Concept Drift Detection and Adaption in Big Imbalance Industrial IoT Data Using an Ensemble Learning Method of Offline Classifiers

In a smart factory, thousands of industrial Internet of Things (IIoT) devices or sensors are installed in production machines to collect big data on machine conditions and transmit it to a cyber-physical system in the cloud center of the factory. Then, the system employs a variety of condition-based maintenance (CBM) methods to predict the time point when machines start to be operated abnormally and to maintain them or replace their components in advance so as to avoid manufacturing enormous detective products. CBM suffers from problems of concept drifts (i.e., the distribution of fault patterns may change over time) and imbalance data (i.e., the data with faults accounts for a minority of all data). Ensemble learning that integrates the diversity of multiple classifiers provides a high-performance solution to address these problems. In practice, most companies may not have a sufficient budget to establish a sound infrastructure to support real-time online classifiers, but may have off-the-shelf offline classifiers in their existing systems. However, most previous works on ensemble learning only focused on supporting online classifiers. Consequently, this work proposes an ensemble learning algorithm that supports offline classifiers to cope with three-stage CBM with concept drifts and imbalance data, in which Stages 1 (training an ensemble classifier) and 3 (creating a new ensemble) employ an improved Dynamic AdaBoost.NC classifier and the SMOTE method to address imbalance data; and Stage 2 (detecting concept drifts in imbalance data) employs an improved LFR (Linear Four Rates) method. The experimental results on datasets with different degrees of imbalance show that the proposed method can successfully detect all concept drifts, and has a high accuracy rate in detecting minority-class data, which is over 94%.

Abstract licence: CC BY-NC-ND

34 more references, whose wording the paper’s licence keeps private.

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.