1. A New Generation of Adaptive Control: An Intelligent Supervisory Loop Approach
A Dissertation Proposal Presentation By
Sukumar Kamalasadan
Department of Electrical Engineering and Computer Science,
University of Toledo,
30th April 2003
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

2. Dynamic Systems Operates in Real time Specifies Performance Quality
Regardless of External Disturbance
Complex Dynamic Systems
Uncertainties: Functional and Parametric
Time Varying and/or nonlinear elements

3. The Control Challenge, A Practical Matter :
Practical Systems are mostly Nonlinear and Shows some degree of Uncertainty
Advances in technology led to highly complex processes, to be controlled with tight specifications and high level of autonomy
Example: Fighter Aircrafts

4. Practical Approaches to the Control Design Problem
Systems that can be modeled “Adequately” with stationary Linear Models:
Fixed Parameters (Stationary) Controllers Designed off line.
Mostly used for Linear Time Invariant Systems
Systems that CANNOT be modeled “Satisfactorily” with stationary Linear Models:
Adaptive Controllers (STR and MRAC)
Sophistication Level # 1
Intelligent Adaptive Controllers (A New Generation)
Sophistication Level # 2
Which Implies Certain Levels of
Learning and Adaptation

5. Research Motivation
Investigate possibilities of some Intelligence based solutions to a major structural problem that exists in the two “conventional” Adaptive Control techniques (MRAC & STR ):
The Problem:
The Designer’s A priori Choices, such as the choice of a “MODEL” as required in either of the two Schemes
Inability to Control functionally nonlinear and Changing systems
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

6. Intelligent Adaptive Control
What is Intelligent Control ?
Controls complex uncertain systems within stringent specification
Features
Ability to Learn: Ability to modify behavior when condition changes
Ability to Adapt: Ability to handle uncertainty by continuously estimating the relevant unknown knowledge
Ability to deal with Complex Systems : Characterized by nonlinear dynamics and multiple mode of operation
Autonomous in Nature: Ability to deal with uncertainty all by itself without human intervention
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

7. Intelligent Adaptive Control : Constituents
Deals with linear or nonlinear parametric uncertain Systems
Needs detailed prior knowledge of the systems to be controlled
Have the ability to adapt
Artificial Intelligence (AI) Techniques
Neural Networks
Ability to learn either offline or online
Adjusts the parametric values allow the network to learn
Fuzzy Systems
Ability to fuzzify a complex system in terms of linguistic rules
Can avoid dealing with complex mathematical models
Create the “long term memory” or learning behavior
Reduce the uncertainty in dealing with models

8. Intelligent Adaptive Control : Applications
Objective
Control of Complex systems which is affine but shows “ Multi Modal” and Sudden parametric ‘Jumps’
Control of Nonlinear Systems which shows “Functional Uncertainty”
Control of Nonlinear Systems which shows “Functional Uncertainty” and “Multi Modal”

9. Statement of Dissertation Objectives
Theoretical Design and Development of Three Intelligent Adaptive Control Schemes
Develop an F-16 Aircraft Model in MATLAB for Investigation and Application
Classification
Development of the F-16 Aircraft MATLAB Model
Fuzzy Switching Multiple Reference Model Adaptive Control Scheme
Neural Network Adaptive Control Scheme
Neuro-Fuzzy Adaptive Control Scheme

10. Current Status of Dissertation
Development of a 6 Degree of Freedom (6 DOF) dynamic F16 Aircraft Model in MATLAB and SIMULINK
Development of a Fuzzy Switching Multiple Reference Model Adaptive Controller
Development of a Neural Network Adaptive Controller
Development of Neuro-Fuzzy Adaptive Controller
Overall Dissertation Status

11. Concluding Remarks
Three Intelligent Adaptive Control schemes are proposed
Objective is to control a class of multimodal nonlinear systems which deals with function and/or parametric uncertainty
Application systems which shows changes influenced by external or internal disturbance
A nonlinear Aircraft Model is developed to simulate as an appropriate application system, and to investigate and verify the effectiveness of schemes
Preliminary Simulation Results appear to be promising

13. Typical Stationary Controller
Regulator
Plant
Control Signal
Command
Signal y
Output
Parameters
Control processor A
stationary
(Fixed
Parameter)
Controller is
designed (
Off Line
) For
The Plant as represented
by a
Stationary M
odel

14. Self Tuning Regulator (STR) Scheme

15. Model Reference Adaptive Control (MRAC) Scheme

16. Development of the F-16 Aircraft MATLAB Model
Developing the Building Blocks
Developing the Algorithm in MATLAB including the subroutine functions and the main equations of motions
Testing with certain developed Trim conditions
Developing the SIMULINK Model 

17. F-16 Aircraft Body System Axes and Variables

18. F-16 Aircraft Model Building Blocks
Aerodynamic
Model
6DOF
Equations
Of Motion
Actuator
Modeling
Atmospheric
Model
Engine Model
Control
deflections

19. Development of the F-16 Aircraft MATLAB Model
Computing Air data
Outputs: - Mach number, Dynamic Pressure Inputs: -Velocity, Altitude
Aerodynamic look-up table and coefficient buildup
Outputs: - Aerodynamic Force (Cxt, Cyt, Czt) & Moments (Cnt, Clt, Cmt) coefficients
Inputs: -Control Variables (elev, ail, rdr) and (alpha, beta)
Computing Engine Model Outputs: - Engine Thrust
Inputs: -Power, Altitude, Mach Number
State Equations
Force Equations Derivative, Inputs: -Moment Rates (P, Q, R), Velocity (UVW), Kinematics (Phi, Theta) and Aerodynamic Force coefficients
Outputs: - Vt, Alpha and Beta Derivatives
Kinematic Equations Derivative, Inputs: -Moment Rates (P, Q and R), Kinematics (Phi and Theta)
Outputs: - Phi, Theta and Psi Derivaties
Moments Equations Derivative, Inputs: -Moment Rates (P, Q, R), Aerodynamic Moment Coefficient (Clt,Cmt.Cnt) and Inertia Constants
Outputs: - Moments Derivatives
Navigation Equations Derivative
Inputs: -Moment Rates (P, Q, R), Aerodynamic Moment Coefficient (Clt,Cmt.Cnt) and Inertia Constants
Control Vector

20. Development of Steady State Trim Conditions
Trim Conditions are Developed based on a Simplex Routine
Table Below Shows the Trim conditions for five cases
Conditions
Variables
Nominal
Xcg=0.38C
VT(ft/sec)
502.0
5020
(rad)
0.2485
0.3006
(rad)
-4.0E-9
4.1E-9
3.1E-8
4.8E-4
4.1E-5
(rad)
1.367
(rad)
P(rad/sec)
Q(rad/sec)
0.2934
0.3000
R(rad/sec)
Thtl(0-1)
0.1385
0.1485
0.1325
0.8499
1.023
El(deg)
-1.931
-6.256
-7.082
Ail(deg)
-1.2E-7
-7.0E-8
-5.1E-7
-6.2E-4
Rdr(deg)
-6.2E-7
8.3E-7
-4.3E-6
Reference Models
(90S+287)
(S S2+115S+28)
(110S+287)
(S S2+115S+287)
(235S+4163)
(S3+40S2+608S+4163)
(10S+287)
(S S2+115S+287
(132S+287)

21. Developed SIMULINK model of F16 Aircraft

22.  Proposed Scheme I Scheme Outline
Developing the Model Reference Control Law
Development of Reference Models for each operating modes
Testing the operation by manually switching the Reference Model
Developing the Fuzzy Logic Scheme depending on the System
Testing overall system with the Dynamic Fuzzy Switching Scheme 
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

23. Fuzzy Logic Switching Scheme (FLSS)
Proposed Scheme I :
Ref. Model 1
Ref. Model 2
Ref. Model n
Command Signal
Control Signal
Aux. Inputs y
Regulator Parameters
Error
Fuzzy Logic Switching Scheme (FLSS)
Output +
Regulator
Plant
Adjustment Mechanism - 
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

24. MRAC Structure
Develops a Control Law looking at the Input and Output of the Plant
Updates the Control law using an Adaptive Mechanism
Use a reference model to effectively model the dynamics and forces the plant to follow that model

25.  Proposed Scheme II Scheme Outline
Design of the Dynamic Radial Basis Neural Network (RBFNN)
Development of overall scheme linking the RBFNN control with Adaptive Control
Testing the Scheme on a Functionally Nonlinear System 
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

26. Neural Network Controller
Proposed Scheme II +
Usl -
Umr
Unn em ym yp
Neural Network Controller
Nonlinear Process
MRAC Controller
Adjustment Mechanism 
Reference Model
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

27. Features of Proposed Neural Network
Centers, Radius and Distance adapt with time looking at input vector
Dynamic in Nature
Starts with three nodes and grows depending on functional complexity
Grows accordingly
RBFNN weights adjust to correct the Output and Reference Error
Learns Online
Features
Neural Network
Radial Basis Function Neural Network

28. RBFNN Structure Consists of Nodes in Input layer
Nodes basically have two elements : Center and Radius
Consists of a basis function which is a Gaussian Function
The output is the summation of each functions times the weights

29. Proposed Scheme II (Nonlinear Functional Uncertain System)
Highlights
RBFNN
Center
Grows depending on new Inputs
Moves close to Input Set
Radius : Changes for each center addition
Weights: Adapts Depending on the Error
MRAC
Stable Direct Model Reference Framework
Sliding Mode Gain and Rate Increase
Reduces Network Approximation Error
Reduces Parametric Drift especially in the Boundary Region 
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

30.  Proposed Scheme III Scheme Outline Design of the RBFNN Control
Design of Fuzzy Logic Scheme depending on the System
Development of the Reference Model
Integrating overall scheme
Testing the system on a Functionally Nonlinear Parametrically Uncertain System 
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

31. Neural Network Controller
Proposed Scheme III +
Usl -
Umr
Unn em ym yp
Neural Network Controller
Nonlinear Process
MRAC Controller
Adjustment Mechanism 
Reference Model ‘1’
Reference Model ‘2’
Reference Model ‘n’ :
Fuzzy Logic Switching
Auxiliary Inputs
Reference Input
Desired Inputs
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

32. Proposed Scheme III (Nonlinear Complex System)
Flight Control System
Sensor Measurements
Pilot Command
Reference
Measurements
Controller Output
(Thtl,Rdr,Elev,Ail)

33. Proposed Scheme III (Nonlinear Complex System)
Neural Network
Flight Pattern
Model 
Adaptive
Control
Adjustment
Mechanism
Fuzzy
Switching 

34. Status: Nonlinear F16 6DOF Model in MATLAB and SIMULINK
Developed the Building Blocks of the Aircraft Model
Developed 6 DOF nonlinear Aircraft Model
Developed Steady State Trim Conditions
Algorithmic Development in MATLAB has completed
Developed Graphical Equivalent in the SIMULINK 
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

35. Status : Scheme 1 and Simulation Results
Problem Formulation has been established
Derived a Stable Model Reference Adaptive Law
Developed a Fuzzy Logic Switching Scheme
Developed a Multiple Reference Model suitable for all ‘modes’
Simulation Results for a Linear ‘Jump’ System
Simulation Results of the Pitch Rate Control of F16 Aircraft 

36. Status: Scheme 2 and Simulation Results
Problem Formulation has been established
Developed a RBFNN Architecture which is dynamic in nature
Derived a Stable Adaptive Law and developed an overall system
Simulation Results to control a Nonlinear Process
Application of the Developed scheme to control F16 aircraft Dynamics is yet to be accomplished 

37. Status : Scheme 3 and Simulation Results
Problem Formulation has been established
Developed an dynamic RBFNN Architecture
Development of a Fuzzy Logic Switching Scheme is yet to be accomplished
Development of a Multiple Reference Model suitable for all ‘modes’ is yet to be done
Integration of Overall Scheme is yet to be done
Application to a Nonlinear Process and F16 Dynamics Control is yet to be done 

38. Proposed Scheme I (Linear Parametric “Jump” System)
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

39. Proposed Scheme I ( Linear Parametric “Jump” System)
Time
T <40
T<70
T<100
Plant Structure
1/(s2+30s-10)
1/(s2+3s-20)
1/(s2+9s-30)
Reference Structure By FLSS
5/(s2+3.51s+1.74)
5/(s2+4.46s+4.11)
5/(s2+7.23s+4.95) I
Time
T <40
T<70
T<100
Plant Structure
1/(s2+9s-30)
1/(s2+30s-10)
1/(s2+3s-30)
Reference Structure By FLSS
5/(s2+7.23s+4.95)
5/(s2+3.51s+1.74)
5/(s2+5.57s+6.32) II
Time
T <40
T<70
T<100
Plant Structure
1/(s2+18s-20)
1/(s2+24s-10)
1/(s2+9s-30)
Reference Structure By FLSS
5/(s2+4.62s+4.93)
5/(s2+3.51s+1.74)
5/(s2+7.23s+4.94)
III
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

40. Proposed Scheme I ( Linear parametric “Jump” System)
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

41. Proposed Scheme I ( Linear parametric “Jump” System)
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

42. Proposed Scheme I ( Pitch Rate Control of F-16 Aircraft)

43. Proposed Scheme I ( Pitch Rate Control of F-16 Aircraft)

44. Proposed Scheme I ( Pitch Rate Control of F-16 Aircraft)

45. Proposed Scheme I ( Pitch Rate Control of F-16 Aircraft)

46. Proposed Scheme II (Nonlinear Functional Uncertain System)
Desired
Position
Actual
Position
Neural Network Inversion
Desired Other States
Single Link Robotic Manipulator with Payload
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

47. Proposed Scheme II (Nonlinear Functional Uncertain System)
Position Trajectory
Time
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

48. Overall Dissertation Status
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach

49. Overall Dissertation Status (Contd.)
A New Generation of Adaptive Control : An Intelligent Supervisory Loop Approach