基于PCA-LSTM的轴承退化趋势预测

由于磨损、腐蚀等因素会导致旋转机械性能的退化甚至失效,若不能及时维修或更换,会影响工程设备的运行,降低设备的工作效率。为了对滚动轴承进行退化趋势预测,避免潜在故障对工程设备造成损害,提高设备工作效率,对滚动轴承振动信号的特征参数使用主成分分析(PCA),对其融合后得出了能充分表现其退化状态的关键特征,降低了预测时数据的输入维度,并使用一种带有随机失活层的长短期记忆(LSTM)神经网络对数据融合后的数据进行预测。将PCA-LSTM模型与非线性自回归(NAR)神经网络的预测结果进行对比,结果表明,基于PCA-LSTM的滚动轴承退化趋势预测模型与NAR神经网络相比,均方根误差和平均绝对百分比误差分别提高了9.1%和8.0163%,预测精度更高,为滚动轴承的退化趋势预测提供了一种新的思路。

Prediction of Bearing Degradation Trend Based on PCA-LSTM

Due to factors such as wear and corrosion,rotating machinery will degrade or even fail.If it cannot be repaired or replaced in time,it will affect the operation of engineering equipment and reduce equipment efficiency.In order to predict the degradation trend of rolling bearings,avoid potential failures to cause damage to engineering equipment and improve equipment work efficiency,principal component analysis(PCA) on the characteristic parameters of rolling bearing vibration signals is used,and after fusion,the key features that can fully express their degradation are obtained,the input dimension of the data during prediction is reduced,and a long short-time memory(LSTM) neural network with a dropout layer is used to predict the data.Comparing the prediction results of the PCA-LSTM model with the nonlinear auto regressive(NAR) neural network,the results show that the root mean square error and the average absolute percentage error of the PCA-LSTM based rolling bearing degradation trend prediction model are 9.1% and 8.0163% higher than those of NAR neural network,respectively,and the prediction accuracy is higher,which provides a new idea for the prediction of the degradation trend of rolling bearings.