Explainable AI for Predictive Maintenance in Industrial IoT: A Transfer Learning Approach
Abstract
In Industrial Internet of Things (IIoT) settings, predictive maintenance has become a vital tool for lowering maintenance expenses, increasing operational effectiveness, and minimising equipment downtime. Nevertheless, conventional deep learning models for failure prediction and detection frequently operate as opaque black boxes that lack interpretability and transparency, two essential components for industrial implementation. In order to improve predictive maintenance in IIoT systems, this study suggests an explainable AI (XAI) architecture that makes use of transfer learning. Even with small labelled datasets, the method's use of pre-trained deep learning models that have been refined using domainspecific sensor and machine data allows for quicker convergence and better efficacy. The system incorporates model-agnostic XAI techniques to handle the interpretability issue and offer useful perspectives into the models' decision-making process. Results from experiments on benchmark IIoT datasets show that the suggested approach provides insightful justifications for maintenance choices while achieving excellent fault prediction accuracy. Field engineers and maintenance planners are more likely to accept the predictive models when transfer learning and XAI are combined since it increases transparency and confidence. The potential for generalising the suggested technique across several industrial sectors for scalable, intelligent maintenance solutions is highlighted in the paper's conclusion.