EverAdapt: Continuous Adaptation for Dynamic Machine Fault Diagnosis Environments

TL;DR

EverAdapt addresses the challenge of continual domain adaptation in machine fault diagnosis by combining Class-Conditional Alignment (CCA) with Continual Batch Normalization (CBN), supplemented by entropy minimization and a lightweight replay strategy. It preserves knowledge from previously seen domains while adapting to new ones, mitigating catastrophic forgetting and achieving state-of-the-art accuracy and backward transfer on real bearing datasets. The framework dynamically balances adaptation and retention through an adaptive objective and minimal replay requirements, demonstrating robust performance in dynamic industrial environments. Overall, EverAdapt offers a scalable, practical solution for continual fault diagnosis under evolving operational conditions.

Abstract

Unsupervised Domain Adaptation (UDA) has emerged as a key solution in data-driven fault diagnosis, addressing domain shift where models underperform in changing environments. However, under the realm of continually changing environments, UDA tends to underperform on previously seen domains when adapting to new ones - a problem known as catastrophic forgetting. To address this limitation, we introduce the EverAdapt framework, specifically designed for continuous model adaptation in dynamic environments. Central to EverAdapt is a novel Continual Batch Normalization (CBN), which leverages source domain statistics as a reference point to standardize feature representations across domains. EverAdapt not only retains statistical information from previous domains but also adapts effectively to new scenarios. Complementing CBN, we design a class-conditional domain alignment module for effective integration of target domains, and a Sample-efficient Replay strategy to reinforce memory retention. Experiments on real-world datasets demonstrate EverAdapt superiority in maintaining robust fault diagnosis in dynamic environments. Our code is available: https://github.com/mohamedr002/EverAdapt

Figures (3)

• Figure 1: Comparison of Conventional and Continual Adaptation Approaches in Domain Adaptation. Top: the conventional adaptation approach, where individual models are independently trained for each new target domain. This often results in a scalability issue as the number of target domains increases, necessitating separate model and training phases for each domain. Bottom: the continual adaptation strategy, which employs a singular model that is sequentially adapted across multiple target domains. This method maintains knowledge from previous domains, effectively mitigating catastrophic forgetting and promoting model adaptation across a series of domain shifts.
• Figure 2: OverAdapt Framework Overview: incorporates input source samples, input target samples from the current target domain, and input memory samples to the feature extractor. It applies conditional entropy loss on the feature space of the target samples, cross-entropy loss on the input source samples, self-training with pseudo-labels on the memory samples, and local alignment loss between the source and target features.
• Figure 3: Model Analysis for Everadapt