An Explainable Ensemble Machine Learning Framework for Predictive Maintenance in Industrial IoT Systems Using Multi-Sensor Data

Abstract

The complexity of modern industrial systems and their need for intelligent maintenance to avoid unexpected failures and to improve the efficiency of production have led to the development of Predictive Maintenance (PM) using data from Industrial Internet of Things (IIoT). PM using IIoT uses a large amount of data gathered continuously by multiple sensors installed on industrial equipment. The use of machine learning (ML) has become common in PM using IIoT. However, existing ML approaches to PM using IIoT have many restrictions, such as sensor data variability, high dimensionality of feature space, class imbalance, and lack of interpretability. This paper proposes an explainable ensemble machine learning framework that uses multi-sensor data fusion for PM using IIoT. The proposed framework consists of data preprocessing, feature engineering, ensemble learning, and explainable AI. It uses three types of ML classifiers, namely, Random Forest, XGBoost, and LightGBM, and combines them using ensemble learning to improve prediction accuracy. It uses SHAP to explain the contribution of individual sensor features to the maintenance decision and provides interpretable results. The performance of the proposed framework is evaluated using several real industrial multi-sensor datasets. The framework provides high accuracy for fault prediction, while it is also explainable. Therefore, it can support effective and reliable industrial maintenance decisions. The results of this study also show that the proposed explainable ensemble ML framework can support the development of intelligent and trustworthy PM using IIoT in Industry 4.0.