This paper presents a comprehensive survey on interpretable intelligent fault diagnosis for rotating machinery, addressing the challenge of the “black box” nature of machine learning techniques that hampers reliability in automated diagnostic processes. It underscores the growing importance of interpretability in intelligent fault diagnosis (IFD), marking a shift from traditional signal processing methods to machine learning-based approaches that necessitate transparency for trustworthiness. Our review systematically collates and examines the spectrum of interpretability in IFD, distinguishing between post-hoc and ante-hoc strategies. We detail mainstream post-hoc methods, their applications, and critique their limitations, particularly the absence of physical significance. The survey then explores ante-hoc methods that incorporate physical knowledge upfront, enhancing interpretability. By categorizing and evaluating three distinct knowledge embedding approaches, we shed light on their unique applications. Conclusively, we highlight emerging research directions and challenges in the field, aiming to equip readers with a nuanced understanding of current methodologies and inspire future studies in making IFD more reliable and interpretable.
1 November 2020 in Mechanical Systems and Signal Processing
The maturity of sensor network technologies has facilitated the emergence of an Industrial Internet of Things (IIoT), which has collected an increasing volume of data. Converting these data into actionable intelligence for fault diagnosis is key to, e.g., reducing unscheduled downtime and performance degradation. This paper formalizes a problem called semantic fault diagnosis–to construct the formal specifications of faults directly from data collected from IIoT-enabled systems. The specifications are written as signal temporal logic formulas, which can be easily interpreted by humans. To tackle the issue of the combinatorial explosion which arises, we propose an algorithm that combines ideas from agenda-based searching and imitation learning to train a policy that searches formulas in a strategic order. Specifically, we formulate the problem as a Markov decision process, which is further solved with a reinforcement learning algorithm. Our algorithm is applied to time series data collected from an IIoT-enabled iron-making factory. The results show empirically that our proposed algorithm is both scalable to the size of the data set and interpretable, therefore allowing human users to take actions for, e.g., predictive maintenance.
17 February 2026 in Engineering Applications of Artificial Intelligence
This study examines the issue of interpretability in fault diagnosis for rolling bearings using a symbolic learning technique. We propose the adoption of weighted signal temporal logic (wSTL) as a formal language and introduce the temporal logic network (TLN) as a neural-symbolic learning architecture capable of encoding symbolic wSTL representations for input signals. TLN is comprised of three sub-networks: a basic predicate network for abstracting features and generating predicates from vibration signals, an autoencoder for identifying significant signal components, and a logic network for constructing a formal language that aids in fault classification and model interpretation. To improve comprehensibility, timed failure propagation graphs (TFPGs) are employed to visually represent the logical relationships and propagation of fault events. Experimental results demonstrate TLN’s ability to extract impulse fault patterns from signals, accurately describe fault events through learned wSTL formulas and enhance understanding of fault events for non-expert individuals through TFPGs. These findings contribute to the field of fault diagnosis in rolling bearings by incorporating symbolic learning techniques, utilizing formal language representation and TFPG for improved interpretability.