A Fault Diagnosis Method using Convolutional Neural Network with Visualized Samples

Convolutional neural network (CNN) is a prominent machine vision modelling method for extracting useful image features. In order to overcome the limitation of CNN used in industrial processes, an image conversion strategy is utilized in this work. Firstly, the scaled data is assigned into different numerical segments by setting the data bounds and the division rate. Subsequently, the RGB image is converted by using the mapping relation from the previously divided segment to the graphical dictionary. Then the converted RGB image of the process measurements can be used as the input of CNN. Application to the benchmark Tennessee Eastman (TE) process validates the performance of the proposed method.