A Feedforward Neural Network Optimized by Fuzzy Particle Swarm for Fault Diagnosis of Hexapod Robots

In fault diagnosis of hexapod robots, a Feedforward Neural Network (FNN) is commonly used to classify various robot states. Traditional optimization methods such as Backpropagation (BP) and Particle Swarm Optimization (PSO) are applied to train the network. However, these methods have limited ability to dynamically adjust the network weights. The adjustment range is narrow and often leads to suboptimal performance. This paper proposes a novel approach called Fuzzy Particle Swarm Optimization-Feedforward Neural Network (FPSO-FNN). It uses fuzzy controllers to dynamically regulate the three key parameters of PSO to optimize the FNN. First, a hexapod robot model is simulated using triangular gait on the CoppeliaSim platform integrated with MATLAB. Faults are then injected into the model to generate a dataset containing fault states. Second, fuzzy controller inputs, rules, and membership functions are defined. These are used to adjust the PSO parameters in a fuzzy logic framework. FPSO selects the best particles based on fitness to optimize the FNN, resulting in the proposed FPSO-FNN algorithm. Finally, the fault dataset is fed into FPSO-FNN, PSO-FNN, and conventional FNN models for comparison. Diagnostic accuracy is evaluated through simulations. The results show that FPSO-FNN achieves an accuracy of 78.54%, which is higher than that of the other two methods. This demonstrates the effectiveness of the proposed approach.