Interpretable Prognostics with Concept Bottleneck Models

TL;DR

This work tackles the interpretability gap in prognostics by introducing Concept Bottleneck Models (CBMs) that predict RUL through high-level degradation-mode concepts. By using component degradation states as intermediate concepts, the authors explore CBMs, their extensions (CEMs, Hybrid CBMs), and test-time interventions, demonstrating competitive or superior performance on the N-CMAPSS turbofan dataset compared with non-interpretable baselines. Key findings show that CEMs achieve top RUL accuracy and concept fidelity, while test-time interventions can further improve predictions with minimal degradation of interpretability. The study highlights the practical value of interpretable prognostics for safety-critical assets and points to robust performance even with incomplete concept labeling, albeit with limitations around concept labeling requirements and deployment considerations.

Abstract

Deep learning approaches have recently been extensively explored for the prognostics of industrial assets. However, they still suffer from a lack of interpretability, which hinders their adoption in safety-critical applications. To improve their trustworthiness, explainable AI (XAI) techniques have been applied in prognostics, primarily to quantify the importance of input variables for predicting the remaining useful life (RUL) using post-hoc attribution methods. In this work, we propose the application of Concept Bottleneck Models (CBMs), a family of inherently interpretable neural network architectures based on concept explanations, to the task of RUL prediction. Unlike attribution methods, which explain decisions in terms of low-level input features, concepts represent high-level information that is easily understandable by users. Moreover, once verified in actual applications, CBMs enable domain experts to intervene on the concept activations at test-time. We propose using the different degradation modes of an asset as intermediate concepts. Our case studies on the New Commercial Modular AeroPropulsion System Simulation (N-CMAPSS) aircraft engine dataset for RUL prediction demonstrate that the performance of CBMs can be on par or superior to black-box models, while being more interpretable, even when the available labeled concepts are limited. Code available at \href{https://github.com/EPFL-IMOS/concept-prognostics/}{\url{github.com/EPFL-IMOS/concept-prognostics/}}.

Figures (16)

• Figure 1: Our proposed concept-based architectures for interpretable prognostics. Intermediate concepts $\mathbf{c}_i$ corresponding to component degradation modes are used to predict the remaining useful life (in this example, four concepts correspond to the Fan, HPC, HPT and LPT components of a turbofan engine).
• Figure 2: Workflow of a potential concept intervention strategy for prognostics evaluated in this study. After a fault is detected by the model, an operator can inspect the asset and confirm the degradation, and proceed to a new estimation of the RUL given this new knowledge. This re-estimation would be impossible with standard prognostics models.
• Figure 3: Example of concept intervention on the LPT in DS07 unit 10. A fault is detected on the LPT when the corresponding concept activates over 0.5 during a cycle. After confirming the degradation through an inspection, we intervene on the LPT concept by setting its activation to 1 for the remaining cycles, resulting in a new RUL estimation.
• Figure 4: Schematic representation of the CMAPSS model arias_chao_aircraft_2021. Air intake enters by the Fan, and a part of the flow passes though the low-pressure compressor (LPC), high-pressure compressor (HPC), combustion chamber, high-pressure turbine (HPT) and low-pressure turbine (LPT).
• Figure 5: Examples of the degradation trajectories and associated RUL in N-CMAPSS dataset DS01 (test units). The concept label $c$ (here, corresponding to HPT) is obtained by binarizing the degradation level $\theta$ using a fixed threshold of $-0.0015$.