1. Lecture Structure and Introduction

Transcription

1 Soft Control (AT 3, RMA) 1. Lecture Structure and Introduction

2 Table of Contents Computer Aided Methods in Automation Technology Expert Systems Application: Fault Finding Fuzzy Systems Application: Fuzzy Control (FC) Neural Networks (NN) Application: Identification and Neural Control Genetic Algorithms (GA), Simulated Annealing (SA) Application: Stochastic Optimization Basic Applications and Limitations of such Methods Soft Control 2

3 What is Soft Control? Three classes of control methods 1. Conventional Control (Classical Control) PID controller 2. Modern Control State-Based Control Model Predictive Control 3. Soft Control (Intelligent Control) Fuzzy Control Neural Network Genetic Algorithms Soft Control refers to those methods of control which use soft computing and computational intelligence. Soft Control = Intelligent Control = Knowledge-Based Scheme 3

4 Problems of Conventional Control To design a conventional controller, a Macroscopic model of the controller process is required The model may be based upon the empirical knowledge about the dynamics of the controlled process This knowledge can be obtained from measurements on control and controlled variables In practice, tuning of the control parameters is performed by the experts on a running system Example: Design of PID controller according to Ziegler and Nichols Advantages: Easy to use(few free parameters to configure, simple process model) Robust Problems: Increased complexity of the requirements and constraints Quality of control for complex controlled processes are often not sufficient 4

5 Problems of Modern Control For the design of modern control, a microscopic model of the controlled process is required. The model is determined through mathematical modeling Alternatively methods of identification can be used to ascertain the model Example: Design of state-based control Advantages: Strong mathematical basis (stability, etc.) High quality of control Possible to include additional constraints Problems: Building a mathematical model of the controlled process is difficult and sometimes impossible Detailed identification of process is often impossible or undesirable Resulting controllers are complex and difficult to understand for the users 5

6 Situation in the Industry Many conventional controllers at lower levels. Human as a controller at higher levels ADA systems (Supervisory Control and Data Acquisition) provides operators with all necessary information and access to the equipment Advantages: Operator can make intelligent decisions Operator can learn by experience Problems: Quality of control depends on the experience of the operator Interventions by the operator are subjective and often incomprehensive, error-prone (especially under stress) Under abnormal process conditions (alarm), the time delay in the decisionmaking by the operator or the wrong decision by him can lead to disasters (Chernobyl) 6

7 Consequences In modern complex systems, it is required that The operator performs the routine tasks that conventional controllers are unable to solve The support of the decision-making process is provided, especially in abnormal situations in which the operator is confronted often with conflicting signals and objectives In developing such systems Analytical process models are generally not available Objectives of the control scheme can often not be formulated precisely In certain cases this results in formulation of conflicting goals This requires intelligent controllers 7

8 Artificial Intelligence (Künstliche Intelligenz) The biggest objective of Artificial intelligence is to emulate the intelligent human behavior by means of computer programs. Symbolic and logic-based AI Systems to solve problems Systems for decision support Knowledge-Based Systems Formalisms for knowledge representation and AI programming languages Knowledge acquisition and machine learning Intelligence through behavioral simulation Turing Test Intelligence by symbol manipulation Chinese Room Philosophical discussion on the concepts of intelligence, perception, awareness is not the aim of the lecture Pragmatic approach 8

9 Computational Intelligence (Soft Computing) Artificial Intelligence Classical methods of artificial intelligence is based on the processing of symbolic data Example: Expert systems Computational Intelligence It refers to the methods that deal with numerical data Example: Fuzzy systems, Neuron Networks, Genetic algorithms Another denomination: Soft Computing Intelligent controllers are based on methods of soft computing, so the name Soft Control 9

10 Expert systems Core idea (Natural Model) Human-like abstract thinking History First expert systems began in 1970's (though faced the problem of high computing expenses) Application in Automation Engineering Today: Manifold industrial use higher levels of automation Examples Expert systems to support process control Expert systems for fault diagnosis Training Systems (Simulators) 10

11 Example of XPS: Diagnostic System in Process Control Source: Polke

12 Fuzzy Systems Core Idea (Natural Model) Dealing with fuzzy (non-crisp) knowledge History In the mid-1960s Zadeh fuzzy logic In the mid-1970s Mandani FuzzyControl Application in Automation Engineering First industrial applications in the early 1980s Fuzzy controller Examples Drying processes Gas heater Fuzzy control of an inverted pendulum Washing machine (AEG) Fuzzy control of a hammer drill 12

13 Example of Fuzzy: Control of a Hammer Drill Task: Automatic control of optimum speed and blow count with respect to drill diameter and material hardness. Solution: In total there are 20 IF-THEN rules for the determination of drill diameter and material hardness based on four measured variables Rule Nr. 11 as example: IF Power=average AND Longitudinal acceleration=high AND Transversal acceleration=high AND Longitudinal frequency=average THEN Drill diameter=24mm 13

14 Neural Networks Core Idea (Natural Model) Connective approach for knowledge, storage and processing (neurons in the brain) History Beginning in the 1970s Problems due to inadequate computing technology New interest in the 1980s Application in Automation Engineering Identification of complex processes Control by inverse model Prediction Examples Identification of nonlinear systems 14

15 Example of NN: Identification of a Two Tank System q Zu q ( k Zu - 1) h 1 h 2 q ( k Zu - h ( k 1 - h ( k 1-1) 2) 2) h ( ˆ1 k ) v 12 v a L 1 L a h (k) 1 h ˆ (k) 1 real Model

16 Genetic Algorithms Core Idea (Natural Model) Stochastic Optimization (Evolution in Nature) History Began in mid-1960s in Holland Application in Automation Engineering From the mid-1990s for complex optimization problems (Offline) Examples Optimizing control parameters especially with multiple degrees of freedom (Fuzzy Controller) 16

17 Interrelation Among the Methods Structure of Knowledge Processing Control Structured minimum (not adaptive) Expert systems Adaptivity Fuzzy Control Genetic Algorithms Neural Networks Fuzzy Rules Topology of Networks maximum Unstructured Populations Structure 17

18 Classification into the Lecture Natural Development If you look at the systems presented so far, we can say that we have looked at the intelligence from top-down : Technical Development Procedure in the lecture Expert Systems (Abstract mathematical thinking) are a further development of Fuzzy Systems ( "Natural" Fuzzy-Schlie sizes) these could only develop on the basis of the neural structures of the brain Neural Networks (Learning and adaptation) in the course of evolution arose from much simpler structures by Genetic Algorithms ( "Survival of the fittest") 18

19 Summary The problems of industrial domain require the use of "smart" controllers The research in the field of artificial intelligence and in particular the Computational Intelligence offers a number of methods The ideas are quite old Found its application only since a some years ( mainly due to computing power) The skepticism of the users has been significantly decreased 19

20 Outline of Lecture 1. Introduction to Soft Control: Definition and Limitations, Basics of Smart" Systems 2. Knowledge Representation and Knowledge Processing (Symbolic AI) Application: Expert Systems 3. Fuzzy Systems: Dealing with Fuzzy Knowledge Application: Fuzzy Control 4. Connective Systems: Neural Networks Usage: Identification and Neural Control 5. Genetic Algorithms: Stochastic Optimization Application: Optimization 6. Summary & Literature 20

21 Literature (Sources Used) General Information about the AI: Comprehensive Reference Book for the Interested Students Götz, Güntzer (Hrsg.): Handbuch der künstlichen Intelligenz. Oldenbourg Verlag, Expert Systems: Application Oriented Interpretation for the Use in Control Engineering: Polke, M.: Prozeßleittechnik. Oldenbourg Verlag, Ahrens, W.; Scheurlen, H.-J.; Spohr, G.-U.: Informationsorientierte Leittechnik. Oldenbourg Verlag, Methods of Computational Intelligence for the Automation Engineering : Fatikow, S.: Neuro- und Fuzzy- Steuerungsansätze in Robotik und Automation. Vorlesungsskript, Karlsruhe, King R.E.: Computational Intelligence in Control Engineering. Marcel Dekker,

22 Objectives of the Course To know what is the meaning of Soft Control To know the AI and specially Computational Intelligence for Automation Engineering related areas: Expert systems Fuzzy Systems Neural Networks Genetic Algorithms To know the application, advantages, and dis-advantages of each method To understand and apply the design methods; specially for Soft Control 22