An interpretable approach based on possibilistic hypothetical case-based reasoning for fault diagnosis

Fault identification (FI) systems based on case-based reasoning (CBR) adopt a pattern recognition (PR) perspective, considering observable features and treating different faults as fault semantic classes. However, their limitations lie in the necessity to establish case bases with observation vectors that are fully informed. This requirement neglects the essential aspect of the sequential reasoning employed by experts based on the observed features, a critical element in human experiential knowledge. To address these challenges, a novel approach called PH-CBR (possibilistic hypothetical-CBR) is proposed for FI. This modeling seeks to capture the human diagnosing expertise facing the burdensome task of FI. PH-CBR focuses on a hypothetical case base (H-CB) representation following the observed sequence of the expert based on hypothetical reasoning. Additionally, it introduces possibilistic reasoning to retrieve a hypothetical case (H-case) incorporating the heterogeneity of features. Furthermore, PH-CBR includes the estimation of a possibilistic state-of-knowledge vector (PSK) to improve the accuracy of the FI. A case study is presented to validate the performance of the PH-CBR approach, utilizing a condenser maintenance database collected from the cleaning device. Experimental results demonstrate its performance, achieving an overall accuracy of 95.49%, outperforming the classical-CBR (C-CBR) approach. PH-CBR delivers encouraging abilities in modeling the H-CB which allows capturing the human expertise leading, thus, to improved quality in retrieved H-cases in a homogeneous space of decision making for FI. Moreover, the PH-CBR approach significantly reduces the number of observed features and, consequently, the time required for FI, which can significantly impact the efficiency of industrial machines.