CMWPE结合SaE-ELM的轮对轴承故障诊断方法

针对DF4型内燃机车轮对轴承不同故障状态的判别问题，提出了一种基于复合多尺度加权排列熵(Composit multiscale weighted permutation entropy, CMWPE)和自适应进化极限学习机(Self-adaptive evolutionary extreme learning machine, SaE-ELM)的机车轮对轴承故障识别方法。CMWPE基于复合粗粒化和加权排列熵的思想，能很好地区分信号的不同模式。SaE-ELM通过自适应进化算法对极限学习机的输入权重、隐含层参数和输出权重进行优化，解决了ELM随机选取网络参数的局限性，提高了网络的泛化性能。计算机车轮对轴承不同健康状态下振动信号的CMWPE,利用SaE-ELM识别轴承所属故障类型及故障程度。在机务段的JL-501轴承检测台上采集了7种不同健康状态的轮对轴承试件的振动信号数据。结果表明：CMWPE特征提取效果优于MPE和MWPE;SaE-ELM模式识别效果优于参数不经优化的ELM。所提方法能够有效诊断机车轮对轴承的不同故障，且故障识别率达到100%。

Fault Diagnosis of Wheelset Bearings Using CMWPE and SaE-ELM

Aiming at the identification of different fault states of DF4 typed diesel locomotives, a fault diagnosis method is proposed by the combination of Composite Multiscale Weighted Permutation Entropy (CMWPE) and Self-Adaptive Evolutionary Extreme Learning Machine (SaE-ELM). CMWPE is based on the idea of composite coarsening and weighted permutation entropy, which can distinguish the different modes of signals very well. SaE-ELM optimizes the input weight, hidden layer parameter and output weight of the extreme learning machine by adaptive evolutionary algorithm, which solves the limitation of ELM random selection of network parameters and improves the generalization performance of the network. The CMWPE of vibration signals of wheelset bearings in different health states is used to identify the type and degree of fault of bearings by SaE-ELM. The vibration signal data of seven wheelset bearing specimens in different health conditions were collected on the JL-501 bearing test bench of the locomotive Depot. The results show that CMWPE feature extraction is better than MPE and MWPE, and SaE-ELM pattern recognition is better than ELM without optimized parameters. The proposed method can effectively diagnose different faults of locomotive wheelset bearings, and the fault recognition rate reaches 100%.