Fault detection and predictive maintenance in IoT-based building management system using machine learning

Abstract

Infrastructures in the modern era are incorporating the Internet of Things (IoT) in everything from complex building automation systems (BAS) to individual small devices, emphasizing the importance of Predictive Maintenance. As a result, early fault detection is required, especially in sensitive and massive structures such as hospitals, industries, and multipurpose buildings. In such infrastructures, even minor failures can result in tragedies such as fires or slow down productivity. In our research, we have used several machine learning fault detection and diagnostics (FDD) algorithms in building fault detection data. We collected two datasets, MZVAV-1 (SET-A) and MZVAV-2-1 (SET-B), which were split into train-test sets to deploy LogisticRegression, KNearestNeighbours, Naive Bayes classifier, Support Vector Classifier, RandomForestClassifier, Decision Tree, MLP Classifier and Extra Tree Classifier. We achieved the highest accuracy of 98.91% using the Decision Tree classifier and the lowest accuracy of 14.17% from Naive Bayes classifier on the MZVAV-1 dataset. RandomForestClassifier and ExtraTree classifier outperformed all other algorithms with 99.91% accuracy on the MZVAV-2-1 dataset.