Abstract: Considering that the performance degradation mode of different tools shows various trends,a single fixed global model is difficult to accurately predict the remaining life of tools with different performance degradation mode.Thus,a tool remaining useful life prediction method based on u-shapelets clustering method and Long Short Term Memory (LSTM) neural network model was proposed.The u-shapelets set was extracted from the tool processing monitoring signals,and the distance between each u-shapelet and the processing monitoring signals was calculated to get the distance matrix.Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering method was used to cluster the distance matrix to obtain the cluster results of process monitoring signals.Different LSTM neural network models were trained to predict the remaining useful life of tools according to different clusters.The validity of the proposed method was verified by processing monitoring signals of wheel groove milling cutter processing,and compared with K-means clustering method,spectral clustering method,hierarchical clustering and DBSCAN clustering method.

Key words: process monitoring data, u-shapelets clustering, clustering algorithm, long short-term memory network, tool remaining useful prediction

摘要： 针对不同刀具的性能衰退规律呈现出多种趋势,单一固定的全局模型难以对不同性能衰退规律的刀具进行准确剩余寿命预测的问题,提出一种基于u-shapelets聚类与长短时记忆网络(LSTM)模型相结合的刀具剩余寿命预测方法。首先,对刀具加工过程监控信号提取u-shapelets集合,并计算各u-shapelet与时间序列的距离得到距离矩阵;其次,通过基于密度聚类方法对距离矩阵进行聚类,得到聚类结果;最后,根据聚类结果基于各类别数据分别训练长短时记忆网络模型进行刀具剩余寿命的预测。以轮槽铣刀加工过程监控数据进行验证,并与K-means聚类、谱聚类、层次聚类、DBSCAN聚类方法进行比较,验证了所提方法的有效性。

关键词: 过程监控数据, u-shapelets聚类, 聚类算法, 长短时记忆网络, 刀具剩余寿命预测