本工作提出了一种基于深度残差网络(DRN)的化工过程故障诊断方法，可从大量化工过程运行数据中自动提取故障特征。模型采用快捷连接缓解传统深度神经网络训练困难的问题，且使用批归一化(BN)方法，可有效缓解梯度消失/爆炸的问题。以田纳西?伊斯曼(TE)过程为实验对象对所提方法进行诊断性能评价实验，并与以往的基于传统深度学习模型的TE过程故障诊断方法进行对比，进一步探究了模型层数、BN技术和残差结构对故障诊断率的影响，最后，通过t分布随机邻域嵌入(t-SNE)方法对网络部分层的输出进行可视化。结果表明，模型对21种工况取得了94%的平均故障诊断率和0.30%的平均假阳率，表现出更加卓越的诊断性能。输出层的二维散点图显示了清晰的聚类，表明所提出的DRN模型能够对故障进行准确诊断。

A fault diagnosis method for chemical processes based on deep residual network (DRN) was proposed, which could automatically extract fault features from a large number of chemical processes operation data. The model adopted the shortcut connections to alleviate the training difficulty in the traditional deep neural network, and adopted the batch normalization (BN) method, which could effectively alleviate the problem of vanishing/exploding gradients. The Tennessee Eastman (TE) process was used as the experimental object to evaluate the diagnostic performance of the proposed method. The proposed method and the previous TE process fault diagnosis method based on traditional deep learning model were compared. Furthermore, the effects of the number of layers, BN technology and residual structure on fault diagnosis rate were studied. Finally, the output of some layers was visualized by the t-distributed stochastic neighbor embedding (t-SNE) method. The results showed that the model achieved an average fault diagnosis rate of 94% and an average false positive rate of 0.30% for 21 working conditions, showing more excellent diagnostic performance. The two-dimensional scatter plot of the output layer showed clear clustering, which indicated that the proposed DRN model can accurately diagnose the faults.