異常値検知手法を用いた液体搬送ラインの異物混入事前予測

抄録

Filters are often used in beverage and other production lines to remove foreign material during fluid transfer. The removal of foreign material causes clogging, which leads to further generation of foreign material due to an increase in the mesh opening or damage to the mesh. Although clogging can be determined from flow rate and pressure changes, it is difficult to detect clogging until after the event has progressed to a certain degree, during which time foreign material is generated. In this study, we extracted feature values from the differential pressure applied to the filter and performed machine learning using One-Class SVM to predict foreign material contamination in advance. We applied this method to an actual fluid transfer line and confirmed its usefulness.