Multichannel one-dimensional convolutional neural network-based feature learning for fault diagnosis of industrial processes

In industrial processes, the noise and high dimension of process signals usually affect the performance of those methods in fault detection and diagnosis. A predominant property of a fault diagnosis model is to extract effective features from process signals. Wavelet transform is capable of extracting multiscale information that provides effective fault features in time and frequency domain of process signals. In this paper, a new deep neural network (DNN), multichannel one-dimensional convolutional neural network (MC1-DCNN), is proposed to investigate feature learning from high-dimensional process signals. Wavelet transform is used to extract multiscale components with fault features from process signals. MC1-DCNN is able to learn discriminative time-frequency features from these multiscale process signals. Tennessee Eastman process and fed-batch fermentation penicillin process are adopted to verify performance of the proposed method. The experimental results demonstrate remarkable feature extraction and fault diagnosis performance of MC1-DCNN and show prosperous possibility of applying this method to industrial processes.

Automated summary

This study introduces a new deep neural network model MC1-DCNN, which learns fault features from high-dimensional process signals using wavelet transform, achieving remarkable feature extraction and fault diagnosis performance.