基于t-SNE和LSTM的旋转机械剩余寿命预测

针对旋转机械的剩余使用寿命预测问题,提出了一种基于t分布随机近邻嵌入(t-SNE)和长短期记忆网络(LSTM)的预测方法。将t-SNE降维方法引入旋转机械振动信号特征提取,实例验证表明无论针对时频域特征或小波包分解得到的能量特征,经t-SNE降维后特征区分度更加明显,利用降维后的特征进行故障模式识别,正确率接近100%;提出利用样本间散度作为旋转机械退化指标,实验表明样本间散度对旋转机械性能退化趋势的表现相比其他指标更加明显;以不同的训练样本量,利用LSTM方法进行剩余使用寿命预测,为了验证LSTM方法的有效性,将其与BP神经网络、灰色预测模型、支持向量机等方法进行比较,结果表明LSTM方法能够预测旋转机械退化趋势,显著提高剩余使用寿命的预测精度,对旋转机械的健康监测和寿命预测具有一定的理论指导意义。

Prediction of remaining useful life based on t-SNE and LSTM for rotating machinery

Aiming at the problem of remaining useful life prediction of rotating machinery, a prediction method based on t-Distributed Stochastic Neighbor embedding (t-SNE) and Long Short-Term Memory network (LSTM) was proposed. First of all, the t-SNE dimensionality reduction method was introduced into the feature extraction of rotating machinery vibration signals, and the example verifies that no matter for the  time-frequency domain features or energy features obtained by wavelet packet decomposition, the feature differentiation is more obvious after t-SNE dimensionality reduction, and the correct rate of fault mode recognition using the dimensionality reduction features is close to 100%. Secondly, it was proposed to use the divergence between samples as the degradation index of rotating machinery. the experimental results show that the divergence between samples has a more obvious performance on the performance degradation trend of rotating machinery than other indexes. Finally, the LSTM method was used to predict the remaining useful life with different training sample sizes. In order to verify the effectiveness of the LSTM method, it was compared with the BP neural network, grey prediction model, support vector machine and other methods. The results show that the LSTM method can predict the degradation trend of rotating machinery and significantly improve the prediction accuracy of the remaining useful life. It has a certain theoretical guiding significance for the health monitoring and life prediction of rotating machinery.