Proceedings of 2019 IEEE

International Conference on Mechatronics and Automation

August 4 - 7, Tianjin, China

Real-Time Flow Control System

Based on Siemens PLC
Zeyu Quan, Xin Yuan, Yuntao Zhu and Zhengyang Wang
College of Automation
Harbin Engineering University
Harbin, Heilongjiang Province, 150001, China
[email protected]

Abstract – Traditional flow control systems are hard to reach For actual flow in industry, we should take more
the goal of high-precision and high-speed due to the volatility, complicated factors into account, which can't be realized and
nonlinearity and large disturbance of the flow. To solve the solved easily only by simulation. In the simulation, the data
problem, a real-time flow control system based on fuzzy PID was calculated by the algorithm can immediately play a role. But in
designed in the paper, which used ID3 algorithm of the decision
reality, data must be sent to a regulating valve or a regulating
tree to calculate information gain and select the best decision to
formulate fuzzy control rules. The design selects Siemens PLC as pump through an electrical signal to take effect ˈ which
control core, combining with electromagnetic Flow Meter, changes the flow by means of physical adjustment in most
Regulating Valve and Human Machine Interface. Finally, results cases. From the time of signal arriving to the completion time
of simulation and experiments of this control method show that of adjustment, it often takes a long time and brings hysteresis,
the control system has favorable dynamic and static which is difficult to embody in simulation [4-7].
characteristics. The design, based on the predecessors', proposes a
comprehensive scheme that can link the complex situation in
Index Terms – Flow control; Fuzzy-PID; PLC; Machine reality with fuzzy PID control, and further explore the
Learning; Decision Tree complex relationship between the 147 rulers of fuzzy PID and
the real situation in production with the help of decision tree
I. INTRODUCTION algorithm [8-10]. In terms of equipment, the design uses a
combination of PLC and HMI. Compared with the AFPIDC
Multi-channel liquid mixing devices has been widely
flow controller composed of PC and DAC and designed by
applied in today's society, especially in liquid dilution in
Zulfadhli Mazlan, the design has a smaller volume and lower
proportion. However, due to the complexity of flow system,
cost, and is better able to handle the task full of high mobility
which has the disadvantages of non-linearity, great inertia,
and computational complexity [11].
hysteresis, time-varying, and easiness to be interfered, when
The following of the paper is divided into two parts. The
the flow needs to change, it is often difficult to control the
first part is the introduction of the fuzzy PID controller based
accuracy, the stability and the response speed of flow. In order
on machine learning. The second tests the real control effect of
to make the flow system appropriate for the complex working
the method through simulation and specific experiments.
environment, especially in the pharmaceutical, refining,
chemical industries, and to keep the flow value remaining II. FLOW CONTROLLER BASED ON FUZZY-PID ALGORITHM
unchanged or changing according to certain rules to ensure the
normal production, it is particularly important to find a method A. Fuzzy Design in the Traditional Way
to control the parameters of flow precisely. Because of the lack of the self-adjusting ability of the flow-
The design analyzes and summarizes the experience and change and the adaptive ability of the external interference, it
shortcomings of predecessors' works in detail. In terms of is difficult for traditional PID to timely adjust the parameters
algorithm, the flow control method developed by Qi Xing Liu of the flow system. Compared with the traditional one, the
with expert PID, which sets the PID parameters by expert improved fuzzy PID control can make up for some defects,
experience, has certain limitations for the fluctuating flow which can use inaccurate fuzzy information to decide the
system [1]. The fuzzy PID method developed by Song Lepeng parameters of the PID. Therefore, the method only needs to
and the adaptive fuzzy PID developed by Zhifei Yan solved care about the actual change of flow, not complex
the problem of the detection and control of small flow [2-3]. mathematical models.
However, little attention has been devoted to the situation The fuzzy PID controller, by applying the fuzzy set theory,
in reality. The fuzzy rules they used are based on long-term can automatically adjust the fuzzy relationship of Kp, Ki and
experience of other experts, which may not adapt well to every Kd from PID according to the control deviation E and the
system in industry. Moreover, the most important point is that deviation change rate Ec. The fuzzy controller in the
they only carried out MATLAB simulation to verify their own experiments takes E and Ec of flow as input, and Kp, Ki,
theories, instead of setting up actual waterways for Kd as output, which is shown in Fig. 1.
experimental testing.

978-1-7281-1699-0/19/$31.00 ©2019 IEEE 1703

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on March 15,2021 at 02:19:48 UTC from IEEE Xplore. Restrictions apply.
 construction process does not require users to master too much
knowledge about the flow domain, and only needs users to
label the flow data to carry out machine learning.

Fig.1 Fuzzy-PID controller

The data signal of flowmeter are sampled uniformly by the

flow control system and compared with the target flow value, Fig.3 Simple Decision Tree of Ki due to E and Ec
which has been set by the system after calculating, to obtain
the flow deviation value E and the deviation change rate Ec, A decision tree is a tree structure, in which each internal
which are used to determine the three adjustments of the PID node represents a test on an attribute, each branch represents a
parameters. The basic domain of E on the fuzzy set is [-100, test output, and each leaf node represents a category. The
100], the one of Ec is [-16, 16], and the quantization factors of traditional fuzzy PID is a decision tree, of which Fig. 3 is a
Ke and Kec are 0.03 and 0.1875. simplified form which determines the value of Ki according to
In the Fuzzy control system, the fuzzy sets such as E, Ec, the difference of E and Ec. The attributes E and Ec of the flow
Kp, Ki and Kd are divided into 7 grades, including {NB, represent selection attributes, based on which the tree structure
NM, NS, ZO, PS, PM, PB}, which stand for change speed of will generate 49 branches in total, because there are 7 kinds of
input variables. The domain of Kp is {-3, -2, -1, 0, 1, 2, 3}, linguistic variables of E, Ec. Results of Kp, Ki and Kd
Ki is {-0.6, -0.4, -0.2, 0, 0.2, 0.4, 0.6} and Kd is {-3, -2, -1, represent 3 different leaf nodes. Therefore, the traditional
0, 1, 2, 3}, of which the membership functions of Kp, Ki, fuzzy PID theory has 147 different decisions of 49×3.
Kd, E and Ec are selected as shown in Fig. 2. Selection attributes of the design in decision-making are
more than the ones of traditional fuzzy PID, because the most
important factor for the flow system, except the flow size, is
the flow change rate and time effect. The data are uniformly
sampled and numbered according to time in each experiment,
of which the flow rate E, the flow change rate Ec, the flow
change acceleration Ecc, and the electric signal of the
(a) regulating valve (Signal), the opening of the valve (Opening),
the pump power (Pump), the pipe inner diameter (Diameter)
and so on, are collected and combined into Prior Data, where
Kp, Ki, Kd are stored as leaf nodes, as shown in TABLE 1.
The more attributes we take into account, the deeper the
decision tree is structured, the more complex the decision rules
(b) will become, and the more appropriate the model will be to the
Fig.2 (a) Membership functions of E and Ec existing flow system.
(b) Membership functions of Kp, Ki, Kd
TABLE 1
B. Fuzzy Control Rulers with ID3 Algorithm FUZZY CONTROL RULE FROM DECISION TREE
Since the flow system is complex and changeable, if only
relying on the experience of existing flow system designs, the
previous experience may not adapt to the system very well.
Once the flow environment changes, for instance, the pipe
diameter changes or the pipeline lengthens, the existing
experimental experience will be immediately invalid. Instead
of summarizing the experience with each new experiment, it is
better to explore a method that can adapt to the environment As shown in TABLE 1, there are 5 sets of Prior Data. Each
faster according to the actual situation of flow. set of different attributes will have different effects on Kp, Ki,
Therefore, the design, based on the predecessors', puts Kd. The flow data of the pipeline interior is collected every
forward a new machine learning method to replace the second, whose data at each moment are different from the
traditional model of manual operation in setting rules of the previous moment, after which machine learning will start
fuzzy PID, in which the ID3 algorithm of the decision tree based on these huge prior data.
(ID3) is used to select attributes and judge the fuzzy Since there are comparisons of the acceleration of flow-
relationship from the prior data obtained from thousands of changes, the opening of the regulating valve and so on, it can
teaching experiments. For this kind of knowledge discovery, display the adjustment force of the electric signal to the
the advantage of the decision tree ID3 algorithm is that the opening of the valve and the hysteresis effect of the valve from

1704

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on March 15,2021 at 02:19:48 UTC from IEEE Xplore. Restrictions apply.
adjustment. By integrating multiple factors in reality to find experiments are different, and each has its own special
out the deeper relationship, we can avoid the situation that the characteristics. When the flow environment has changed, the
theory does not match the actual environment, when relying best method is to carry out machine learning by ID3 again.
only on E and Ec to make decisions in the simulation. In this Although taking a certain amount of time, it can produce more
way, different control rules will be generated for different flow suitable effects for the new flow environment.
environments, which greatly increase flexibility and relatively In short, the method is not based on purely theoretical
reduce the operators' huge workload in selecting parameters. research, but rather focuses on actual work.
Of course, these data of results are obviously not suitable
C. Defuzzification
enough to be transformed into the implementation of a
Fuzzy inference is a kind of uncertainty inference
program. So a quantitative investigation is needed to evaluate
method ˈwhose foundation is fuzzy logic. There are many
the performance of each branch of the decision tree. The
design uses the ID3 algorithm, which adopts the information kinds of inference method. In the design, fuzzy ratiocination
entropy gain as not only the method to select the optimal uses the centre-of-gravity method of the minimum and
splitting property but also the standard to measure the purity of maximum, namely the Mamdani method. For the established
the node, to choose the best decision of Kp, Ki and Kd. The fuzzy variables and fuzzy control rules, the control variables
formula of the information entropy gain is as follows: are determined by the fuzzy inference. Because the fuzzy rule
is the form of “Ai 䴒 Bi => Ci”, the final comprehensive
Info _ Gain = Entropy −  pi ⋅ Entropyi (1) ratiocinating result is as follows:

In the formula, Entropy represents the entropy of the parent μC ( z ) = μC1 ( z ) ∨ μC 2 ( z ) ∨ …… ∨ μCn ( z ) (2)
node, Entropy i represents the entropy of the node I, pi
In the formula, ' 䴒 ' indicates an operation of logical OR.
represents the ratio of the data volume of the child node i to
An accurate analog or digital system cannot be controlled
the data volume of the parent node. The larger the entropy is,
precisely by the result of inference, so defuzzification must be
the more information the node will have, and the less pure the
converted to an accurate output, of which the center of gravity
result will be. The larger the Info_Gain is, the smaller the
of C from the fuzzy sets can be calculated as:
entropy is after splitting, and the purer the child nodes and the
better the classification effect will be. Therefore, the largest n

attribute of Info_Gain is selected as the split attribute.


i =1
μ C ( zi ) zi
z0 = n
According to the ID3 algorithm, PLC can automatically  μ C ( zi )
calculate the information entropy gain of each attribute i =1 (3)
separately, and then selects the root node with the largest gain
In summary, after obtaining the various adjustment
to split. If the information entropy has the most rapid decline
parameters of the PID, the fuzzy PID adjustment can be
after data partitioning through the selected attribute, then this
realized, of which the calculation formula is as follows:
attribute is the optimal choice.
Since ID3 algorithms can be implemented in both Kp = Kp 0 + ΔKp
MATLAB and Python, of which the Python implementation is 
 Ki = Ki0 + ΔKi
much cleaner. So the Python's Sklearn of tree library is put  Kd = Kd + ΔKd
 0
(4)
into use in the design to realize functions. After thousands of
learning processes, the fuzzy control rules of three parameters Kp0, Ki0, Kd0 in the formula are initial values, and Kp,
of PID had been finally formulated. The fuzzy rules of Kp Ki, Kd are PID parameter values obtained after the
are shown as TABLE 2. adjustment of fuzzy inference. In the experiment, only the
TABLE 2 deviation values E and Ec of flow are needed to be input into
FUZZY CONTROL RULE OF KP the fuzzy controller. After fuzzy processing has been input into
the controller, the appropriate value is entered into the fuzzy
rule base, after which the output of the fuzzy controller can be
obtained. After all of these, the appropriate parameters of the
PID controller can be acquired by the adjustment of
appropriate scale factor.
III. SIMULATION AND EXPERIMENTS OF REAL-TIME FLOW
CONTROL SYSTEM

However, because these are rules generated by means of A. Simulation

machine learning, the table has superior adaptability to the After determining the exact solution of the flow
flow environment built in the design. From the analysis above, experiment, the design simulates the traditional PID, the
the control rules derived from the actual situation must have general fuzzy PID and the one based on machine learning
certain particularities. After all, the water paths of disparate separately in the Simulink of MATLAB, to test its feasibility

1705

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on March 15,2021 at 02:19:48 UTC from IEEE Xplore. Restrictions apply.
and theoretical control effect. Firstly, run the module of Fuzzy control effect. In order to better verify that it also has the same
Logic Controller in the command window, enter the fuzzy effect in actual industrial control, it will be tested with PLC in
logic editor, and select the 'Mamdani' method from the detail in actual waterway experiments.
controller type. Last but not least, input the membership Open the Window of Fuzzy controller surface observation,
functions and quantization intervals of E, Ec, Kp, Ki and Kd in which we can see the output surface of Kp, Ki and Kd on
respectively according to the fuzzy control rules of Table 2. the domain respectively. The change situation of Ki is shown
in Fig. 6. The surface display shows that during the simulation
process, the fuzzy controller will continuously modify the
numerical values of Kp, Ki, Kd according to the input data to
determine the parameters for the best control performance.

Fig.4 Main Interface of Simulink

As shown in Fig. 4, the Simulink interface has been set up,

in which the fuzzy controller and the PID controller are
combined to form a Fuzzy-PID controller. The Scope is Fig.6 Output surface of Ki on the domain of discourse
connected afterwards, on heels of which the period of the pulse
signal is set to 20s. By checking the adjustment time of each B. Experiments of Real-Time Flow Control System
PID signal for each step signal, we can observe the difference After finishing the simulation, in order to verify whether
of its waveform precisely. the system can work well in reality and have the same control
effect as the simulation. The design uses the Siemens S7-1200
PLC, whose full name is "programmable logic controller",
who's internal CPU has powerful computing power and control
functions, to replace the PC as the main controller, to imitate
the industrial production environment and set up the waterway
to conduct experiments.
The main structure of the hardware structure of the real-
time flow control system used in the design is shown in Fig. 7,
which uses the simplest two-channel liquid automatic mixing
device as an example to carry out the flow control experiment.
The water path uses the simplest pipeline to reduce the
influence of complex passages on the flow velocity, in order to
ensure the accuracy of the experiment.

Fig.5 Verification of Simulation results Fig.7 Main view of a two-channel real-time flow control system

The result of simulation shows that the fuzzy-PID control In order to facilitate the upgrade and maintenance, all the
based on ID3 is significantly shorter than the traditional PID programs adopt modular structure. The system consists of a
control in the respect of the rise and fall times. As shown in main controller PLC and several auxiliary devices. As shown
Fig. 5, there is about 8% overshoot in the traditional PID in Fig. 7, the complete flow control system includes Siemens
control. Compared with the former, the fuzzy PID can control S7-1215 PLC, KTP400 HMI, TR-M1708 electric regulating
the overshoot within 3%, by which the fluctuation of the flow valve (No.3), and FX-T1308 electromagnetic flowmeter
does not exceed ±2.5. The method has not only a small setting (No.4), which can be replaced by K24 turbine flowmeter
time, but also higher accuracy, strong robustness and better (No.5) according to actual conditions.

1706

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on March 15,2021 at 02:19:48 UTC from IEEE Xplore. Restrictions apply.
 In the experiment, the external waterway is built in the way it, many experiments were carried out and compared, after
of Fig. 7, and the PLC program is written in the form of a which the experimental results show that compared with the
traditional ladder diagram, whose external circuit uses current traditional PID, the new-fashioned fuzzy PID can speed up the
signal of 4-20mA for output and input. PLC's communication system response, has not only smaller adjustment time and
adopts the form of network cable to interact with KTP400 overshoot, but also good dynamic performance.
HMI, whose visual interface adopts Chinese mode, to transmit
C. Results and Analyses
data streams to each other. Even in the offline state, flow-
The experimental data can demonstrate the effect of the
changes on the experimental site can be monitored by HMI in
experiment clearly. But due to the defect that PLC can't
real time. The actual pipeline environment and the interface of
perform visual statistics and analysis of large data stream, for
HMI are shown in Fig. 8.
the data processing and drawing in the later stage of
experiments, we use Python's Matplotlib to collect, analyze
and plot the data stream, which is uniformly sampled and
exported inside the PLC in detail.

(a)

Fig.9 Data curves of Simple flow rate tests.

Fig. 9 shows that when the actual flow rate of the pipeline
is 6000 dm3/h and the target flow rate is set to 4800 dm3/h,
there are 3 response curves of the three control modes, of
which it is traditional PID, fuzzy PID and the one based on the
(b) ID3 algorithm.
Fig.8 (a) Waterway environment of real-time flow control system
(b) Visual interface of SIEMENS HMI A large volatility in the flow in reality has been fully shown
in the picture, which is a phenomenon that cannot be observed
The real water environment has more complex physical by simulation. It is difficult to grasp the regularity of these
principles and experimental procedures than simulation. Firstly fluctuations and emulate it in the simulation. However,
the liquid flows through the water pump for pressure. After the according to the actual conditions, as long as the fluctuation is
flow rate becomes larger, it passes through the regulating within ±100 dm3/h, it is in the permissible error range.
valve, whose opening can be controlled by the current signal From Fig. 9, we can see that when the fuzzy PID control
from the PLC. After the change of opening degree and the rate improved by the ID3 algorithm is adopted, the adjustment time
of water flow, the electromagnetic flowmeter can accurately of the system is about 7 seconds, and the overshoot is about
detects the flow rate, collecting the analog signal of the flow in 5.5%. Fast response times and small overshoots are well
the pipeline and converting it into a digital quantity and storing adapted to large-scale process control requirements.
it in the CPU of the PLC.
After receiving the data, the CPU performs fuzzy PID
calculation, calculates the appropriate opening value of
regulating valve, and converts the digital quantity into a
current signal to control the opening degree of the regulating
valve to adjust the water flow rate in time. With the high-speed
operating power of PLC, it can realize high-efficiency flow
control. Finally, the two regulated water flows flow into the
electric turbine mixer to complete the mixing process.
In the design, the PLC's processing of related events is
completed by calling each subroutine, of which it invokes the
traditional PID, the fuzzy PID, and the new-fashioned one
based on the ID3 algorithm in different subroutines. Based on Fig.10 Tests with a diverse gradient.

1707

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on March 15,2021 at 02:19:48 UTC from IEEE Xplore. Restrictions apply.
 In complex industrial situations, the system is required to In summary, the flow control system has good dynamic
have the ability to continuously and quickly adjust the flow in quality, fast adjustment time and small overshoot. The fuzzy
a short period of time. Fig. 10 shows the effects of the three rules selected by the ID3 algorithm have achieved more
methods in multiple adjustments according to different significant control effects and more adaptive capacity than by
gradients within 50 seconds. The picture demonstrates that in traditional fuzzy method.
the process, the fuzzy PID improved by the ID3 algorithm has
IV. CONCLUSIONS
a significantly faster response speed than the other two.
Especially when encountering large-scale adjustment, the In the paper, by combining fuzzy PID control with machine
method is obviously faster than the traditional PID, which can learning algorithm, we have designed a new PLC flow control
compensate for some hysteresis caused by the regulating valve system with a visual interface by HMI, and a new-fashioned
and effectively guarantee the stability of the work. fuzzy PID control method that does not rely on accurate
mathematical models and does not require users to master
much background knowledge. The control method can be
applied not only to the flow system but also to other domains
that require parameters to participate in decision making, such
as steelmaking and metallurgical industries.
Results of simulation and experiments show that the PLC
system has good adaptive ability, rapidity and stableness. It
will play an important role in improving the accuracy of
current flow system.
REFERENCES
[1] Qi Xing Liu, “Design of Flow Control System Based on Expert PID,”
2016 International Symposium on Computer, Consumer and Control
Fig.11 Tests with sharply increasing or reducing vibration (IS3C), pp.1031-1034, July 2016.
[2] Song Lepeng, Dong Zhiming, Xiang Lijuan, and Xing Siyuan, “Variable
Fig. 11 is a test conducted under the condition of sharply universe adaptive fuzzy PID control of spray flow valve,”
increasing or reducing vibration, to test the system's ability of TRANSACTIONS OF THE CHINESE SOCIETY OF AGRICULTURAL
ENGINEERING, vol. 26, pp.114-118, November 2010.
adjustment in more complicated industrial control, such as [3] Zhifei Yan, Chaoying Liu, and Xueling Song, “Application of Fuzzy
when its maximum adjustment extent is over 4000 dm3/h. The Adaptive PID Control in Chlorine Flow Control System,” 2013 Sixth
larger the adjustment of the flow is, the longer the time is International Symposium on Computational Intelligence and Design,
required for the traditional PID, and the more obvious the vol.2, pp.53-56, October 2013.
[4] Jörg Kiesbauer and Heinfried Hoffmann, “Modern regulating valves with
advantages of fuzzy PID will have. Compared with the fuzzy failure diagnostics,” 1999 European Control Conference (ECC), vol. 64,
rules according to expert experience, the response speed and no. 9, pp.670-674, September 2017.
overshoot of the ones according to the ID3 algorithm are also [5] Bin Zhang, Shuang Wang, and Yuting Liu, “Design of flow and pressure
having great superiority, due to its better adaptability to the coupling controller for independent metering valves control system,”
2016 12th IEEE/ASME International Conference on Mechatronic and
current flow environment. Embedded Systems and Applications (MESA), vol. 64, no. 9, pp.670-674,
The method combines fuzzy control and the algorithm of September 2017.
decision tree, which has not only the stability of traditional [6] Liang Zhang, Chunming Xia, and Jiabin Cao, “Physical-based modeling
PID control and the ability of eliminating steady-state error, of nonlinearities in process control valves,” 2012 International
Conference on Control Engineering and Communication Technology,
but also the advantage of fuzzy control, which has good 2013, pp.75-78.
dynamic performance and does not depend on complex [7] Si Huang, Guangqi Qiu, and Xianghui Su, “Flow performance analysis
mathematical models. To some extent, the new-fashioned flow on shutoff condition in centrifugal pump based on CFD simulation,”
system has improved the shortcomings of overshoot and large 2014 ISFMFE - 6th International Symposium on Fluid Machinery and
Fluid Engineering, vol. 2014, October 2014.
flow fluctuations of traditional PID. The experimental result [8] Sachin S. Gavankar and Sudhirkumar D. Sawarkar, “Eager decision
demonstrates that fuzzy PID control collaborating with ID3 tree,” 2017 2nd International Conference for Convergence in
algorithm can achieve good results for time-varying, pure time Technology (I2CT), vol. 2017-January, pp.834-840, December 2017.
delay, nonlinear and great inertia of first-order system. [9] Phaisarn Sutheebanjard and Wichian Premchaiswadi, “Fast convert OR-
decision table to decision tree,” 2010 Eighth International Conference on
The design, which synthesizes the advantages of PLC,
ICT and Knowledge Engineering, 2010, pp.37-40.
fuzzy PID, machine learning algorithm and so on, is combined [10] Hongtao Xie and Fuhua Shang, “The study of methods for post-pruning
with self-adaptive control, nonlinear control, robust control, decision trees based on comprehensive evaluation standard,” 2014 11th
and the latest artificial intelligence control, and successfully International Conference on Fuzzy Systems and Knowledge Discovery
designs a full-automatic liquid mixing device with high (FSKD), pp.903-908, December 2014.
[11] Zulfadhli Mazlan and R. Ibrahim, “Development and implementation of
stability and accuracy, high flexibility and timeliness, which Adaptive Fuzzy PID Controller (AFPIDC) for flow control application,”
can transform manual work into automation of PLC and allow 2012 4th International Conference on Intelligent and Advanced Systems
operators to get rid of complicated and rigorous work and (ICIAS2012), vol. 1, pp.1-6, June 2012.
guarantee safety and reliability of manufacture.

1708

Authorized licensed use limited to: Institut Teknologi Sepuluh Nopember. Downloaded on March 15,2021 at 02:19:48 UTC from IEEE Xplore. Restrictions apply.