Fault Diagnosis Based on Multi-scale LSTM-FCNs for Industrial Process

Aiming at the characteristics of multi-scale, time series and nonlinear of the complex industrial process data, in order to extract process information more efficiently for improving the diagnosibility, a novel fault diagnosis method based on multi-scale long short term memory- fully convolutional networks (LSTM-FCNs) for industrial process is proposed. Firstly, variational mode decomposition (VMD) is used to obtain the multi-scale components of the process data. Then a parallel LSTM and FCNs optimized by batch normalization and random inactivation is built to extract features for each component. Finally, based on the fusion of multi-scale features, a fully connected layer is used for diagnosis. The LSTM-FCNs method is evaluated by the Tennessee Eastman (TE) benchmark industrial process, and compared with other methods, its effectiveness and superiority are verified.