > Performance of the perfect > Extensive backup systems are operator can be programmed required > Computers never sleep or make > Extensive testing is required errors > Computer based programs can have bugs ‘As a PID replacement > Control will be much better. > Setup cost is relatively high ‘than PID (response is quicker (uch more than PID) and more stable than PID) ‘Asanon-linear controller > By makingthe shape of the > Knowledge of how to control input and output “triangles” the process in all situations is ‘on-symmetrical the controller required. ‘becomes non-linear > Setup cost is relatively high > Can control processes that PID (much more than PID) alone can not, Asa multivariable controller > The fuzzy logic controller can > Setup cost is relatively high control multiple Toops to cut (much more than PID) cost or maximize production. > Can maintain tight control of loops that would normally interact with each other. ‘TOP Control 1999 206 DAHRARHRARKRAAAKRA RARFundamentals of process control 6.7.5. Neural Networks Neural networks with a single hidden layer and trained by least-squares are statistically consistent estimators of arbitrary square-integrable regression functions under certain practically-satisflable assumptions regarding sampling, target noise, number of hidden units, size of weights, and form of hidden-unit activation function. (White, 1990) With most introductory texts on neural networks beginning with statements such asthe one above, itis no wonder why there is much confusion about neural networks, Fal neivorts solve problems tht most traditional computer systems can not. Speech recognition, data sorting in {he presence of noise. mathematical approximations of complex calculations and many other intricate tasks cay te solved using neural network solutions. Figure 20 Neuron connections in humans ‘The computer version of a neuron system links nodes (called neurons) together inthe same manner, ‘The input to cach node is one or more pieces of data, normally measurements or outpus of other neurons. The output of 2 neuron is a function of the sum of each input times a changeable weight, or gain, Input 1 —of Gain Input 2 asia i — Output or to eee another node Input 8 Input 4 — Figure 21 Neural Network ©TOP Control 1999 207Fundamentals of process control The equations that govern the behaviour of each node are as follows: Sum of n inputs = ¥° Input, xGain, Node Output = F (Sum of n inputs) 1 (Sum) = aie = ‘When a large network is connected, it consists of several inputs and one or more outputs plus many gains. The gains ‘must be adjusted so thatthe nodes behave as a system, giving the desired outputs. Instead of going through and ‘tweaking each gain, which would take days to weeks, the network adjusts the gains on its own, using training data Large amounts of training data are fed into the network. The network adjusts its own gains after each data pass ‘based on how far off ofthe known solution its answer was. The reason that this is possible is due to the function used to compute the output of each node. The function used is normally a sigmoid function such as 1/(-+e""") ‘This non-linear function allows the non-linear relationship between the inputs and outputs to form. As back Propagation of the solution is done using the derivative of the sigmoid function, it also allows the network to learn. ‘When a human leams to ride a bicycle, the first few attempts result in spectacular crashes, but eventually the body ‘movement is refined until a semblance of control is attained. Similarly, the neural network needs a lot of time (and data) before it comes up with the correct solutions, ‘Once trained, a neural network will find the answer in the data if the right data is given. However, the network can not create information that is not contained in the training data. Neural networks show a lot of promise in process ‘control, not as individual loop controllers, but as orchestrators of the entire process, speeding up and slowing down production rates and modifying setpoints. The problem with using neural networks for process control is that they take a lot of time and money to put in place, and when the system changes the controller will need retraining, In other fields neural networks are providing never imagined solutions as well as frustration. There are many ‘mathematical problems that known algorithms can solve yet neural networks can not (or they can only solve them afier extensive training), Neural networks are similar to fuzzy logic in that the mathematical model relating inputs to outputs does not need to be known, the behaviour response is known instead. One major difference between fuzzy logic and neural networks is that we can not modify the gains or functions in the neural network, we can train the system with data but can not. train it with human reasoning. Fuzzy logic allows for modification of each controller parameter. ‘TOP Control 1999 208Fundamentals of process control 6.7.6. Multivariable Controllers Ina multivariable control system, the control algorithm is designed for the specific system. For example, a mode! is ‘computed that relates each variable, This is similar to adding feedfoward between all loops and potential disturbances in a system. The controller, in reality, is an equation set that allows each component to know the impacts of each move it makes in advance, Hence, the controller will send a new CO to each valve (or a new SP to cach loop) each time it makes a move, When properly setup this permits the entire system to maintain stability when applying a correction. To build such a controller, alot of models, or equations, need to be defined. For example, a3 Input /3 Output system, 6 models are needed; with N variables, N*(N-1) models are needed; with 10 variables, you need 90 models. If the number of variables is high, a very complex controller is needed as well as a lot of data to obtain the models, Loop A (Zone | Temp) lk Loop B (Zone 2 Temp) B ed Loop eC (Air Flow) Each product is an equation that represents the effect of one loop on another _ Figure 23 3x3 multi-variable system Due to the number of relationships that need to be determined, the multivariable controller is relatively expensive to setup. Once it is up and running it will do a good job until something changes. For example ifthe process is re- designed or equipment starts to wear (hysteresis, stiction, noise, etc.) the controller performance will decrease or become unstable. In many cases a multivariable controller controls entire production rates and loop setpoints rather than individual loops. If any of the individual PID loops are re-tuned, system performance will change. 6.7.7. Comparison of advanced controllers See summary in figure 23. ©TOP Control 1999 209Sl aiicecdeececend td) sdcesseebaeeeeseed Controller Brief Description / Comparison Pros and Cons Various Modified D on PV - Will not improve performance, just solves a potential undesired effect. © Always helps (except on the inner Algorithms loop of a cascade system) D and P on PV — Gives the ability to have a quicker response time without overshoot on © Improves control setpoint changes. ® Can cause confusion when tuning Setpoint filtering — Gives the ability to have a quicker response time without overshoot on © Improves control setpoint changes. i ___ © Can cause confusion when tunin PID / PED P'ID — Controller will aggressively for large errors, slugishly for small errors ® Does not improve speed that greatly Can improve speed by a factor of 2 on some loops. for the time it takes to impliment and test PED — Controller integral mode will act only on large errors, eliminating cycling caused by integral on some processes, © Causes problems when not required Can improve steady state stability in some loops. Does not affect speed too much, (seldom required) Smith Predictor Controller gets feedback from a model of the process without the deadtime rather than the real © _No overshoot on setpoint changes. process. Real process variable is compared with models to correct errors and handle © As fast or only slightly faster than disturbances. Good for processes with large deadtime that have setpoint changes. normal PID control on load changes Dahlin Controller / Controller tries to behave as one over the process, giving a closed loop response that hasthe _- @ No overshoot on setpoint changes Model Predictive same form as the open loop response. This is similar to lambda tuning. ® Very sluggish on load changes Conti Beneficial for loops that can tolerate no overshoot on a setpoint change. . a: Fuzzy Logic Programs the behaviour (or decision making) of the perfect operator into the controller © Improves performance on individual Performance is fairly robust inthe face of changes to the process or improper definition of one or loops or for the whole system more rules. ® Not well know or accepted in North Is less difficult to implement and test compared to neural networks or multivariable systems. ‘America (yet) Performance on par with neural networks and multivariable, but more robust. oa Neural Networks [Network of nodes that are similar to human neurons in that they can be taught to respond ina @ Time consuming and must have alot certain way, Network is rained on data and learns the correct responses without having to know of traning data the complex mathematical solution. © Improves system performance (not Is more difficult to implement and test than fuzzy logic, but easier than multivariable controllers (for large systems). Performance on par with neural networks and multivariable. System can be re-taught, but not tuned (black box). normally applicable to individual loops) ‘TOP Control 1999 ATA 210SEE ee ee ee eet eel el oan ae | ae Fundamentals of process control Multivariable Controller links al of the control loops and disturbances together. Compensates where required @ Time intensive to setup and test Controllers ‘o maintain stability in all linked loops when manipulating a variable. ® Needs to be re-tuned anytime process 1s most difficult to implement and test compared to fuzzy logic and neural network systems changes UNLESS system is small (33 to 5x5), © Works very well when in place and Performance on par with neural networks and multivariable. system is constant System must be re-tuned when process changes. Adaptive Model ‘Same as mode! predictive controllers but the model is continuously updated. Very good in © Very good in mechanical systems Controller mechanical systems but does not handle process problems very well. Hysteresis, stiction, non- Does nat handle process problems linearities cause the system to “re-model” continuously. very well Model Free Controller At time of publishing, there have been no model free controllers that truly work without any + Tobe determined prior knowledge (or approximation) of the process model, Figure 24 Comparison of advanced control methods TOP Control 1999 2Fundamentals of process control 6.8. Definitions and Formulas Interacting Loops PID loops that impact each other, For example a setpoint change in one loops causes a disturbance for the other. Synchronising Loops PID loops that need to act at the same speed. Cascade Systems Feedforward Split Range Ratio Control ‘The controller output of one loop becomes the setpoint for another loop (or loops). Compensation for a disturbance. Normally used when the effect of the disturbance can be calculated and is repeatable. A split range controller controls two or more final control elements. ‘The final control elements are only active over a percentage of the controller outputs total range, A control set-up where the setpoint ofa PID loop is a ratio of another PID loops measured variable (PV), Coarse-Fine Control Using two PID loops with two final control elements to give range and precise PID Error Squared prp Smith Predictor control PID algorithm that uses the square of the error instead of the error as the input. PID algorithm that uses the square of the error instead of the error as the input PID algorithm that squares the input to the integral calculation. Control algorithm that takes feedback from a model of the process without the deadtime, Model Predictive Controller Any type of control algorithm that basses controller parameters based on the inverse of the process parameters, Internal Model Controller Any type of control algorithm that basses controller parameters based on the inverse of the process parameters. Fuzzy Logic Controller A controller that uses fuzzy logic (rather than traditional finite state logic) to determine the appropriate control response. Neural Networks Controller A controller who's internal structure is linked together in the Multivas Rule of thumb when Rule of thumb when tuning synchronising loops: same way that human neurons are linked. This structure can “lear” a desired response. ble Controller A controller that attempts to de-couple all loops by computing the effect that each loop has on the others, tuning interactive loops: When tuning interactive loops, select response times that differ by a factor of 3 to 5 ‘When tuning loops that need to act in harmony, select tuning parameters that give similar response times. Rule of thumb when Rule of thumb when tuning ratio systems: tuning cascade loops: When tuning cascade loops, select response | times that differ by a factor of 3 to 5. [parameters that give similar response times. When tuning loops are set-up in ratio, select tuning ‘TOP Control 1999 212Fundamentals of process control 6.9. Exercises |-When tuning the first loop in a series of interacting loops, all others should be in: Auto Manual 2-When tuning the outer loop of a cascade system, the inner loop should be in: Cascade Local 3-When tuning a loop with feedforward, the feedforward should be: Active Ina 4-Choose the best answer(s). When tuning a split range controller, tune as you would any loop ‘uy and find parameters that will work in both ranges of operation ‘use multiple sets of tuning parameters if required ensure that the control zones do not overlap SEE EEE Ee eee eee Caer eee eee eee ee eee eee ee a ©TOP Control 1999 213Fundamentals of process control 6.10. Computer Exercises 1- Using the standard simulated flow loop, find tuning parameters that will give a 180 second response to a setpoint change or disturbance. Test the parameters found. If the result is quicker than 160 seconds or slower than 200 seconds, refine the parameters and try again. P found - I found - D found - F found - Response Time Found - e e e e ¢ e é e @ € e é ¢ 6 6 6 6 @ g ‘GTOP Control 1999 214see OO CC CeCTCeCeEESSE Fundamentals of process control 6.11. Notes ©TOP Control 1999 215Fundamentals of process control 7. PID controller in PLCs ©GTOP Control 1999 218Fundamentals of process control 7.1. Introduction Although the PLC was relative latecomer tothe field of industrial controls, they have been used to implement PID Control since the early to mid 1980's. Newer versions allow for almost all of the advanced functions offered in Stand-alone or DCS controllers, This being said, using PLCs versus “conventional” PID solution is sila hot topic of debate in some instrumentation shops and boardrooms. In Top Control's opinion, almost all ofthe well-known PL.C’s can be configured to implement good PID control. However, they all require a decent programmer with a good knowledge PID to implement the PID correctly. This is not to say that DCSs and stand-alone controllers are easy to set-up. They can be tricky also, but they are usually pre-configured or installed by experienced technicians. This chapter wil: ‘+ List some of the known pitfalls for setting up PID control using a PLC * “Translate” common terminology used by PLC manufacturers into the terms used throughout this book + Explain how scan time can be important to the PID calculation Timer#1 | | a = So 300 msce a _ = see = recuaf J] : Timer 1 tt done bit Figure 1- Typical PID logic and setup screen Usually, a list of parameters is displayed and the programmer selects the appropriate values, One must be careful since sometimes, the default values are unacceptable ©TOP Control 1999 219Fundamentals of process control 7.2. The importance of sampling time on the PID calculation 7.2.1. Analog and didital jehapter 3 of iis book, the PID controllers P,Tand D modes were explained by analyzing them separately. The ‘input to the P, 1and D blocks were continuous analog signals as were the outpute ‘Analog signals ~ Continuous Digitally sampled signals — Discrete ‘Theres a separate Value associated with every estan of ime ‘Theres valu sample ake every“ seconds, Value] — Value} xXXX ae x* x a KKK Time Time Figure 2- Analog versus digital signals Fneumatic and analog electronic controllers work off of analog signals, however, most modern controllers sample {he contol signal to allow them to be processed by a microprocessor. With stand-alone controllers the sample rate is almost always quick enough that we can consider it to be “true to life” ‘With PID control, ifthe PV and SP are not sampled often enough, two problems result: J: The effective process dead time increases, making process more instable and harder to control 2- When the sampling time gets too large, aliasing can result, making the contollr fight a cycling that is not resent 70.26 what is going on inthe loop, the controller needs to sample faster than the loop can move; hence the Saunpling time shouldbe faster than the damped period of the loop or faster than the closed loop tine conan One pak fe accomplish this and to minimize sampling error is to sample the loop ata rate of 1/10th to Usth erika Drocess dead time (1/10th or faster being preferred), Tins 2200p is sampled at any given rate, the effective process dead time is increased anywhere from ts tots. nus sample rate of 1/10" ofthe dead time increases the dead time by 5-10%. A sample ote of Use ofa dead perbrmmmaats Ie dead time by 10-20%, An increase i dead time wil always result ina detease in loop performance. Thus, the faster the sample rate, the better. Having a constant sampling time is also important for good control, This is because the Integral and Derivative ‘modes of the PID controller are time-based functions. ‘The Integral block calculates its output based on how long the error has been there. * The Derivative block calculates its output based on how fast the error is changing _hihen we sample a signal and then present it tothe PID algorithm for computation, it needs to know exactly how tong it has been since it saw the previous sample. When the PV and SP are not sampled at acres interval, a miscalculated CO value can result. ©TOP Control 1999 220