Latest revision as of 14:10, 11 February 2026

Abstract

Uncertainty and conflicting information are pervasive in artificial intelligence (AI)-driven engineering systems, especially in multisensor fault diagnosis. Dempster-Shafer theory (DST) has garnered significant interest across various fields as it provides a powerful framework for modeling uncertainty. However, despite its advantages, the application of Dempster’s rule can lead to paradoxical outcomes when it encounters highly conflicting evidence. To address this limitation, this paper first presents a new evidential Jensen-alpha divergence (EJ AD) to quantify the discrepancy between the evidence items based on DST. Furthermore, an advanced version, the reinforced evidential Jensen-alpha divergence (REJ AD) is developed, which takes into account the quantity of potential propositions. We demonstrate thatREJ ADcan be transformed into various divergences such as the χ2divergence, Jensen-Shannon divergence, Hellinger distance, and arithmetic-geometric divergence under certain conditions. Also, we show the key properties ofREJ AD, including non-negativity, non-degeneracy and symmetry. Additionally, we design a new multisensor fault diagnosis method utilizingREJ ADand belief entropy. The superior performance of the proposed method is tested in three distinct fault diagnosis cases, and analysis shows robust performance across a range of its key parameter α, offering a computationally feasible, scalable and interpretable solution for AI-based decision-making in real-world engineering applications.OPEN ACCESS Received: 15/10/2025 Accepted: 17/11/2025

Document