Dynamic neural network switching for active control of nonlinear systems.

Area of Science:

• Control Engineering
• Applied Mathematics
• Signal Processing

Background:

• Traditional feedforward active noise and vibration control (ANC/AVC) systems often use linear digital filters, such as the filtered reference least mean squares (LMS) algorithm.
• Linear controllers struggle to accurately model and control systems exhibiting nonlinear dynamics, leading to suboptimal performance.
• Neural network (NN) controllers show promise for nonlinear control but can be computationally intensive, limiting their real-time application.

Purpose of the Study:

• To develop a computationally efficient control strategy for active noise and vibration control in nonlinear systems.
• To maintain high control performance across a range of nonlinear system behaviors.
• To address the computational expense associated with traditional neural network controllers in dynamic environments.

Main Methods:

• A novel control strategy employing dynamically switching neural networks (NNs) is proposed.
• Individual NNs are trained for specific operating ranges of the nonlinear system.
• The system dynamically switches between these smaller, specialized NNs to adapt to varying nonlinearities.

Main Results:

• The proposed switching NN approach demonstrated superior control performance compared to a single, larger generalized NN controller.
• Performance advantage was also observed over a functional link artificial neural network (FLANN) based controller.
• Simulations included a system with nonlinear stiffness and offline tests on a physical nonlinear dynamical system.

Conclusions:

• Dynamically switching between smaller, specialized NNs is an effective strategy for improving active noise and vibration control in nonlinear systems.
• This method offers a practical solution to the computational challenges of NN-based control in nonlinear dynamic environments.
• The proposed approach provides a significant performance advantage over existing NN and linear control techniques for nonlinear applications.