Dynamically Slided Chemometric Models for Robust On-line Prediction of Cloud Point in Melamine Resin Production

In the considered batch process of melamine resin production, the essential process parameter to be regularly supervised is the cloud point. Monitoring the value of that parameter indicates the best point of time to turn off heating in order to stop the condensation and is of utmost importance to assure high product quality. Currently, the cloud point is measured manually, which takes high efforts for the operators: walking to the basin, drawing the probes and analyzing them. To automatize the supervision of the cloud point a FT-NIR spectrometer has been installed by RECENDT GmbH, whose optics contains a monolithic Michelson interferometer, which helps with temperature stability and expands the spectral range from 4000cm-1 to 10000cm-1. Standard linear calibration models, in particular PLS [1][2], have been extracted from an initial calibration set and applied onto a further separate on-line test set, both drawn over a time frame of about 6 months from the process. However, they did not lead to satisfactory results, especially in case of lamp changes and intensity downtrends of the halogen light source, a dynamics which cannot be fully controlled by static models. Furthermore, some internal dynamics such as changing compositions of the educt etc. are assumed.