张量奇异谱分解与极限学习机的故障诊断方法

Fault Diagnosis of Rolling Bearing Based on Tensor Singular Spectrum Analysis and Extreme Learning Machine

针对滚动轴承故障诊断问题,提出张量奇异谱分解(TSSA)与极限学习机(ELM)相结合的诊断方法。TSSA将一维时域振动信号转换成三阶张量,使用标准张量分解对三阶张量进行分解并重构回一维时域振动信号;为了验证TSSA的有效性,将奇异谱分解作为对比方法,仿真结果表明:TSSA重构后的信号能够找到故障特征倍频,其效果优于奇异谱分解。从重构时域信号中提取时域特征参量,并使用ELM网络对其实施诊断;为验证ELM的有效性,将BP、SVM作为对比算法,诊断结果表明:从诊断准确率、样本比例、诊断时间方面而言,ELM的性能比BP、SVM要好,ELM更适宜于轴承故障诊断。

In this paper,in order to solve the problem of rolling bearing fault diagnosis,a diagnostic method based on tensor singular spectrum analysis(TSSA)and extreme learning machine(ELM)is proposed.The one-dimensional time-domain vibration signal is converted into three tensor by TSSA.The standard tensor decomposition is used to decompose third-order tensor and reconstructs one-dimensional time-domain vibration signal.In order to verify the effectiveness of TSSA,the singular spectrum analysis(SSA)is used as a comparison method.The simulation results show that fault characteristic frequency doubling can be fund from the reconstructed signal of TSSA,and its effect is better than that of SSA.The time-domain characteristic parameters can be extracted from reconstructed vibrational signals,and ELM is used fault diagnosis.In order to verify the effectiveness of ELM,BP and SVM are used as a comparison method.The diagnostic results show that ELM is better than BP and SVM in terms of diagnostic accuracy,sample proportion and diagnostic time.In contrast,ELM is more suitable for bearing fault diagnosis.