Deep Learning Based Fault Diagnosis for Chemical Process with Statistical Feature Fusion

The existing fault diagnosis methods based on deep neural networks mostly rely on process data for fault classification and diagnosis, which makes it difficult to distinguish minor faults with similar data features. Therefore, this paper proposes a deep learning-based fault diagnosis model that combines statistical indicators and neural networks. By analyzing and calculating the relevant information entropy to screen the statistical signal features of time domain and frequency domain, we can extract more hidden information from the raw data, and use the convolutional autoencoder to compress the statistical indicators for feature extraction, and use CNN network to extract features from raw data. Finally, the two sets of features are fused and input into the LSTM network to obtain temporal information for fault diagnosis. Tests conducted on the Tennessee Eastman (TE) process show that the proposed model outperforms typical classifiers and improves the diagnostic accuracy of minor faults.