基于PSO和GSA的神经网络轴承故障诊断

Bearing Fault Diagnosis with Optimized Neural Network Based on PSO and GSA

针对原始BP神经网络诊断方法存在初始权值和阈值随机选取而导致识别率低的问题,提出了一种基于粒子群优化算法(PSO)与引力搜索算法(GSA)优化的神经网络诊断方法。上述方法先从原始信号中提取特征向量,再利用PSO的记忆能力和信息共享能力对GSA进行改进,并以此双优化算法来优化BP神经网络的初始权值及阈值,形成一种适用于轴承故障诊断的双优化神经网络模型。实验结果表明,上述方法与原始BP法、GSA-BP法相比,能准确地识别出多种滚动轴承故障,具有比较理想的诊断效果。

Focus on the problem of randomly selection of initial weights and thresholds in the original BP neural network may lead to poor diagnosis results, this paper proposed an approach based on the fusion of Particle Swarm Optimization Algorithm（PSO） and Gravitational Search Algorithm （GSA）. This method firstly extracts the feature vectors from the initial signals, and then introduces the PSO memory ability and information sharing ability into the GSA improvement. The fusion algorithm can optimize the initial weights and thresholds for BP neural network, and a- chieve an improved neural network model for rolling bearing fault diagnosis. The rolling bearing faults experimental results demonstrate that, compared with the original BP or the GSA-BP neural network diagnosis model, the proposed method can obtain more accurate diagnosis results with promising validity and practicability.