基于谱熵梅尔积和改进VMD的轴承故障预警

针对传统轴承故障预警实时性较差、故障特征提取准确性影响预警效果的问题，将语音端点识别思想进行迁移，采用谱熵梅尔积特征的双门限法实时追踪故障起始点.为克服变分模态分解（variational mode decomposition，VMD）参数选取不当和端点效应对提取效果造成的影响，提出能量差网格搜索法对VMD进行参数寻优，并用支持向量回归机对端点效应进行抑制，结合多尺度加权排列熵在检测振动信号随机性方面的优势，充分发挥VMD对信号的重构能力，对起始点后的故障段进行特征捕捉.通过实际轴承故障信号的实验及数据分析，验证了该方法在轴承故障预警中的有效性. 

Bearing Fault Warning Based on MFPH and Improved VMD

To solve the problem of bad real-time performance of traditional bearing fault early warning and the accuracy affection of fault feature extraction on the early warning effect, transferring the idea of speech endpoint recognition, a double threshold method was used based on the MFPH feature to track the fault starting point. Firstly, in order to overcome the influence of parameter selection and endpoint effect of variational mode decomposition (VMD)on the feature extraction, based on the grid search method of energy difference, the parameters were optimized, and the breakpoint effect was suppressed by SVR. Then, combined with the advantages of MWPE in detecting the randomness of vibration signals, the ability of VMD to reconstruct the signals was fully utilized and the fault signal after the starting point was extracted. Finally, the effectiveness of this method in bearing fault warning was demonstrated by the experiment of bearing fault signal.