基于鲁棒最小二乘支持向量机的电机振动故障诊断

Motor Vibration Fault Detection Based on Robust LS-SVM

对时序数据建模与辨识技术进行了分析,提出了使用鲁棒最小二乘支持向量机(LS-SVM)算法建立自回归移动平均(ARMA)时序预测模型。该模型是在LS-SVM的约束条件中考虑了鲁棒特性和时序模型参数,使之在求解的过程中对孤立点与噪声不敏感,并且能准确地辨识时序模型参数。考虑到电机振动故障诊断的输入输出数据集间存在着复杂非线性时序上的关系,通过用基于鲁棒LS-SVM的ARMA模型预报电机的振动值,从而预测电机振动故障。实验表明该模型在对非线性时间序列预测精度和稳定性上具有明显的优越性,为确保电机正常运行创造了良好条件。

The time series modeling and identification techniques have been analyzed and the autoregressive moving average （ARMA） time series model based on robust least square support vector machine method （LS-SVM） has been presented. In the model, the robust character and time series model parameters have been considered in constrain condition of LS-SVM. In the process of computation, the presented model is insensitive to the outliers and noises and can identify parameters of time series model accurately. Considering the complex nonlinear relationship in time series between the input and output data sets, the ARMA model based on robust LS-SVM has been used to predict the vibration value of motor in industrial production process and detect the fault of vibration. Finally, the experiments with practical data prove that the presented model has obvious superiority in the precision and robust of nonlinear time series prediction. Thus it provides the good environment to ensure motor normal operation.