Adaptive Neural Network-Based Control and Fault Detection for Soft Bending Pneumatic Actuators

This paper outlines the development and implementation of a neural network-based controller for a soft bending pneumatic actuator, incorporating an actuator fault detection and estimation technique utilizing neural networks. Soft robot manipulators, known for their flexibility and compliance with their surrounding environment, are increasingly utilized in applications requiring gentle and precise movements, such as rehabilitation. However, their nonlinear dynamics pose significant challenges for classic control methods. To overcome these challenges, we propose an adaptive neural network-based control strategy that leverages the learning capabilities of neural networks to model and predict the actuator's behavior with minimal error. This network controls the curvature angle. Furthermore, we introduce a real-time fault detection and estimation mechanism using neural networks to diagnose potential faults. This enhances the reliability of the system through identifying abnormalities and enabling countermeasures. Simulation results demonstrate the effectiveness of the proposed control and fault estimation methods. The neural network-based controller achieves high precision in trajectory tracking, while the actuator fault estimation system accurately identifies and diagnoses faults, ensuring robust and reliable operation.