On Data Science for Process Systems Modeling, Control and Operations

Abstract

Data science is emerging as a multidisciplinary field with tremendous recent development in theoretical foundations and expanded applications in both science and engineering. Engineering applications include industrial data analytics, autonomous systems, energy analytics, environmental applications, economic data modeling, image sequence modeling, and other high dimensional time-series data analytics. This paper is concerned with the integration of data science with dynamic systems, monitoring and control. The development of machine learning is reviewed in both a neural-mimic learning route and a learning control route, which deals with intrinsically uncertain dynamic systems. The paper then reviews the interaction of data with process manufacturing systems modeling and control, involving both data and first principles models with varying proportions. Problems include data reconciliation, state and disturbance estimation, system identification, process monitoring, and inferential property estimation. For time series data in process manufacturing systems, we present latent dynamic variable modeling methods to extract the principal dynamics in a low dimensional subspace of the data. The approaches effectively distill latent dynamic features from the data for easy interpretation, prediction, and visualization. Case studies are presented to illustrate how these latent dynamic analytics extract important features for process interpretation, troubleshooting, and monitoring.