Conference article

Partial least squares PLS1 vs. PLS2 – optimal input/output modeling in a compound industrial drying oven

Maths Halstensen
Department of Electrical Engineering, IT and Cybernetics, University of South-Eastern Norway

Ulrich Hundhausen
Norwegian Institute of Wood Technology (NTI), Oslo, Norway

Kim H. Esbensen
KHE Consulting, Copenhagen, Denmark

Download articlehttps://doi.org/10.3384/ecp20176458

Published in: Proceedings of The 61st SIMS Conference on Simulation and Modelling SIMS 2020, September 22-24, Virtual Conference, Finland

Linköping Electronic Conference Proceedings 176:65, p. 458-465

Show more +

Published: 2021-03-03

ISBN: 978-91-7929-731-2

ISSN: 1650-3686 (print), 1650-3740 (online)

Abstract

A feasibility study was carried out to assess the possibility of developing prediction models for monitoring drying conditions of wood coatings in one of Europe’s largest and most modern coating plants for exterior cladding. These models were based on data from real-time Process Analytical Technology (PAT) sensors, measuring airflow and air direction, temperature and relative humidity). The study revealed that the information from the PAT sensors gave sufficient input to accurately model the complex drying conditions and their interrelations. Modelling was carried out using both Principal Component Analysis (PCA) and PLS-regression in both its PLS1 and PLS2 manifestations. In addition, the diagnostic prediction performance RMSEP between PLS1 and PLS2 models were not significantly different. This is advantageous for an industrial implementation concerning re-calibration operations: PLS1 requires 40 separate calibrations whereas PLS2 requires only one, because PLS1-R is a regression of a singular output variable (y-variable) and PLS2-R of several simultaneous, correlated output variables. While a single calibration based on PLS2 will take approximately one hour, the PLS1 approach will take more than a week.

Keywords

process monitoring; exterior wood coatings; wood cladding; oven drying; multivariate modeling; PLS1-R, PLS2-R

References

No references available

Citations in Crossref