Advanced Intelligent Adaptive Control Systems

• Elective Course (Type D) in English at the Master Degree LM-32 INGEGNERIA INFORMATICA E DELL'AUTOMAZIONE

• Google Classroom Course Code: viwlscb (for course enrollment only)
• In-Person Lectures in PC Laboratory or Classrooms with PCs

Required Background

• Basic concepts of mathematics, differential and integral computation;
• knowledge of the basic concepts of Physics;
• knowledge of dynamic systems, their behaviour, and their practical application; methods to analyse dynamic systems in steady and transient states;
• knowledge of the frequency tools for the analysis of dynamic systems;
• ability to analyse and design digital systems.

Main Objectives and Knowledge

• Basic nonlinear control and identification techniques for complex processes from a dynamic point of view, considering the information from its input to its output variables
• knowledge related to the analysis of nonlinear dynamic systems in steady and transient states and their advanced simulation tools;
• knowledge of nonlinear blocks and system identification techniques;
• knowledge of the tools to tackle the study of complex systems and their interconnections under the constraints imposed by performances in terms of cost, speed, computational cost, and robustness;
• knowledge of the nonlinear mathematical tools for the analysis of nonlinear dynamic systems;
• basic knowledge of nonlinear dynamic system software simulation tools;
• fundamentals about system identification, neural networks and fuzzy logic for control, nonlinear methods for control, adaptive and learning algorithms, basics of optimization methods and tools, nonlinear prediction and filtering tools;
• fundamentals of simulation and control for nonlinear dynamic systems.

Main Skills

• Analysis of the behaviour of nonlinear systems in steady and dynamic conditions;
• design of nonlinear dynamic and adaptive controllers for a given nonlinear dynamic system to meet proper transient and steady-state constraints;
• design the most suitable system identification algorithm for the design of intelligent control solutions using fuzzy systems, neural networks, adaptive systems, nonlinear filters, adaptive schemes, and purely nonlinear prototypes;
• identification of the most suitable process model, as well as the most suitable parameters for a specific control design and its application, given its model prototype;
• use of simulation numerical programs to analyse nonlinear systems, their stability properties and the control performance.

General Description and Main Goals

• The course represents an advanced course of nonlinear and adaptive control for complex systems, and it studies advanced elements of a control and adaptive system from the dynamic point of view, by considering nonlinear dynamic processes from their input-state-output and input-output points of view.
• The course thus aims at presenting those adaptive control and data-driven methodologies currently required and expected by modern industries and practical application activities.
• The main goal of the course consists of providing advanced topics and tools for the study of adaptation, learning, and control of complex dynamic systems, as well as their interconnections under proper design constraints imposed by cost, speed, computational cost, robustness, reliability, sustainability, and power consumption.

Course Contents

The course consists of 60 hours of teaching activities divided into frontal lectures (45 hours) and guided tutorials in the computer laboratories (15 hours). In more detail, the following topics will be analysed and investigated:
• Introduction. Adaptive, intelligent, autonomous, distributed, embedded and cyber-physical systems. Key tools and multidisciplinary techniques for the comprehension, analysis, representation and synthesis of complex physical phenomena. Introduction to adaptive systems and adaptation theory for control.
• Basics of optimization tools. Constrained and constrained optimization; gradient method; stochastic approaches; genetic algorithms.
• Dynamic system identification. Parametric and nonparametric identification; recursive algorithms for linear system identification; online models and mathematical tools for nonlinear dynamic system identification; adaptive identification and control approaches; adaptive PID and classical controllers.
• Fuzzy logic for control. Definitions and properties of fuzzy logic; fuzzy model identification; fuzzy logic for control; fuzzy control: automatic learning and adaptation for fuzzy models; adaptive fuzzy control; ANFIS tool - Adaptive Neuro-Fuzzy Inference System.
• Neural Networks. Fundamentals and properties. Algorithms for supervised and unsupervised learning, with application to identification and control of dynamic systems; stochastic search algorithms; recurrent neural networks; adaptive neural networks; convolutional neural networks for identification and control; design of the adaptive neural controller employing the Model Reference Adaptive Control (MRAC) principle.
• Computer Aided Hands-on (computer-aided design). Hands-on with the identification of nonlinear dynamic systems, fuzzy logic for control, neural networks, and design of adaptive control schemes.

Teaching Structure

The course is organised as follows:
• 45 hours of lectures on all the course topics;
• 15 hours of practical exercises in the Informatics Laboratory concerning the analysis and the simulation of nonlinear dynamic systems.
• After the guided tutorials the students will have free access to the computer laboratories for additional individual tests and hands-on.

Examination Procedure

The exam aims to verify at which level the learning objectives previously described have been acquired. The examination is divided in 2 sections that will take place on the same day.
• A project regarding the simulation and the control design for a nonlinear system using the Matlab and Simulink environments, which aims at understanding if the student has the skills in the analysis and the synthesis of a complex process. To pass this test it is required to get at least 18 points out of 22. The time allowed for this test is 1.5 hours. It is allowed to consult the Matlab and Simulink program manual only.
• One test (8 open and multiple choice questions) based on all the topics tackled in the class or on the basic concepts of the course, to evaluate how deeply the student has studied the subject and how he can understand the topics analysed. To pass this test it is required to get at least 1 point out of 8. The time allowed for this test is 0.5 hours. It is not allowed to consult any textbook or use any PC, smartphone, or calculator...
• The final mark is the sum of the 2 marks. To pass the exam it is necessary to get at least 18 points out of 31. If the first test fails or if the final mark is below 18, it is necessary to repeat all the exam steps.
• Passing the exam is proof of having acquired the ability to apply knowledge and the required skills defined in the course training objectives.

Course Material: Lecture Notes, Slides and Recorded Lectures

• Lecture 0. Course Introduction: Analysis of the course Syllabus and teaching material. Lecture 0 recorded movie (YouTube Channel).
• Lecture 1. Introduction - General introduction to modelling and system identification. Lecture 1. (Lesson Slides PDF Format, 3.9MB). Lecture 1 recorded movie (YouTube Channel).
• Lecture 2. Non-recursive (off-line) system identification methods. Lecture 2. (Lesson Slides PDF Format, 2.5MB). Lecture 2 - Least Squares and Regression Models (part 1) recorded movie (YouTube Channel); Lecture 2 PEM and IMV Estimation Schemes (part 2) recorded movie (YouTube Channel).
• Lecture 2.5. Nonparametric Identification. Input Signals for Identification. Identification Conditions. Lecture 2.5. (Lesson Slides PDF Format, 1.4MB). Lecture 2.5 - Nonparametric Models (part 1) recorded movie (YouTube Channel); Lecture 2.5 Input Signals (part 2) recorded movie (YouTube Channel).
• Lecture 3. Recursive (on-line) methods. Lecture 3. (Lesson Slides PDF Format, 1.7MB). Lecture 3 recorded movie (YouTube Channel).
• Lecture 3.5. Recursive (on-line) methods: 2 application examples in Matlab. Matlab script file 1 for Lecture 3; Matlab script file 2 for Lecture 3; Lecture 3.5 recorded movie (YouTube Channel).
• Lecture 4. Prediction Error Methods. Lecture 4. (Lesson Slides PDF Format, 2MB). Lecture 4 recorded movie (YouTube Channel).
• Lecture 4.6. Model Structure Determination. Model Validation. Lecture 4.6. (Lesson Slides PDF Format, 2.6MB). Lecture 4.6 recorded movie (YouTube Channel).
• Lecture 5. An Application Example with the System Identification Toolbox of Matlab. Lecture 5. (Lesson Slides PDF Format, 6.5MB). Lecture 5 recorded movie (YouTube Channel).
• Lecture 6. Remarks on System Identification and Data Analisys. Lecture 6. (Lesson Slides PDF Format, 2.8MB). Lecture 6 recorded movie (YouTube Channel).
• Lecture 6.5. The System Identification Toolbox by MathWorks. Tutorial from MathWorks Website. Lecture 6.5 webinar by Lennart Ljung (YouTube Channel). Lecture 6.5 recorded movie (YouTube Channel).
• Lecture 7. Control Refresher. Lecture 7. (Lesson Slides PDF Format, 232kB). Lecture 7. (Lesson Slides PDF Format, 2.8MB). Lecture 7 recorded movie (YouTube Channel).
• Lecture 8. Introduction to Fuzzy Logic and Fuzzy Control. Lecture 8. (Lesson Slides PDF Format, 386kB). Lecture 8 - Introduction to Fuzzy Logic (part 1) recorded movie (YouTube Channel); Lecture 8 - Introduction to Fuzzy Models (part 2) recorded movie (YouTube Channel).
• Lecture 9. Fundamentals of Fuzzy Control. Lecture 9. (Lesson Slides PDF Format, 1.2MB). Lecture 9 - Introduction to Fuzzy Control (part 1) recorded movie (YouTube Channel); Lecture 9 - Data-Driven Fuzzy Control (part 2) recorded movie (YouTube Channel); Lecture 9 - Examples of Designs of Fuzzy Controllers (part 3) recorded movie (YouTube Channel).
• Lecture 10. Fundamentals of Model-Based Control. Lecture 10. (Lesson Slides PDF Format, 240kB); Lecture 10 - Overview of Model-Based and Data-Driven Control Techniques recorded movie (YouTube Channel).
• Lecture 11. Introduction to Artificial Neural Networks for Control. Lecture 11. (Lesson Slides PDF Format, 1.2MB); Lecture 11 - Introduction to Machine Learning and Neural Networks (part 1) recorded movie (YouTube Channel); Lecture 11 - Fundamentals of Deep Learning (part 2) recorded movie (YouTube Channel); Lecture 11 - Neural Network Training (part 3) recorded movie (YouTube Channel).
• Lecture 12. Artificial Neural Networks for Control and Convolutional Architectures. Lecture 12. (Lesson Slides PDF Format, 4.5MB); Lecture 12 - Neural Network Regularization (part 1) recorded movie (YouTube Channel); Lecture 12 - Fundamentals of Deep Neural Networks and Reinforcement Learning (part 2) recorded movie (YouTube Channel).
• Lecture 13. Remarks and Application Examples. Lecture 13. (Lesson Slides PDF Format, 6.4MB); Lecture 13 - Examples of design of fuzzy regulators for different processes recorded movie (YouTube Channel).
• Lecture 14. Fuzzy Models and Neural Networks for Control: Fuzzy Clustering, and Neural Design Examples. Lecture 14. (Lesson Slides PDF Format, 2.8MB); Lecture 14 - Notes on the design of dynamic fuzzy systems and neural networks for control recorded movie (YouTube Channel).

General Application Examples and Control Design Cases

• Real Chemical SISO Process (CSTR, Prof. Robert Babuska, 1998) and Examples of Controller Designs (PID, MPC). (zipped folder .zip); (.txt file with a few details); Example of identification of ARX models and design of PID/MPC controllers recorded movie (YouTube Channel).
• Recursive Estimation for Linear Varying Parameters of a Dynamic Process: Matlab and Simulink Solutions. Implementation of the RLS algorithm with forgetting factor (example); Matlab script for result display; Simulink file for data generation with linear varying parameters; RLS Simulink block and data generation; Recursive estimation example of identification of ARX models and design of PID/MPC controllers recorded movie (YouTube Channel).
• Neural Network Example for the Estimation of a Static Nonlinear Function. Data generation in Simulink for neural network training; Neural network training Matlab script file (feedforward MLP and Back-Propagation); Simulink file for neural network validation (NN to be included); Matlab script for result comparison; Simulink file for result validation (NN included); Neural network training for the approximation of a static nonlinear function recorded movie (YouTube Channel).
• Examples of Adaptive Neuro-Fuzzy Inference System - ANFIS and NN for Dynamic Process Modelling: Chemical Reactor Case Study. Compressed folder with files used in the simulations (.zip file); Neural network and fuzzy system for the approximation of a dynamic nonlinear process recorded movie (YouTube Channel).
• Example of a Simple Adaptive PID Controller. Compressed folder with files used in the simulations (.zip file); Design of a Simple Adaptive PID Controller Example recorded movie (YouTube Channel).
• Design of a Reference Model and Training of the Neural Controller for the Chemical Reactor Case. Reference Model Design: compressed folder with files used in the simulations - part 1 (.zip file); Design of the Reference Model recorded movie - part 1 (YouTube Channel); Data Augmentation example for Reference Model design: compressed folder with files used in the simulations - part 2 (.zip file); Design of the Neural Controller recorded movie - part 2 (YouTube Channel); Neural Controller Design from the Reference Model: compressed folder with files used in the simulations - part 3 (.zip file); Neural Controller Design from Reference Model recorded movie - part 3 (YouTube Channel).
• Design of an Adaptive Reference Model for the Training of Intelligent Controllers Applied to a Directional Antenna Nonlinear Process. Files for the Design of the Reference Model and the Intelligent Controllers (.zip file); Design of the Linear Model and the RLS with FF Solutions - part 1 (YouTube Channel); Derivation of the Neural Model of the Process - part 2 (YouTube Channel); Estimation of the Fuzzy Model of the Process - part 3 (YouTube Channel); Design of the Intelligent Controller with Neural Networks - part 4 (YouTube Channel); Design of the Intelligent Controller with the Fuzzy Models - part 5 (YouTube Channel); Derivation of the Intelligent Controller and Final Validation - part 6 (YouTube Channel).

Matlab and Simulink Basic Docs

• K. Sigmon, "Matlab Primer". University of Florida, Florida, Second Edition ed., 1993. (PDF format file).
• "Matlab Tutorial", Getting Started with MATLAB. (PDF format file).
• Simulink Main Features, "SIMULINK. Dynamic System Simulation for MATLAB". (PDF format file).
• Simulink for Dynamic System Modelling. Dynamic System Simulation for MATLAB and SIMULINK. (PDF format file).
• Simulink Basic Features. SIMULINK for Beginners. (PDF format file).

References and Textbooks

• Applied nonlinear control, J.J. Slotine, W. Li. - Prentice Hall, 1991.
• A Course in Fuzzy Systems and Control, L.-X. Wang - Prentice Hall, 1997.
• Neural Networks for Identification, Prediction, and Control, D.T. Pham and X. Liu - Springer Verlag, 1995.
• Model-based Fault Diagnosis in Dynamic Systems Using Identification Techniques, S. Simani, C. Fantuzzi, R. J. Patton - Springer, 2003.
• Robust and Adaptive Control with Aerospace Applications, E. Lavretsky, K. A. Wise, Springer-Verlag, London, 2013
• Adaptive Control. Second Edition, K. J. Astrom,, B. Wittenmark, Addison Wesley, 1995
• System Identification, T. Soderstrom, P. Stoica, Prentice Hall International, Cambridge, 1989
• Adaptive Control. Algorithms, Analysis and Applications, I. D. Landau, R. Lozano, M. M'Saad, A. Karimi, Springer, London, 2011
• Stable Adaptive Systems, K. S. Narendra, A. M. Annaswamy, Dover Publications, New York, 2005
• Robust Adaptive Control, P. Ioannou, J. Sun, Dover Publications, New York, 2012
• Adaptive Control Tutorial, P. Ioannou, B. Fidan, Advances in Design and Control, SIAM, Philadelphia 2006

Previous Page

Simani Home Page

Dipartimento di Ingegneria