Self-healing machine learning system for fault detection using predictive maintenance

Abstract

Machine learning-powered fault detection systems are the backbone of modern predictive maintenance solutions in the context of Industry 4.0, as they allow organizations to prevent expensive breakdowns and unnecessary machine downtime. Still, it is a known fact that machine learning systems are prone to model drift, which happens when a learning model performs worse due to changes in input data patterns. In dynamic industrial environments where machines operate under varying conditions, such changes are frequent and inevitable. Therefore, predictive maintenance systems must have built-in self-learning and self-healing mechanisms that allow them to adapt to new patterns and maintain the desired performance level. This paper proposes a Self-Healing Machine Learning System for Fault Detection to perform these tasks and demonstrates the concept’s feasibility within the Python/Flask environment. The proposed system utilizes Scikit-learn and the AI4I 2020 Predictive Maintenance Dataset to train an isolated random forest classifier and test its healing mechanism’s performance. The results demonstrate that the proposed approach can keep the random forest classifier’s performance stable by re-training it with new data whenever the performance metric falls below a certain threshold. Therefore, the self-healing ML system described in this paper can be a viable solution for industrial applications, as it ensures continuous learning and adaptability in the ever-changing environment of Industry 4.0.