Unsupervised Tool Anomaly Detection Based on AGRU Autoencoder
At present,most processing enterprises still rely on manual experience or regular shutdown inspections to monitor CNC machine tools,which not only reduces production efficiency,but also leads to obvious data imbalances in the tool processing.Therefore,a gated recurrent unit(GRU)autoencoder model that integrating attention mechanism was proposed for tool anomaly detection.In this model,the encoders and decoders were set up through the gated loop unit to extract deep features of the time series data.The attention mechanisms was integrated into the encoder reconstruction part to select key features,so as to improve model efficiency.In addition,an anomaly detection model that combined long-term evaluation window mechanism was proposed to further enhance detection capability and stability.Finally,the effectiveness and feasibility of the proposed method were proved by experiments on the data set obtained from the experiment and the open data set.The results show that the accuracy of the proposed method is more than 98％on different data sets.Compared with other methods in the field of tool condition monitoring,this method can obtain tool life cycle data and wear label data without a lot of experiments,which is convenient for the development and application of tool detection systems.
NETL
NSTL
万方数据
