Explainable AI Models for Predictive Maintenance in Smart Manufacturing Systems

Abstract

Predictive maintenance (PdM) is one of the determinants in the smart manufacturing in which the data-driven information is being applied to cut down the amount of maintenance costs and schedule with the down time and increase the overall management of the systems. Using conventional machine learning (ML) models on predictive maintenance on the other hand is not devoid of the issue on transparency and explainability. This current paper is an argument on how to adopt Explainable Artificial Intelligence (XAI) models in predictive maintenance of smart manufacturing systems and is addressed to bridge the gap between the accuracy of predictive maintenance and the interpretability of the models. XAI models seize an opportunity of not only jumping up the quality of decision making processes but also building trust in maintenance predictions among the human operators owing to the clearness and simplicity of the explanations of the processes. Evaluated methods In this paper, some of the XAI methods, such as the decision trees, LIME (Local Interpretable Model-agnostic Explanations), and SHAP (Shapley Additive Explanations), will be analyzed to evaluate and identify their applicability to the predictive maintenance systems. We will also discuss the consequences of implementing the said models in the real manufacturing environments both with regard to the gains as well as the hurdles. The point of the desired result is to demonstrate how explainability in AI models can be applied to ensure that predictive maintenance strategies in smart manufacturing become more solid and acceptable.