A Method to Estimate the Remaining Useful Life of a Filter Using a Hybrid Approach Based on Kernel Regression and Simple Statistics

##plugins.themes.bootstrap3.article.main##

##plugins.themes.bootstrap3.article.sidebar##

Published Jul 22, 2020
Roman Łomowski Szczepan Hummel

Abstract

This paper describes the method used by the Uptime team for the estimation of the remaining useful life of a filter during the 2020 PHM Conference Data Challenge. The proposed method is a hybrid of two methods: (1) based on median lifetime of a filter for a particular contamination profile and (2) kernel regression for a sensor-based prediction after a certain threshold of differential pressure is reached. The threshold value was chosen based on visual assessment followed by grid search for fine tuning. Median lifetime of a filter for unseen contamination profiles was estimated using interpolation. Choosing the right interpolation method was a challenge as training data contained samples with only two values of contamination particle size. Interpolation was chosen based on other publicly available information about the relationship between contamination profile and filter lifetime. The results (ranked 1st with the total penalty score of 49.67) showed that an observation made based on one dataset can be useful for solving similar problems in the case of limited data availability. This suggests that there is a potential for using transfer learning in PHM applications.

Abstract 86 | PDF Downloads 109

##plugins.themes.bootstrap3.article.details##

Keywords

data challenge

Section
Data Challenge Winners