Canonical Correlation Analysis aided Dynamic Graph Autoencoder Method for Industrial Process Monitoring

Industrial processes monitoring characterized by nonlinear and dynamic behaviors presents significant challenges. Recently, autoencoders (AEs) and their advanced variants have been proposed for nonlinear process monitoring. Canonical correlation analysis (CCA)-based methods have also emerged, leveraging correlations between process inputs and outputs to streamline fault detection design. To address these limitations, we propose a novel approach that combines a dynamic graph autoencoder (DyGraphAE) with CCA for monitoring nonlinear dynamic processes. Unlike static graph autoencoders, such as graph dynamic auto encoder (GDAE), which assume static relationships, DyGraphAE uses graph convolution and gated recurrent unit (GRU) layers to represent the dynamics of an evolving process. CCA improves fault detection by analyzing the correlation between the reconstructed model and the original model, resulting in better performance compared to methods fault detection rate (FDR) and false alarm rate(FAR) for fault 4 on the tennessee eastman process (TEP) dataset. This integrated approach provides a novel way to investigate different aspects of industrial process monitoring.