A simplified adaptive neural network prescribed performance controller for uncertain MIMO feedback linearizable...

Area of Science:

• Control Theory
• Nonlinear Systems
• Artificial Intelligence in Control

Background:

• Designing controllers for multi-input, multi-output (MIMO) feedback linearizable systems presents significant complexity.
• Current methods often involve intricate designs and can suffer from issues like model controllability loss due to approximations.

Purpose of the Study:

• To develop a simplified, continuous state-feedback controller for MIMO feedback linearizable systems.
• To reduce design complexity and improve performance bounds on tracking errors.
• To address and avoid the model controllability loss problem inherent in approximation singularities.

Main Methods:

• A novel control scheme employing a single neural network to approximate a scalar function representing system nonlinearities.
• Development of an adaptive law that avoids additional complexity.
• Simulation-based verification of the proposed controller's efficacy.

Main Results:

• The proposed scheme successfully achieves prescribed bounds on transient and steady-state output tracking errors.
• Controller simplification and reduced design complexity compared to existing methods.
• Avoidance of model controllability loss without added complexity to the control or adaptive law.

Conclusions:

• The single neural network approach offers a more streamlined and effective method for controlling complex nonlinear MIMO systems.
• The findings demonstrate a significant advancement in reducing the complexity of feedback linearizable system control.
• The proposed controller maintains robust performance despite system uncertainties and avoids common pitfalls in approximation-based control.