Skip to content
Conference Open access

A data-driven and knowledge-informed interpretable model for fault prediction

Oct 2026 · International Conference on Information and Software Technologies · pp. 258-267 · 0 citations · 21 references

Abstract

With the development of Industry 4.0, Predictive Maintenance (PdM) usually uses Artificial Intelligence (AI) technologies to predict the probability of future failures by detecting the operating status of equipment or systems, thus helping engineers to explore maintenance strategies significantly. Although many PdM methods have obtained good performance, especially those based on Deep Learning (DL), their black-box decision process often makes it difficult to provide a reasonable and understandable explanation, causing users’ distrust of the prediction results. To solve this problem, this paper proposes an interpretable method that integrates knowledge and data to predict the Remaining Useful Life (RUL), explain the prediction results, and identify the degradation type and location. First, an artificial neural network is developed to predict potential faults in the target equipment. To bridge the gap between model outputs and domain semantics, a mapping mechanism is established to associate the activation patterns of specific neurons in the artificial neural network with pre-defined bottom events, thereby enabling the extraction of semantically meaningful fault-related information from raw input features. Furthermore, a logic-based inference mechanism is explored, leveraging a Knowledge Graph (KG) constructed from real-world datasets, to generate structured explanations for the network’s predictions. Experimental evaluations on the turbofan engine dataset indicate that the proposed approach effectively improves the interpretability of the fault predictions while maintaining high predictive accuracy.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.