基于深度学习理论的刀具状态监测及剩余寿命预测方法

Deep learning-based method for tool condition monitoring and remaining useful life prediction

针对机械大数据的特点及深度学习的优势,提出了一种新的刀具磨损状态监测及剩余使用寿命预测方法。该方法首先利用稀疏自编码器及皮尔逊相关系数对原始切削力信号自适应提取敏感特征;之后将得到的显著性特征与其对应的刀具磨损值训练反向传播(BP)神经网络;最后使用预测的刀具磨损值作为观测值,利用指数平滑算法预测刀具剩余使用寿命。为了解决样本数量不足带来的过拟合问题,对原始样本进行加噪处理,同时在特征提取过程中引入dropout训练技巧。通过刀具全寿命周期实验实现了大量样本下刀具磨损特征自适应提取与剩余寿命预测,证明了所提方法的有效性。

A new method for tool condition monitoring and remaining useful life prediction is proposed with harnessing the properties of machinery big data and the advantages of deep learning theory. First, the sensitive features of the original cutting force signal are extracted by sparse auto-encoder and Pearson correlation coefficient;Then using the obtained features to train back propagation ( BP) neural network. Finally, the exponential smoothing prediction method uses the predicted tool wear value to predict the tool remaining useful life. In order to solve the problem caused by the insufficient number of samples, the noiseis addedto original samples. Meanwhile, dropout method is introduced in the process of feature extraction. The tool life experiment shows that the proposed method is effective, which is able to not only adaptively extract sensitive features, but also predict the tool remaining useful life.