基于长短时记忆循环网络的塑料编织机故障诊断研究

Research on Rault Diagnosis of Plastic Braiding Machine Based on Long-Short Memory Convolutional Neural Network

长短时记忆(LSTM)循环神经网络的塑料编织机故障诊断法通过提取振动信号的能量矩,突出信号在时间轴上的分布特征,降低输入模型的向量维度。从多个特征向量构成的样本集中选择80%作为训练样本,训练LSTM循环神经网络模型,并利用剩余样本验证模型的检测精度;以准确率、查准率和查全率作为评价指标,利用多组不同的振动数据样本,对BP神经网络模型、卷积神经网络(CNN)模型和LSTM循环神经网络模型进行比较分析。结果表明:LSTM循环神经网络模型在不同样本中能够同时达到较高的准确率、查准率和查全率,其平均值分别可达95.69%、86.96%、96.89%,证明LSTM循环神经网络能充分学习具有时序特性的故障信息,对塑料编织机的故障诊断具有可行性和有效性。

The fault diagnosis method of plastic knitting machine based on long short-term memory(LSTM)cyclic neural network is to extract the energy moment of the vibration signal,highlight the distribution characteristics of the signal on the time axis,and reduce the vector dimension of the input model.A sample composed of multiple feature vectors chooses 80%as the training sample,trains the LSTM recurrent neural network model,and uses the remaining samples to verify the detection accuracy of the model.With accuracy,precision and recall as evaluation indicators,and multiple sets of different vibration data samples to BP neural network model,convolutional neural network(CNN)model and LSTM recurrent neural network model are compared and analyzed.The results show that the LSTM recurrent neural network model can achieve high accuracy,precision and recall in different samples at the same time,and the average value can reach 95.69%,86.96%and 96.89%,which proves that the LSTM recurrent neural network can fully learn the fault information with sequential characteristics,which is feasible and effective for fault diagnosis of plastic weaving machine.