Adaptive Unknown Input Estimation by Sliding Modes and Differential Neural Network Observer.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Nonlinear Dynamics

Background:

• Estimating dynamics and unknown inputs in perturbed uncertain nonlinear systems is challenging.
• Existing methods often require full state access or exact system models.
• Robust observers and differentiators are crucial for system analysis and control.

Purpose of the Study:

• To propose a robust observer based on a differential neural network (DNN) for estimating dynamics of uncertain nonlinear systems.
• To reconstruct exogenous unknown inputs using a second-order sliding mode supertwisting algorithm.
• To demonstrate the applicability of the method for both identification (full state access) and estimation (partial state access) problems.

Main Methods:

• A differential neural network (DNN) is employed as a robust observer to estimate system dynamics.
• An identification error convergence is achieved in the first stage.
• A second-order sliding mode supertwisting algorithm acts as a robust exact differentiator to reconstruct unknown inputs.

Main Results:

• The DNN observer successfully estimates system dynamics and reconstructs unknown inputs.
• The approach is validated for both full and partial state vector access scenarios.
• Numerical examples with a spatial minisatellite and a flexible robot manipulator demonstrate high accuracy, comparable to methods with complete plant knowledge.

Conclusions:

• The proposed DNN-based observer and differentiator method effectively handles perturbed uncertain nonlinear systems.
• It offers a robust solution for estimating system dynamics and unknown inputs, even with partial state information.
• The technique shows significant promise for applications in robotics and aerospace systems.