Adaptive feedback control by constrained approximate dynamic programming.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Aerospace Engineering

Background:

• Adaptive control systems are crucial for maintaining performance under varying conditions.
• Neural networks (NNs) offer powerful function approximation capabilities for complex control problems.
• Ensuring closed-loop stability and performance guarantees remains a challenge in adaptive NN control.

Purpose of the Study:

• To develop a constrained approximate dynamic programming (ADP) approach for designing adaptive neural network (NN) controllers.
• To guarantee closed-loop stability and performance for adaptive NN controllers.
• To ensure baseline performance preservation in the presence of unmodeled dynamics or failures.

Main Methods:

• Utilizing prior knowledge of linearized equations of motion to establish performance and stability objectives within a linear parameter-varying (LPV) regime.
• Implementing an adaptive NN controller that optimizes performance online when facing unmodeled dynamics or failures.
• Applying constrained ADP to guarantee the preservation of LPV baseline performance at all times.

Main Results:

• Demonstrated the effectiveness of the adaptive NN flight controller in simulated scenarios.
• Successfully managed control failures, parameter variations, and near-stall dynamics.
• Validated the closed-loop stability and performance guarantees of the proposed constrained ADP approach.

Conclusions:

• The constrained ADP approach effectively designs adaptive NN controllers with guaranteed stability and performance.
• The method ensures robust control by preserving baseline performance despite system uncertainties.
• The adaptive NN flight controller shows significant promise for real-world aerospace applications.