基于时频分析和迁移学习的齿轮故障诊断方法

Gear Fault Diagnosis Method Based on Time-frequency Analysis and Transfer Learning

针对齿轮故障诊断问题,提出了基于平滑伪Wigner-Ville时频分布和AlexNet迁移学习的故障诊断方法。原始齿轮故障信号是一维振动信号,通过平滑伪Wigner-Ville时频分析+灰度级-RGB变换伪彩色增强变换可以转化为二维时频图像,从而获得比一维振动信号更直观和丰富的特征;进而利用AlexNet神经网络模型进行迁移学习,显著提升开发效率。采集了齿轮系统正常运转、3种单一故障(齿面点蚀、齿面磨损、断齿)和2种复合故障(点蚀+磨损、断齿+磨损)共6种特征的振动信号,构造了齿轮故障数据集,并通过AlexNet迁移学习获得迁移AlexNet齿轮故障诊断模型。研究结果表明,提出的基于时频分析和迁移学习诊断方法对齿轮故障诊断的总体准确率为95.79%,因此可用于工业现场的齿轮系统故障诊断。同时,试验结果表明,齿面磨损单一故障和含有齿面磨损的复合故障造成所提方法的诊断正确率相对较低,说明对于齿面磨损的信号特征的研究和提取方法仍有进一步提升的空间。

Aiming at the problem of gear fault diagnosis,a fault diagnosis method based on smooth pseudo Wigner-Ville time-frequency distribution and AlexNet transfer learning is proposed.The original gear fault signal is a one-dimensional vibration signal,which can be transformed into a two-dimensional time-frequency image through smooth pseudo Wigner-Ville time-frequency analysis and grayscale RGB transformation pseudo color enhancement transformation,thus obtaining more intuitive and rich features than one-dimensional vibration signals.Furthermore,the AlexNet neural network model is used for transfer learning,which significantly improves the development efficiency.The vibration signals with six characteristics including normal operation of gear system,three kinds of single faults(tooth surface pitting,tooth surface wear,tooth breakage)and two kinds of composite faults(pitting&wear,tooth breakage&wear)are collected,and a gear fault data set is constructed,and a transferred AlexNet gear fault diagnosis model is obtained through AlexNet transfer learning.The research shows that the overall accuracy rate of gear fault diagnosis based on time-frequency analysis and transfer learning diagnosis method proposed is 95.79%,so it can be used for fault diagnosis of gear system in industrial field.At the same time,the experimental results indicate that the diagnostic accuracy of the method proposed in this paper is relatively low due to single tooth surface wear faults and composite faults containing tooth surface wear,indicating that there is still room for further improvement in the research and extraction methods of signal features of tooth surface wear.