基于深度CNN-LSTM神经网络的加工过程实时异常监测模型

Real-time Anomaly Monitoring Model in Machine Process Based on Deep CNN-LSTM Neural Network

为了有效兼顾加工过程实时信号的多维特征和序列特征,以实现精度更高的加工状态异常监测,本文提出了一种基于深度卷积-长短时记忆神经网絡(Convolution Neural Network-Long Short Time Memory,CNN-LSTM)的异常监测模型。该模型以数控系统中采集的实时数据为输入,先利用CNN提取其高维特征并生成特征向量,再利用LSTM进行序列特征提取,最后通过逻辑回归得到异常诊断结果。实验表明,该模型的准确率、精确率、召回率等性能指标均超过98.5%,明显优于其他异常监测模型。

In order to extract both muli-dimensional features and sequential features of the real-time signal of the machining process ffectively and achieve higher-precision anomaly monitoring of the machining stale,an anomaly monitoring model based on deep Convolution Neural Network-Long Short Time Memory(CNN-LSTM)is proposed.The model takes real-time data collected in the CNC system as input.Firstly,CNN is used to extract high-dimensional features and then LSTM is used for sequential feature extraction.Finally,the anomaly diagnosis result is obtained through logistic regression.Experiments show that the accuracy.precision,and recall rate of this model are all over 98.5%,which is significantly better than other anomaly monitoring models.