Cross-condition hybrid dual-path adaptation network for fault diagnosis of rotating machinery

Abstract Under complex variable-speed operating conditions, the condition monitoring of rotating machinery is severely hampered by speed-induced temporal distortion of fault impulse sequences and feature distribution discrepancies among diverse operating states. To tackle these challenging issues, this study proposes a novel Cross-Condition Hybrid Dual-Path Adaptation Network (CHDAN). First, relative shaft synchronous resampling (RSSR) is employed to correct speed-induced temporal distortions, and noise injection is integrated during training to enhance signal consistency and robustness, thereby establishing a unified angular-domain reference. Taking the aligned angular-domain signals as input, a Channel-Recalibrated Multi-Scale Tokenizer is designed to extract and fuse local multi-scale fault information through parallel convolutions with multiple receptive fields and adaptive channel re-weighting, thereby generating compact and discriminative token representations. To capture long-range dependencies across these local tokens, a hybrid dual-path transformer encoder (HybDualFormer) is constructed, which embeds physical fault-order information matched with the angular reference into the data-driven feature space for cross-speed feature alignment, while adopting Grouped-Query Attention to reduce parameter redundancy in correlation modeling. Building upon the resulting correlation matrix, a Dual-Path Gated Fusion Attention mechanism is developed to decouple discriminative enhancement and retentive accumulation paths, enabling the joint extraction of high-energy dominant impacts and low-energy weak responses. Finally, by retaining the front-end generalizable features, a progressive local unfreezing transfer learning strategy is implemented to achieve efficient parameter adaptation and manifold reconstruction with limited target-domain samples. Cross-speed experiments on a laboratory-built RV reducer and a publicly available aero-engine dual-rotor casing dataset confirm that CHDAN outperforms advanced diagnostic methods in classification accuracy and cross-domain adaptive generalization capability, providing a reliable end-to-end diagnostic paradigm for rotating equipment.

Authors

Institutions

Publication Details

Journal
Measurement Science and Technology
Published
2026-11-17
DOI
https://doi.org/10.1088/1361-6501/aec31b
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

Cross-condition hybrid dual-path adaptation network for fault diagnosis of rotating machinery

Linlin Xue, Huageng Luo, Zhuo Xue, Fei Zhang et al.
3 links
Measurement Science and Technology
Machine Fault Diagnosis Techniques
article

Cross-condition hybrid dual-path adaptation network for fault diagnosis of rotating machinery

Linlin Xue, Huageng Luo, Zhuo Xue, Fei Zhang, Qianyin Zheng
article en

Abstract

Abstract Under complex variable-speed operating conditions, the condition monitoring of rotating machinery is severely hampered by speed-induced temporal distortion of fault impulse sequences and feature distribution discrepancies among diverse operating states. To tackle these challenging issues, this study proposes a novel Cross-Condition Hybrid Dual-Path Adaptation Network (CHDAN). First, relative shaft synchronous resampling (RSSR) is employed to correct speed-induced temporal distortions, and noise injection is integrated during training to enhance signal consistency and robustness, thereby establishing a unified angular-domain reference. Taking the aligned angular-domain signals as input, a Channel-Recalibrated Multi-Scale Tokenizer is designed to extract and fuse local multi-scale fault information through parallel convolutions with multiple receptive fields and adaptive channel re-weighting, thereby generating compact and discriminative token representations. To capture long-range dependencies across these local tokens, a hybrid dual-path transformer encoder (HybDualFormer) is constructed, which embeds physical fault-order information matched with the angular reference into the data-driven feature space for cross-speed feature alignment, while adopting Grouped-Query Attention to reduce parameter redundancy in correlation modeling. Building upon the resulting correlation matrix, a Dual-Path Gated Fusion Attention mechanism is developed to decouple discriminative enhancement and retentive accumulation paths, enabling the joint extraction of high-energy dominant impacts and low-energy weak responses. Finally, by retaining the front-end generalizable features, a progressive local unfreezing transfer learning strategy is implemented to achieve efficient parameter adaptation and manifold reconstruction with limited target-domain samples. Cross-speed experiments on a laboratory-built RV reducer and a publicly available aero-engine dual-rotor casing dataset confirm that CHDAN outperforms advanced diagnostic methods in classification accuracy and cross-domain adaptive generalization capability, providing a reliable end-to-end diagnostic paradigm for rotating equipment.

Measurement Science and Technology
Xiamen University (CN)
Industry, innovation and infrastructure
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