Fault detection and diagnosis using hybrid artificial neural network based method

Abstract

This thesis proposes a novel approach to fault detection and diagnosis (FDD) that is focused on artificial neural network (ANN). Unlike traditional methods for FDD, neural networks can take advantage of large amounts of complex process data and extract core features to help detect and diagnose faults. In the first part of this work, a hybrid model was developed to improve efficiency and feasibility of neural networks by combining Kernel Principal Analysis (kPCA) and deep neural network. The hybrid model was successfully validated by Tennessee Eastman Process. The second part of the research focuses on a specific application to gas leak detection and classification. In this scenario, a convolutional network (ConvNet) was used as a feature extraction tool prior to network training due to the visual nature of data. The model was shown to accurately predict leaks and leak sizes; furthermore, further model optimizations were performed and evaluated. The proposed approach is superior to other FDD approaches due to its performance and optimization flexibility.