基于卷积降噪自编码器和CNN的滚动轴承故障诊断

Fault Diagnosis of Rolling Bearing Based on Convolutional Denoising Auto-Encoder and CNN

针对滚动轴承故障诊断中,振动信号往往具有非线性与非平稳性、且振动信号一般含有噪声的问题,提出一种卷积降噪自编码器和深度卷积网络(CNN)相结合的滚动轴承故障诊断方法。首先将滚动轴承时域信号输入到卷积降噪自编码器中,用无监督方式训练并在原始数据中加入噪声,提取自编码器隐含层数据作为特征实现降噪与降维,接着将提取的特征输入到CNN中进行模式识别。使用时域信号直接输入到CNN、人工特征输入到CNN两种方法作为对比,并对三种方法提取的特征进行主成分分析(PCA)。实验结果表明,基于卷积降噪自编码器和CNN的诊断方法在滚动轴承故障诊断中正确率较高、时间复杂度较低,验证了该方法的优越性。

In rolling bearing fault diagnosis, the fault signal is often non-stationary and contains noise. In recent years, convolutional neural network (CNN) has a high time complexity in processing large data. To solve these problems, a fault diagnosis method based on convolutional denoising auto-encoder and CNN is proposed. Firstly, the time-domain signals are input into the convolutional denoising auto-encoder, which is trained unsupervised and added noise to the original data. The hidden layer data from the encoder is extracted as features to achieve noise reduction and dimensionality reduction, and then the extracted features are input into CNN for pattern recognition. The time-domain signals are directly input to CNN and the artificial features are input to CNN as a comparison, and the principal component analysis (PCA) of the features extracted by the three methods is carried out. The experimental results show that the proposed method has higher accuracy and lower time complexity, which verifies the superiority of the method.