Fast real-time SDRE controllers using neural networks.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Aerospace Engineering

Background:

• State-dependent Riccati equation (SDRE) control is a powerful technique for nonlinear systems.
• Implementing SDRE controllers can be computationally intensive, limiting their application in high-speed scenarios.
• Artificial neural networks (ANNs) offer potential for approximating complex control functions.

Purpose of the Study:

• To implement fast state-dependent Riccati equation (SDRE) control algorithms using artificial neural networks (ANNs).
• To evaluate the efficacy of ANNs in replicating SDRE controllers for satellite attitude dynamics.
• To demonstrate the potential for reduced computational complexity and higher execution rates with neural network controllers.

Main Methods:

• Training shallow and deep artificial neural networks (ANNs) to emulate a pre-designed SDRE controller.
• Utilizing a satellite attitude dynamics simulator (SADS) for developing and testing the SDRE controller.
• Validating a trained neural network controller through practical experimentation on the SADS.

Main Results:

• Trained ANNs successfully replicated the functionality of the original SDRE controller.
• Neural network controllers exhibited significantly reduced computational complexity.
• Experimental validation confirmed that the neural controller performed equivalently to the original SDRE controller with small training errors.

Conclusions:

• Fast SDRE control algorithms can be effectively implemented using ANNs.
• ANN-based controllers offer a computationally efficient alternative to traditional SDRE methods for satellite attitude control.
• The proposed technique demonstrates practical viability for high-rate control applications.