Uncertainty-aware open-set transfer diagnosis of rotating machinery via class-wise adaptive decision boundaries and enhanced target clustering

Open-set transfer diagnosis of rotating machinery aims to classify target samples from shared fault classes while rejecting unseen target classes under domain shifts. Although open-set domain adaptation has been introduced to handle unseen fault classes in the target domain, reliable separation between known and unknown samples remains challenging. This challenge becomes more pronounced when unknown samples are close to known-class distributions or when different known classes exhibit distinct uncertainty patterns, making a uniform rejection criterion insufficient for robust open-set diagnosis. To address these problems, this paper proposes an open-set domain adaptation with class-wise adaptive entropy threshold and enhanced target clustering (OSDACE) method for open-set transfer diagnosis of rotating machinery. First, an entropy-minimized prototype-neighbor alignment mechanism is integrated with a masked Transformer encoder to achieve robust feature extraction and unsupervised feature alignment. Specifically, the masked Transformer encoder extracts robust fault representations from vibration signals, while the alignment mechanism exploits source prototypes and target neighborhood structures to promote cross-domain alignment. Second, a class-wise adaptive entropy threshold is designed to assign class-specific rejection boundaries according to intra-class entropy statistics and uncertainty variations, thereby improving the discrimination between known and unknown samples. Finally, a confidence-aware and feature-attention-weighted KL clustering strategy is further introduced to compact target-domain clusters and refine decision boundaries. Experimental results on gearbox and bearing datasets demonstrate that OSDACE achieves the best overall performance among the compared methods, with improved known-class recognition and unknown-class rejection.

Authors

Institutions

Publication Details

Journal
Measurement Science and Technology
Published
2026-11-19
DOI
https://doi.org/10.1088/1361-6501/aec46b
Primary Topic
Machine Fault Diagnosis Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Uncertainty-aware open-set transfer diagnosis of rotating machinery via class-wise adaptive decision boundaries and enhanced target clustering

Yulei Gao, Lingli Cui, Dongdong Liu, Zhijun Jin
Measurement Science and Technology
Machine Fault Diagnosis Techniques
article

Uncertainty-aware open-set transfer diagnosis of rotating machinery via class-wise adaptive decision boundaries and enhanced target clustering

Yulei Gao, Lingli Cui, Dongdong Liu, Zhijun Jin
article en

Abstract

Open-set transfer diagnosis of rotating machinery aims to classify target samples from shared fault classes while rejecting unseen target classes under domain shifts. Although open-set domain adaptation has been introduced to handle unseen fault classes in the target domain, reliable separation between known and unknown samples remains challenging. This challenge becomes more pronounced when unknown samples are close to known-class distributions or when different known classes exhibit distinct uncertainty patterns, making a uniform rejection criterion insufficient for robust open-set diagnosis. To address these problems, this paper proposes an open-set domain adaptation with class-wise adaptive entropy threshold and enhanced target clustering (OSDACE) method for open-set transfer diagnosis of rotating machinery. First, an entropy-minimized prototype-neighbor alignment mechanism is integrated with a masked Transformer encoder to achieve robust feature extraction and unsupervised feature alignment. Specifically, the masked Transformer encoder extracts robust fault representations from vibration signals, while the alignment mechanism exploits source prototypes and target neighborhood structures to promote cross-domain alignment. Second, a class-wise adaptive entropy threshold is designed to assign class-specific rejection boundaries according to intra-class entropy statistics and uncertainty variations, thereby improving the discrimination between known and unknown samples. Finally, a confidence-aware and feature-attention-weighted KL clustering strategy is further introduced to compact target-domain clusters and refine decision boundaries. Experimental results on gearbox and bearing datasets demonstrate that OSDACE achieves the best overall performance among the compared methods, with improved known-class recognition and unknown-class rejection.

Measurement Science and Technology
Beijing University of Technology (CN)
Machine Fault Diagnosis Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon