Integrating process data and topology information through edge-group sparse principal component analysis for process analytics

Despite rapid advancements in deep learning, traditional methods like principal component analysis (PCA) remain indispensable in chemical process analysis due to their strong mathematical foundations and powerful visualization capabilities, which uncover variable correlations and reveal process variations. This study introduces edge-group sparse PCA (ESPCA) for process analytics, integrating process topology while enforcing sparsity on loading vectors to enhance interpretability. A systematic application procedure is demonstrated through illustrative examples. In these applications, ESPCA proves particularly effective in identifying key process units and variables associated with faults or disturbances, providing a solid foundation for root cause analysis. Visualization tools play a crucial role in integrating available process knowledge, facilitating the interpretation of results, and enabling engineers to derive conclusions in a clear and intuitive manner. Additionally, statistical causality analysis methods like transfer entropy can be used alongside ESPCA to trace propagation paths and pinpoint root causes of process anomalies.