Basics of Control Systems Modeling and Identication Analysis Synthesis

A Tutorial Overview of Control Theory for Non-Engineers

M. Vidyasagar
Cecil & Ida Green Professor, and Head, Bioengineering Department The University of Texas at Dallas [email protected]

ICM Satellite Meeting Hyderabad, 16 August 2010

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Outline
1

Basics of Control Systems What is a Control System? Examples of Control Systems

Modeling and Identication Various Types of Control Systems Models Considerations in Choosing a Model

Analysis Feedback and Feedforward Control Stability Analysis

Synthesis Two Kinds of Synthesis

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

What is a Control System? Examples of Control Systems

Outline
1

Basics of Control Systems What is a Control System? Examples of Control Systems

Modeling and Identication Various Types of Control Systems Models Considerations in Choosing a Model

Analysis Feedback and Feedforward Control Stability Analysis

Synthesis Two Kinds of Synthesis

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

What is a Control System? Examples of Control Systems

What is a Control System?

A control system consists of Inputs, which are things that we can not only measure, but to which we can assign chosen values (constants or functions of time). Examples: Drug dosages and treatment regimens. Outputs, which are things that we can measure, but to which we cannot assign values. Examples: Concentrations of administered drug in urine, blood, etc. States, which are things that aect the outputs, but which cannot even measure because we cannot directly access them. Examples: Concentrations of drug in targeted organ. A typical control problem consists of choosing the inputs so as to achieve one or more (possibly conicting) objectives.
M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

What is a Control System? Examples of Control Systems

Abstract Depiction of a Control System

Inputs (Assignable)

States (Not Measurable)

Outputs (Measurable)

M. Vidyasagar

Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

What is a Control System? Examples of Control Systems

Outline
1

Basics of Control Systems What is a Control System? Examples of Control Systems

Modeling and Identication Various Types of Control Systems Models Considerations in Choosing a Model

Analysis Feedback and Feedforward Control Stability Analysis

Synthesis Two Kinds of Synthesis

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

What is a Control System? Examples of Control Systems

Examples of Control Systems

Traditional examples: Thermostats for controlling room (or furnace) temperature Pilot control systems for ghter planes (which are statically unstable) Automotive: anti-skidding, fuel eciency, etc. Biological examples: Gene regulatory networks Insulin delivery and blood glucose regulation system (natural and articial) Any orally administered drug passing through the body
M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

What is a Control System? Examples of Control Systems

A Typical Gene Regulatory Network

Source: Faigle/Schrader Group, University of K oln

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

What is a Control System? Examples of Control Systems

An Angiogenic Regulatory Network

Source: Transcriptional network governing the angiogenic switch in human pancreatic cancer by A. Abdollahi et al., PNAS, July 31, 2007, Vol. 104, No. 31, pp. 12890-12895.
M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Various Types of Control Systems Models Considerations in Choosing a Model

Outline
1

Basics of Control Systems What is a Control System? Examples of Control Systems

Modeling and Identication Various Types of Control Systems Models Considerations in Choosing a Model

Analysis Feedback and Feedforward Control Stability Analysis

Synthesis Two Kinds of Synthesis

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Various Types of Control Systems Models Considerations in Choosing a Model

Classes of Control Systems

We can classify control systems according to: Nature of time: Continuous-time versus discrete-time Nature of quantities: Continuous-state versus discrete-state Nature of behavior: Deterministic versus stochastic All eight possible combinations have been (are being) studied. Most books on control theory are too mathematically demanding for biologists. I recommend Feedback Systems: An Introduction for Scientists and Engineers by Karl-Johan Astrom and Richard M. Murray, Princeton University Press, 2008. (Freely downloadable!)

M. Vidyasagar

Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Various Types of Control Systems Models Considerations in Choosing a Model

Various Types of Control Systems Models

For the most part (though not always), models of control systems fall into one of two types: State-space, and input-output. State-space models usually arise from rst-principles modeling or bottom-up modeling. In the continuous-time, continuous-state case, the models usually take the form = f (x, u), y = h(x, u), x where u, y, x denote respectively the input variables, the output variables, and the state variables. If f , h are linear functions of their arguments, then the system itself is said to be linear.
M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Various Types of Control Systems Models Considerations in Choosing a Model

Role of Physiological Parameters

If we wish, we can separately identify a set of physiological parameters p and display the state equations in the form = f (x, u, p), y = h(x, u, p). x Example: Michelis-Menten kinetics (compartmental) model of drug concentrations in various organs. c i (t) = where i (t) =
j =1 n

Vmax i (t) , Ki + i (t)

ij cj (t) + ui (t), i.

Here Ki , ij are physiological parameters.

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Various Types of Control Systems Models Considerations in Choosing a Model

Outline
1

Basics of Control Systems What is a Control System? Examples of Control Systems

Modeling and Identication Various Types of Control Systems Models Considerations in Choosing a Model

Analysis Feedback and Feedforward Control Stability Analysis

Synthesis Two Kinds of Synthesis

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Various Types of Control Systems Models Considerations in Choosing a Model

Identication of Parameters
Once a model of the form = f (x, u, p), y = h(x, u, p) x is formulated, the next step is to identify the parameters. Useful rule of thumb: If there are n parameters in the vector p, at least n2 data points are needed for a statistically signicant estimate of p. If data size is too small, we get into a problem known as memorization the model perfectly reproduces observed data but generalizes very poorly.

M. Vidyasagar

Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Various Types of Control Systems Models Considerations in Choosing a Model

Input-Output (Black-Box) Models

In this kind of modeling, we perform experiments by applying various input signals u() and measure the resulting output signals y(). Then we t some models to the observations. Advantages of I/O-models: Model size is chosen in accordance with available data; so memorizing data is avoided. Advantages of rst-principles models: Every number in the model has a physical interpretation. But usually the models are far too complex.

M. Vidyasagar

Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Various Types of Control Systems Models Considerations in Choosing a Model

An Overly Complex Model of a Control System

Source: Transcriptional network governing the angiogenic switch in human pancreatic cancer by A. Abdollahi et al., PNAS, July 31, 2007, Vol. 104, No. 31, pp. 12890-12895.
M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Various Types of Control Systems Models Considerations in Choosing a Model

Pitfalls in Overly Complex Models

When the model is overly complex, all attempts to t the model to data will only result in garbage! System simplication must be practiced before trying to t parameters (models) to data. Need of the hour: A model simplication methodology that keeps in tact the subsystem of interest while simplifying everything else. Well-established with linear input-output models and state-space models. Not so well-established for nonlinear models. In most problems involving drug delivery, simple input-output models are more than adequate. Example: Work of Prof. Frank Doyle on insulin control systems.
M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

Outline
1

Basics of Control Systems What is a Control System? Examples of Control Systems

Modeling and Identication Various Types of Control Systems Models Considerations in Choosing a Model

Analysis Feedback and Feedforward Control Stability Analysis

Synthesis Two Kinds of Synthesis

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

What is Feedback?
Feedback refers to the process whereby the measured output is processed and then used to adjust the input to the control system.

Input System

Output

Feedback Signal Controller

M. Vidyasagar

Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

Kinds of Feedback

Feedback can be of two kinds: Positive and negative. Negative feedback is by far the most common and used in problems of stabilization and regulation. Positive feedback is used to cause controlled instability, which is required to maintain periodic processes, for example circadian rhythm.

M. Vidyasagar

Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

Tracking Via Unit Negative Feedback

To ensure that the output signal tracks the input signal, the following conguration is used.

Input +

Error Controller System

Output

The tracking error, processed by the controller, drives the system.

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

Type 1 Control Systems

Unit feedback systems were used by James Watt in his governor in 1788! Contemporary examples are: Thermostats used in air conditioners, and almost all instances of homeostasis in biology. In order to ensure that the tracking error goes to zero in response to a step input, even if the system is imprecisely known, the controller must incorporate an integrator. This is known as a Type 1 control system. Much of what is called homeostasis in biology can be given this interpretation.

M. Vidyasagar

Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

Feedback Reduces Sensitivity and Uncertainty

G + K

Suppose G, K are just constants. Overall (static) gain is H= S := G 1 if GK 1 + GK K 1. 1.

dH/dG 1 d log H = = 0 if GK d log G H/G 1 + GK

M. Vidyasagar Overview of Control Theory

Closed-loop gain is virtually insensitive to G! (Black 1927)

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

Disturbances: Measured and Unmeasured

In addition to the control inputs, measured outputs, and non-measurable states, there can also be disturbances in the system. Disturbances can be of two kinds: Non-measurable, such as measurement errors (noise). Measurable (or at least, estimable), such as food intake for a diabetic. For the former, ltering is the only approach (not discussed here). For the latter, feedforward is a widely used approach. The next slides show a control system with disturbance, and with feedforward control.
M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

Control System with Disturbance

Disturbance

Input Controller System

Output

If disturbance cannot be measured, the controller has to incorporate ltering to attenuate eect of disturbance.
M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

Control System with Feedforward Control

Disturbance

Input Controller System

Output

If disturbance can be measured, it can be another input to the controller.

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

Outline
1

Basics of Control Systems What is a Control System? Examples of Control Systems

Modeling and Identication Various Types of Control Systems Models Considerations in Choosing a Model

Analysis Feedback and Feedforward Control Stability Analysis

Synthesis Two Kinds of Synthesis

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

Lyapunov Stability
For the usual parametrized model = f (x, u, p), y = h(x, u, p), x suppose that (i) u = 0, so that the system is unforced (no input), and (ii) a state vector x0 has the property that f (x0 , 0, p) = 0. Then x0 is called an equilibrium. Lyapunov stability theory addresses the question of what happens if the initial state of the system x(0) is close to the equilibrium. Specically, is it the case that x(t) x0 as t ? If so the equilibrium x0 is said to be asymptotically stable.

M. Vidyasagar

Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

The Lyapunov Function

The basic idea of Lyapunov stability theory is that of the energy function, usually denoted by V . If V achieves its (local) minimum at the equilibrium x0 and if V (x(t)) is strictly decreasing along the solution trajectories, then x(t) x0 as t and the equilibrium x0 is asymptotically stable. The same methodology can also be used to analyze the stability of periodic orbits also.

M. Vidyasagar

Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Feedback and Feedforward Control Stability Analysis

Input-Output Stability
Given the system = f (x, u, p), y = h(x, u, p), x we can also ask, what kind of state trajectory x() or output trajectory y() results from applying an input signal u()? Such questions form the domain of input-output stability theory. Though connections exist between Lyapunov and I/O stability, in general the latter requires more advanced mathematics. But its applicability is also broader. For instance, the presence of delays (omnipresent in biology) is easily handled in I/O framework, not so easily in Lyapunov framework.
M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Two Kinds of Synthesis

Outline
1

Basics of Control Systems What is a Control System? Examples of Control Systems

Modeling and Identication Various Types of Control Systems Models Considerations in Choosing a Model

Analysis Feedback and Feedforward Control Stability Analysis

Synthesis Two Kinds of Synthesis

M. Vidyasagar Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Two Kinds of Synthesis

Conventional Controller Design

In conventional control theory, synthesis usually refers to the process of designing a controller (as shown on the next slide) for a given system, so as to make the overall system behave as desired (or close to it). The premise is that before coming to this stage, the system has already been optimized (ne-tuned) as much as possible. This kind of philosophy is quite pertinent in problems where there is an external control signal being applied, e.g. insulin delivery (or more generally drug delivery of any sort), pacemakers, etc. Standard methods exist, ranging from very elementary (PID, MPC) to very advanced (H control).

M. Vidyasagar

Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Two Kinds of Synthesis

Synthesis in the Context of Biological Systems

In biological contexts, often synthesis is the mirror image of analysis whereby one keeps tinkering with the system until it behaves as desired without any controller. Example: Synthetic biology, design (not control ) of genetic regulatory networks. In this case synthesis becomes essentially iterative analysis.

M. Vidyasagar

Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Two Kinds of Synthesis

Special Features of Biological Control Systems

Control enters via parameters and is often just a constant, not a function of time. (Example: HIV). Strong role of physical topology (e.g. compartmental models in PK/PD) Control and state variables restricted to be nonnegative-valued (e.g. drug can only be injected, not taken out; concentrations can only be nonnegative) Presence of signicant measurement delays. Existing control theory does not always take into account these special features.

M. Vidyasagar

Overview of Control Theory

Basics of Control Systems Modeling and Identication Analysis Synthesis

Two Kinds of Synthesis

Conclusions
Control theory can only provide the methodology and broad concepts needed to analyze and synthesize biological systems. Specic procedures and results can only come from thinking deeply about the problems at hand. In the 1960s (my Ph.D. days) we discovered that solutions to control theory problems arent found in mathematics books we had to invent our own theories! Today the same is also true about computational (or systems of synthetic) biology too! Thank You!

M. Vidyasagar

Overview of Control Theory