Applying Software Engineering for Robust Early Fault Detection in Industrial Systems Using Heterogeneous Ensemble Learning Models

Abstract

Software engineering plays a critical role in developing robust and scalable systems to monitor industrial systems to predictive maintenance. Industrial systems should have early fault detection and prevent equipment failures and downtime. Single classifiers do not however work well in identifying subtle and rare defects especially in imbalance data set. In this paper, I am proposing a heterogeneous ensemble learning model consisting of Random Forest, Gradient Boosting and Support Vector Machine (SVM) with a stacking method in order to enhance the detection accuracy. Data pre-processing methods were used to improve model performance such as feature scaling and class balancing. The experimental results demonstrate that the proposed ensemble had the accuracy of 0.95, precision of 0.93, recall of 0.91, and F1-score equal to 0.92 with the original dataset (1000 samples). As the number of samples grew to 5000, the accuracy, precision, recall and F1-score improved to 0.96, 0.94 and 0.93 respectively. This paper presents a heterogeneous stacking ensemble structure that is scalable, which incorporates multiple classifiers in a modular software design, and is capable of providing robust early fault detection during small and large data sets. The analysis of the importance of features showed that Temperature and Vibration are the most important indicators of early faults. The suggested solution is very robust and reliable in fault detection, it has low false alarms and it offers a good predictive maintenance solution to industrial and IoT enabled systems.