Overcoming Negative Transfer by Online Selection: Distant Domain Adaptation for Fault Diagnosis

TL;DR

This work introduces distant domain adaptation for fault diagnosis and proposes Online Selective Adversarial Alignment (OSAA) to counteract negative transfer when source and target domains are substantially different. OSAA combines online gradient masking to exclude distant source samples, construction of an intermediate domain to ease cross-domain alignment, and conditional adversarial alignment that leverages label information during domain adaptation. Through detailed experiments on PU and CWRU datasets, OSAA outperforms state-of-the-art unsupervised DA methods and demonstrates robust ablations and sensitivity analyses, illustrating practical utility under severe domain shifts. The approach offers a scalable, fully unsupervised solution that reduces negative transfer and improves fault-diagnosis accuracy in real-world industrial settings.

Abstract

Unsupervised domain adaptation (UDA) has achieved remarkable success in fault diagnosis, bringing significant benefits to diverse industrial applications. While most UDA methods focus on cross-working condition scenarios where the source and target domains are notably similar, real-world applications often grapple with severe domain shifts. We coin the term `distant domain adaptation problem' to describe the challenge of adapting from a labeled source domain to a significantly disparate unlabeled target domain. This problem exhibits the risk of negative transfer, where extraneous knowledge from the source domain adversely affects the target domain performance. Unfortunately, conventional UDA methods often falter in mitigating this negative transfer, leading to suboptimal performance. In response to this challenge, we propose a novel Online Selective Adversarial Alignment (OSAA) approach. Central to OSAA is its ability to dynamically identify and exclude distant source samples via an online gradient masking approach, focusing primarily on source samples that closely resemble the target samples. Furthermore, recognizing the inherent complexities in bridging the source and target domains, we construct an intermediate domain to act as a transitional domain and ease the adaptation process. Lastly, we develop a class-conditional adversarial adaptation to address the label distribution disparities while learning domain invariant representation to account for potential label distribution disparities between the domains. Through detailed experiments and ablation studies on two real-world datasets, we validate the superior performance of the OSAA method over state-of-the-art methods, underscoring its significant utility in practical scenarios with severe domain shifts.

Figures (5)

• Figure 1: Classic domain shift versus distant domain shift.
• Figure 2: Left: Existence of distant samples causes conventional domain adaptation to learn a noisy decision boundary. Right: Correct classifications are attained by excluding distant samples from the adaptation.
• Figure 3: The overview of our OSAA method to overcome negative transfer. Top: We illustrate the pipeline of the online selection mechanism. The source domain and the intermediate domain training samples are under selection. Bottom: We identify distant samples by observing training loss magnitudes (normalized) and conduct gradient selection by thresholding for the online gradient masking procedure.
• Figure 4: F1-score against different combinations of loss weights, with $\lambda_1$ and $\lambda_2$ ranging from 0.03 to 30 respectively.
• Figure 5: F1-score (vertical, bottom to top) versus selection portion ranging from 0 to 100$\%$ for the source (front axis, right to left) and intermediate (right axis, front to back) domain.