A heuristic fault diagnosis approach for electro-hydraulic control system based on hybrid particle swarm optimization and Levenberg–Marquardt algorithm

In this paper, a novel heuristic neural network model based on hybrid particle swarm optimization and Levenberg–Marquardt (HHPSOLM) algorithm was proposed for fault diagnosis in electro-hydraulic control system. In this algorithm, the characteristics of strong local searching capability in LM algorithm were adopted to increase the diversity of population. In the first stage, the proposed method searched ten steps with the Levenberg–Marquardt (LM) algorithm for a random particle, and replaced the worst particle with the search result to increase the diversity. In the second stage, the HHPSOLM algorithm employed an inspiration to search the optimal solution with the LM algorithm for 30% of the particles to improve diversity of the particles. In the last stage, the feed-forward neural networks were trained with HHPSOLM to achieve the optimization of its weights and thresholds, then the fault diagnosis model of an electro-hydraulic control system was established with the HHPSOLM neural network. Its application in electro-hydraulic control system fault diagnosis was simulated. Experiment results showed that the diagnostic accuracy of the proposed method was higher than those of the Particle Swarm Optimized BP neural network (PSO-BP) or BP algorithm, thus the HHPSOLM algorithm was an efficient algorithm for optimizing neural networks, and suitable for fault recognition of electro-hydraulic control system.