基于图像识别的机械振动信号特征提取与寿命预测方法研究

滚动轴承作为许多机械设备的关键组件,被广泛应用于机械制造、航空航天等领域,其健康状态直接影响了相应设备的剩余寿命,因此在设备故障预测与健康管理(Prognostics and Health Management, PHM)领域,滚动轴承寿命预测具有很高的研究价值。目前基于数据驱动的轴承寿命预测方法主要利用特征提取并构造健康因子(Health Indicator, HI),然而在这一过程中特征的选择与融合依然依赖于专家先验知识,并且健康因子也很难从复杂的时序数据中进行提取。因此,提出了一种新型的数据驱动寿命预测算法,在特征提取方面,通过连续小波变换(Continuous Wavelet Transform,CWT)将传感器振动信号转换为时频谱图,再通过深度残差网络(Deep residual network, ResNet)结合时空卷积网络(Temporal Convolutional Network, TCN)将时频谱图中的时域频域特征构造成为健康因子,最后完成剩余寿命预测。本研究在PRONOSTIA数据集上与现有的数据驱动算法进行了对比,证明了该算法可以更准确地完成剩余寿命预测。

Research on Feature Extraction and Remaining Useful Life Prediction for Mechanical Vibration Signal Based on Image Recognition

As a key component of modern manufacturing industry, rolling bearings directly affect the remaining useful life (RUL) of corresponding mechanical equipment. Consequently, in the field of Prognostics and Health Management (PHM), the prediction of rolling bearings has become an increasingly crucial research area. Currently, the data-driven methods of bearing RUL mainly focus on extracting features and constructing health indicator (HI). However, the process of selecting and fusing features still relies on the prior knowledge of experts, and the health factors are difficult to extract from complex time series. Therefore, this research proposes a new data-driven remaining useful life prediction algorithm. In terms of feature extraction, the sensor vibration signal is converted into a time-frequency image by continuous wavelet transform (CWT). After that, a Temporal Convolutional Network (TCN) is adopted to construct health factors. We evaluate the proposed method with the existing data-driven methods on a real bearing dataset provided by PRONOSTIA, and the results shows that our method is able to achieve higher accuracy without intensive labor.