(394b) Denoising Dynamic Composition Data from a Multi-Sample Instrument in a Pilot Unit

In today’s world, data and data analysis is gaining an ever more important role. In the chemical industry, it is common to analyze the compositions of N streams with a single analytical instrument. The stream flowing into the instrument is changed from stream 1 to N over N sequential sampling intervals, and this cycle is repeated with time. While such measurement methods are widespread, they introduce noise due to switching of the streams and fluctuating process conditions. Such noise hinders the analysis of time-series data for process monitoring and control.

In this poster, we employ robust local weighted regression and stochastic regression techniques to denoise each signal separately and benchmark their performance against simple averaging within each sampling interval. We explore the benefits and drawbacks of each method, emphasizing how advanced denoising significantly improves accuracy, particularly in dynamic measurement scenarios. This research offers practical recommendations to enhance reliability and precision in dynamic composition measurements obtained from multi-stream instruments used in pilot-scale chemical processes.