Study on shearer fault diagnosis and classification methods based on hybrid of PSO and BP neural network
Zhao Shuanfeng · Mining & Processing Equipment · 2011
The essence of fault diagnosis is to extract and classify single features,and BP neural network is a typical classification method.As the traditional BP algorithm easily forms local minimum and lacks for global search trait,whereas PSO(particle swarm optimization) achieves rapid and efficient global search among intricate,multi-peaked,non-linear and non-differentiable space,a new method of training neural network based on hybrid of PSO and BP neural network is proposed in combination with the traditional BP algorithm.The new algorithm at first uses the global search ability of PSO to optimize weights of BP neural network.Meantime,the concept of particle swarm entropy is introduced to judge the individual diversity of particle swarm.As the estimated value of particle swarm entropy exceeds a set threshold,BP algorithm is used to train the neural network.Finally,fault data of shearer bearings are taken to verify the algorithm,and the results show that the algorithm is able to diagnose the shearer faults.