An artificial intelligence approach to fault isolation based on sensor data in Tennessee Eastman process

An effective fault diagnosis scheme can improve system's safety and reliability. Artificial Intelligence (AI) provides a good framework to deal with this issue. Deep learning is a successful implementation of AI that its superior isolation performance find its way in fault diagnosis area. In this study, based on feature extraction abilities of Convolutional Neural Network (CNN), a deep network have been developed in order to isolate different kinds of faults in Tennessee Eastman process. This network has an end-to-end structure with 13 layers that takes raw sensor's data and has isolation performance of more than 98 percent. A comparison between our proposed method and a linear classifier that uses Principal Component Analysis(PCA) for feature extraction and a Neural Network (NN) with 2 hidden layers as nonlinear classifier have been conducted to show the performance of the proposed fault isolation scheme.