Self-organizing CMAC control for a class of MIMO uncertain nonlinear systems

Area of Science:

• Robotics and Control Engineering
• Artificial Intelligence
• Nonlinear System Dynamics

Background:

• Multiple-input-multiple-output (MIMO) uncertain nonlinear systems pose significant control challenges.
• Existing control methods often struggle with adaptability and real-time parameter tuning.
• Cerebellar Model Articulation Controller (CMAC) offers adaptive learning capabilities but requires careful structural design.

Purpose of the Study:

• To develop a self-organizing control system for MIMO uncertain nonlinear systems.
• To enhance the adaptability and robustness of CMAC-based controllers.
• To simplify CMAC input space and enable automatic structural adjustments.

Main Methods:

• Integration of CMAC with Sliding-Mode Control (SMC) to simplify CMAC input dimensions.
• Development of a Self-Organizing CMAC (SOCM) with automatic layer growth/pruning and receptive function adjustment.
• Utilization of a CMAC uncertainty observer within the SOCM and a robust controller to mitigate approximation errors.
• Application of gradient-descent for online parameter tuning and Lyapunov functions for stability guarantees.

Main Results:

• The proposed control system demonstrated favorable tracking performance in simulations of an inverted double pendulum system.
• Experimental results with a linear ultrasonic motor confirmed the system's effectiveness in motion control.
• The self-organizing nature of the CMAC allowed for systematic structural adaptation.

Conclusions:

• The proposed self-organizing CMAC-based control system effectively manages MIMO uncertain nonlinear systems.
• The combined CMAC and SMC approach offers a robust and adaptive control solution.
• The system's ability to self-organize and adapt ensures favorable tracking performance and stability.