Data-driven dynamic modeling and online monitoring for multiphase and multimode batch processes with uneven batch durations

Abstract

Batch processes are often characterized by piecewise linear dynamics due to varying operating conditions. Multiphase and multimode modeling of batch processes is a common technique that offers insight into the process operation and improved online monitoring. However, existing monitoring methods have several drawbacks such as neglecting process dynamics, requiring separate treatment of transient behavior, and relying on uniformity between batches. These challenges are addressed here by proposing a new strategy to construct a dynamic model for monitoring multimode and multiphase batch processes. A linear dynamic system partitions phases and describes local dynamic behavior before modes of operation are clustered based on the global differences between batches. Lastly, an expectation maximization algorithm for multibatch data in the same mode is applied to estimate phase parameters. Process monitoring results on a benchmark penicillin fermentation data set suggest a significant improvement over previous methods.

Download paper here

Recommended citation: Wang, K., Rippon, L., Chen, J., Song, Z., & Gopaluni, R. B. (2019). “Data-driven dynamic modeling and online monitoring for multiphase and multimode batch processes with uneven batch durations.”; Industrial & Engineering Chemistry Research. 58(30), 13628-13641.