基于EMD和改进TCN的滚动轴承剩余寿命预测方法

Prediction Method of Remaining Life of Rolling BearingBased on EMD and Improved TCN

鉴于当前在滚动轴承的剩余使用寿命预测领域中,轴承振动数据难以提取有效特征、数据维度小而难以满足需求、预测模型趋于复杂化而造成计算成本高的问题,本文提出了一种基于经验模态分解(empirical mode decomposition,EMD)的特征提取方法以及基于改进时间卷积网络(temporal convolutional network,TCN)的剩余寿命预测方法,并在PHM 2012轴承数据集上进行了验证。实验结果显示:改进的时间卷积网络在均方误差(mean square error,MSE)指标上比其他时间卷积网络降低46.43%,在评分函数(Score)指标上比其他时间卷积网络提升4.06%。此外,本文改进的时间卷积网络在MSE上比其他4种模型方法降低84.74%;在Score指标上比其他4种模型方法提升了163%,充分验证了本文改进TCN模型的有效性。

Considering that in the current field of predicting the remaining useful life of rolling bearings,it is difficult to extract effective features from bearing vibration data,the data dimension is small and difficult to meet the demand,and the prediction model tends to become complex,resulting in high computational costs.Therefore,in this article we propose a feature extraction method based on empirical mode decomposition(EMD)and a residual life prediction method based on improved temporal convolutional network(TCN),and also validate it on the PHM 2012 bearing dataset.The experimental results showed that the improved temporal convolution network reduced the mean squared error(MSE)index by 46.43%compared with other temporal convolution networks,and increased the score function index by 4.06%compared with other temporal convolution networks.Moreover,the improved temporal convolutional network in our study reduced the MSE by 84.74%compared to the other four model methods.Compared to its four model methods,the score index increased by 163%.The experimental result fully verifies the effectiveness of improving the TCN model in the article.