基于AC-CNN模型的过程故障识别

针对复杂工业过程中故障变量特征提取效率低,分类数量较少且故障识别率较低等问题,提出基于非对称卷积核(asymmetric convolutions)的卷积神经网络(CNN)的工业过程故障识别模型。采取故障变量重构对故障数据进行预处理;引入非对称卷积核模型对重构后的输入故障变量进行特征提取,提高特征提取的效率;根据CNN模型改进得到具有AC架构的AC-CNN模型,识别TE(田纳西-伊斯曼)过程故障的在线测试集样本,实验结果表明,所提方法对TE过程故障数据集的识别效果明显,验证了模型的有效性和优异性。

Fault recognition based on AC-CNN model

In the complex industrial processes,fault variables feature extraction is low-efficient,the number of classifications is small and the fault recognition rate is low,an industrial process fault identification model based on convolution neural network(CNN)of asymmetric convolution kernels was proposed.The fault variable reconstruction was adopted to preprocess the fault data.The asymmetric convolution kernel model was introduced to extract the features of the reconstructed input fault variables,the efficiency of feature extraction was improved.The TE(Tennessee-Eastman)process fault online test set sample was identified based on the CNN model improved AC-CNN model with AC architecture.Experimental results show that the proposed method has obvious recognition effects on the TE process fault dataset,which demonstrates the effectiveness and excellence of the model.