基于MGWO-SCN的滚动轴承故障诊断方法

Rolling bearing fault diaqnosis method based on MGWO-SCN

为了提高滚动轴承故障诊断模型的鲁棒性和泛化能力,提出了一种基于改进灰狼算法优化随机配置网络(MGWO-SCN)的滚动轴承故障诊断模型。首先,在随机配置网络(SCN)中引入L2范数惩罚项,提高了SCN在实际应用中的泛化能力;然后,在灰狼算法(GWO)中融入差分进化机制,构建了改进灰狼算法(MGWO),并用其对SCN的惩罚项系数C进行了优化;最后,通过分析美国凯斯西储大学(CWRU)轴承振动信号数据集的频域特征信息,构造了基于频域特征参量的振动数据集;并分别用BP神经网络(BPNN)、极限学习机(ELM)和支持向量机(SVM)诊断模型,以及MGWO和粒子群优化算法(PSO)对所提模型进行了对比仿真测试。研究结果表明:在30次重复实验中,采用基于改进灰狼算法优化随机配置网络(MGWO-SCN)的方法,可以准确地识别出12种轴承运行状态,相比于BPNN、ELM和SVM轴承诊断方法,该方法的诊断平均准确率分别提高了7.27%、6.47%和8.67%;另外,MGWO-SCN在优化故障诊断模型方面具有更强的全局搜索能力,相比于GWO-SCN和PSO-SCN,该模型预测结果的偏差值更小,测试集准确率更高。

In order to improve the robustness and generalization ability of the fault diagnosis model of rolling bearing,a rolling bearing fault diagnosis model based on the modified gray wolf algorithm to optimize stochastic configuration networks(MGWO-SCN)was proposed.Firstly,in order to improve the generalization ability of SCN in practical applications,the L2 norm penalty term in the SCN was introduced.Then,the differential evolution mechanism was integrated into the gray wolf algorithm(GWO),and the modified gray wolf algorithm(MGWO)was constructed,which was used to optimize the penalty coefficient C of SCN.Finally,by analyzing the frequency domain characteristic information of the bearing vibration signal data set of Case Western Reserve University(CWRU),a vibration data set based on the frequency domain characteristic parameters was constructed,and BP neural network(BPNN),extreme learning machine(ELM)and support vector machine(SVM)diagnostic model,as well as MGWO and particle swarm optimization algorithm(PSO)were used to compare and simulate the model.The experimental results show that the method can accurately identify 12 kinds of bearing operating states in 30 repeated experiments.Comparing with the BPNN,ELM and SVM bearing diagnosis methods,the average accuracy is respectively increased by 7.27%,6.47%and 8.67%.In addition,MGWO has stronger global search ability in optimizing the proposed model,comparing with GWO-SCN and PSO-SCN,the deviation value of the prediction results of the proposed model is smaller,and the accuracy of the test set is higher.