Monitoring of caliper sensor fouling in a board machine using self-organising maps

This paper presentes a process-monitoring scheme utilising adaptive self-organising maps (SOM) to detect process conditions that lead to the fouling of a caliper sensor in a board machine. The scheme is based on mapping on a SOM the process measurements and the calculated variables which provide insight into the chemical phenomena involved in fouling to classify faulty process conditions. The time-variant nature of the board making process was taken into account by regularly re-training the SOM. The monitoring scheme is demonstrated with industrial data, and the results are presented and discussed.

► Monitoring scheme using SOM for caliper sensor fouling in a board machine is proposed.
► Fouling is detected by classifying process states into faulty and normal conditions.
► Process knowledge is augmented to SOM monitoring in terms of calculated variables.
► The SOM is adapted to time-variant process behaviour with regular re-training phases.
► The SOM is trained, tested and validated using industrial data from the board machine.