基于LSTM和CRF的加工过程运行状态识别

Process Status Recognition Based on LSTM and CRF

针对加工过程中时序信号截取成本高和运行状态自动识别困难的问题,提出一种结合双向循环神经网络BiLSTM和条件随机场(CRF)的加工过程信号状态识别模型,适用于变参数加工场景。采用LSTM来捕捉时间序列数据的前后依赖性关系,并利用该网络对振动信号进行深层特征提取。在此基础上,为了进一步捕捉状态数据,将LSTM的输出特征输入至条件随机场(CRF)模型和多层感知机(MLP)中进行边界帧分类,进行预测并输出结果标签序列。最后以变参数下的铣削实验为例,验证了基于BiLSTM-CRF的信号状态识别模型在复杂变参数场景下的有效性。

In view of the high cost of timing signal interception and the difficulty of automatic identification of operating status during processing,a sequential signal status recognition model combining bidirectional recurrent neural network BiLSTM(bi-directional long-short term memory)and conditional random field(CRF)was proposed,which was suitable for variable parameter machining scenarios.LSTM was used to capture the dependency relationship of time series data,and the deep feature extraction of vibration signals was carried out by this network.On this basis,in order to further capture status data,the output features of LSTM were input into the conditional random field(CRF)model and multi-layer perceptron(MLP)for boundary frame classification,prediction and output of the result label sequence.Finally,taking the milling experiment under variable parameters as an example,the validity of the signal status recognition model based on BiLSTM-CRF in complex variable parameter scenarios was verified.