Paper Machine Press Felt Monitoring : A Case Study on PM2 in Karlsborg

Sammanfattning: Press felts are highly critical components of the paper machine. A degraded press felt could lead to paper web breaks, which requires the paper machine to be restarted. Moreover, a degraded felt influences the quality of the paper, leading to paper disposal. Condition monitoring aims at minimising the risk of paper web breaks, unsatisfactory paper quality and other types of production loss while maximising the useful life of the press felts. However, installing a new condition monitoring system is expensive and the installation can be difficult to fit into the scheduled maintenance stops. This thesis investigates the possibility of using existing monitoring systems instead of installing a new one. Four possible approaches of monitoring the degradation of the press felts have been explored. The identified approaches of press felt monitoring were tested by using data acquired through existing monitoring systems of the paper machine PM2 at BillerudKo-rsnäs in Karlsborg, located in the north of Sweden. The first approach is based upon process parameters. This approach could, however, not be properly investigated due to a malfunctioning sensor. The second approach revolves around the natural frequencies of the felt and the frequency changes as the felt degrades. The remaining two approaches originates from the hypothesis that felt degradation could lead to impacts as the possibly uneven felt passes the rollers. One approach is to detect these possible impacts by using the time domain feature kurtosis. The other approach is to monitor the harmonics these impacts could lead to. Neither the natural frequency nor the kurtosis approach was deemed promising, partly based on the results of the analysed data but also due to intrinsic deficiencies of these approaches. The approach based on felt harmonics did, however, exhibit indications that it might be a feasible monitoring technique. The felt harmonics approach should be further investigated. Furthermore, a python program that can synchronise data from different sources was developed. This program enables degradation features to be extracted using machine learning algorithms. However, due to the lack of vibration data and labels of the current felt condition, machine learning was not applied.