改进多尺度幅值感知排列熵与随机森林结合的滚动轴承故障诊断

Rolling bearing fault diagnosis by improved multiscale amplitude-aware permutation entropy and random forest

针对滚动轴承故障识别准确率较低的问题,本文提出了一种新型滚动轴承故障诊断方法。该方法能够在准确识别滚动轴承故障类型的基础上,进一步分析故障的严重程度。首先,通过固有时间尺度分解提取滚动轴承振动信号的最佳固有旋转分量,突显故障信号的冲击特征;然后,利用改进多尺度幅值感知排列熵对信号幅值和频率变化敏感的特性,计算不同时间尺度下的幅值感知排列熵作为故障特征向量,改善了多尺度分析中的粗粒化过程,提升了故障特征提取的稳定性;最后,利用故障特征集构建随机森林多分类器,实现对滚动轴承不同故障类型的识别及严重程度分析,具有较强的泛化能力。实验结果表明,与现有滚动轴承故障诊断方法相比,平均故障识别准确率达到99.25%。该方法能够稳定而有效地提取滚动轴承的故障特征且具有较好的实时性。

Aiming at the problem of low fault identification accuracy of rolling bearing, a novel fault diagnosis method of rolling bearing was proposed. This method can identify fault types of rolling bearing accurately and analyze the fault severity. Firstly, the optimum proper rotation(PR) component of rolling bearing vibration signal was extracted by intrinsic time-scale decomposition(ITD) to highlight the impact characteristics of fault signal. Then, using the characteristic that improved multi-scale amplitude-aware permutation entropy(IMAAPE) was sensitive to signal amplitude and frequency changes, the AAPE values in different time scales were calculated as the fault feature vector, which improved the coarse-grained process in multi-scale analysis and increased the stability of fault feature extraction. Finally, the random forest multi-classifier was constructed by using the fault feature set to realize the fault type identification and severity analysis of rolling bearing, which had a strong generalization ability. Experimental results show that compared with the existing fault diagnosis methods of rolling bearing, average fault identification accuracy is 99.25%. This method can extract the fault characteristics of rolling bearing stably and effectively with good real-time performance.