基于引力搜索RBF神经网络的机车齿轮箱故障诊断

Fault Diagnosis of Locomotive Gearbox Based on Gravitational Search RBF Neural Network

为解决神经网络结构及参数的优化选择问题,以提高机车齿轮箱故障诊断的精度,提出一种基于引力搜索RBF神经网络的机车齿轮箱智能故障诊断方法。基于高斯RBF神经网络建立机车齿轮箱故障诊断模型,采用减聚类算法确定RBF神经网络结构,并结合混沌优化策略及人工蜂群搜索算子提出自适应混合引力搜索算法对故障诊断模型进行优化求解,避免了参数选择的盲目性。采用国际标准测试数据集对该方法进行分类性能测试,结果表明其分类精度明显优于经GA算法、SPSO算法、QPSO算法和GSA算法优化的RBF神经网络。将该方法应用于机车齿轮箱故障的诊断,应用实例验证了该方法的有效性。

In order to solve the issue of the determination of neural network structure and the optimization of neural network parameters to improve the accuracy of fault diagnosis of locomotive gearbox,an intelligent fault diagnosis method based on the gravitational search algorithm and RBF neural network was proposed.When the locomotive gearbox fault diagnosis model was established based on Gaussian RBF neural network,subtractive clustering algorithm was used to determine the structure of RBF neural network.By reference to the artificial bee colony search operator and chaos optimization strategy,an adaptive hybrid gravitational search algorithm was proposed and applied to solve and optimize the fault diagnosis model,to avoid the blindness of parameter selection.Results of the classification performance test on the proposed method using UCI testing data sets showed that the classification accuracy of the proposed method was significantly better than the RBF neural network optimized by GA algorithm,SPSO algorithm,QPSO algorithm and GSA algorithm.The application of the proposed method in fault diagnosis of locomotive gearbox demonstrated the effectiveness of this method.