Machine learning application for sensor failure detection in polymerization process

This work analysed the time signals from 5 instruments distributed in an industrial polymerization facility: two temperature instruments, a water level instrument, a weight instrument and a flow instrument. The dataset is composed by a five years history with a sample rate of 1 minute. A specialist using the event related industry report labelled the signals. The results using random forest as the machine learning classifier reached significant performance in detecting failures independent of the instrument, preliminary indicating the suitability of the framework.