Hierarchical Deep LSTM for Fault Detection and Diagnosis for a Chemical Process

A hierarchical structure based on a Deep LSTM Supervised Autoencoder Neural Network (Deep LSTM-SAE NN) is presented for the detection and classification of faults in industrial plants. The proposed methodology has ability to classify incipient that are difficult detect diagnose with traditional many recent methods. Faults grouped into different subsets according degree difficulty them accurately structure. External pseudo-random binary signals (PRBS) injected system enhance identification faults. approach illustrated benchmark process (Tennessee Eastman Process) order compare across methodologies. efficacy method shown by comprehensive comparison between fault diagnosis methods literature Tennessee Process. work results significant improvements over both multivariate linear model-based strategies non-hierarchical nonlinear strategies.