(14e) Dealing With Missing Data and Synchronization Issues in Batch Process Monitoring

After visiting professor John F. MacGregor in Fall 1997 I got really fascinated by the potential of multivariate latent variable methods for batch process monitoring. Based on the inspiring contributions of John I went on studying new methods to ease and improve batch process monitoring.

In this talk I will present some of our novel contributions in this field, based on previous John´s work. In particular I will review two issues: missing data and real-time batch synchronization.

In process chemometrics missing data can appear for a number of reasons: failures in sensors or in the communication between the instrumentation and the digital control system (DCS), sensors taken offline for routine maintenance, manual samples not collected at the required times, data discarded by gross measurement errors, and sensors with different sampling periods. In real-time-batch process monitoring and control, at each time t the future behaviour of the process trajectories is unknown, and then can be treated as missing values. Out of the proposed methods, Trimmed Scores Regression (TSR) [1] turns out to be a simple an efficient method to estimate missing data.

Usually batch pace evolution is different batch-to-batch yielding misaligned key process events. Real-time batch synchronization is a critical issue for real-time batch process monitoring. For this purpose the Relaxed-Greedy Dynamic Time Warping (RGTW) algorithm [2] will be introduced.

[1] Arteaga F., Ferrer A. Dealing with missing data in MSPC: several methods, different interpretations, some examples. J. Chemometrics 2002; 16: 408–418.

[2] González-Martínez J., Ferrer A., Westerhuis J. Real-time synchronization of batch trajectories for on-line multivariate statistical process control using dynamic time warping, Chemometrics and Intelligent Laboratory Systems 2011; 105: 195-206.