Dynamic data feature engineering for process operation troubleshooting

Sustained dynamic variations in process data are usually indicators of poor control performance or malfunctioning of control valves. In this chapter, dynamically evolving latent feature analysis (DELFA) is introduced to extract interesting latent dynamic features for troubleshooting and diagnosing plant-wide process anomalies. Composite loadings and composite weights are derived and applied for diagnosing the root cause of a dynamic feature contained in one or more latent variables. The DELFA procedure is applied to an industrial plant dataset to troubleshoot process anomalies and diagnose the causes in a plant-wide setting. Dynamic inner canonical correlation analysis is shown to be superior to slow feature analysis in extracting clean dynamic features and diagnosing the root causes.