An Incremental Learning Method for Fault Diagnosis Based on Random Forest

Abstract

AbstractEquipment fault diagnosis is an important part of the modern industrial process. However, with the increase of working time, new types of faults will appear continuously. It is hard to take into account all types of faults previously due to the complexity and variability of industrial processes and environment. Thus, it is necessary to build a fault diagnosis system that can continuously detect and classify new type faults. In this paper, we propose Incre-RF, an incremental learning method for fault diagnosis based on Random Forest (RF), which can detect and classify new type of fault timely by RF-based incremental learning. We test Incre-RF on a popular open data set. The result shows Incre-RF performs better in both new type fault detection and new type fault learning ability compared with a state-of-the-art method.KeywordsFault diagnosisNew class classificationEnsemble systemOnline detection