基于改进Holt双指数模型的轴承剩余寿命预测方法

Prediction Method of Bearing Residual Life Based on Improved Holt Double Exponential Model

针对Holt双指数模型参数难以确定导致预测精度较低的问题,提出一种基于深度残差收缩网络(Deep Residual Shrinkage Network, DRSN)和Holt双指数模型的轴承剩余寿命预测方法。首先,无需任何人为经验的干扰,利用DRSN模型对轴承原始信号进行特征自提取,挖掘退化特征并构建健康指标;然后,利用Holt双指数模型构建剩余寿命预测模型,针对Holt双指数预测模型超参数α,β难以确定的问题,采用麻雀搜索算法对两个参数进行优化。将DRSN提取出的健康指标进行平滑处理后,利用优化后的Holt双指数预测模型进行预测,完成轴承的剩余寿命预测。利用轴承全寿命实验数据对所提方法进行了验证,实验结果表明,本方法预测误差比双指数预测模型低12%,有效提高了轴承剩余寿命预测精度。

Aiming at the problem that the parameters of Holt double-exponential model are difficult to determine, resulting in low prediction accuracy, a bearing residual life prediction method based on deep residual shrinkage network(DRSN) and Holt double-exponential model is proposed. First, without any interference of human experience, the DRSN model is used to extract the features of the original bearing signals, mine the degraded features and build the health indicators;Then, the remaining life prediction model is built by using Holt double exponential model. Aiming at the problem that the Holt double exponential prediction model is difficult to determine, the sparrow search algorithm is used to optimize the two parameters. After smoothing the health indicators extracted from DRSN, the optimized Holt double-index prediction model is used to predict the remaining life of the bearing. The proposed method is verified by the experimental data of bearing life. The experimental results show that the prediction error of this method is 12% lower than that of the double-exponential prediction model, and the prediction accuracy of bearing remaining life is effectively improved.