A Multi-Scale Mamba–Transformer Model with Physical Parameter Interaction for Cross-Domain Fault Diagnosis of Rolling Bearings

Rolling bearings, as the core load-bearing and transmission components of rotating machinery, directly determine the reliability and safety of equipment operation; however, their cross-domain fault diagnosis is challenged by the scarcity of labeled fault samples, time-varying noise statistics accompanied by impulsive outliers, and the disconnection between physical prior knowledge and data-driven feature extraction. To address these challenges, this paper proposes a multi-scale Mamba–Transformer model with physical parameter interaction (MSMT-PPI). At the physical parameter perception level, a Huber–Sage adaptive robust iterated extended Kalman filter (HS-IEKF) is developed by integrating the Huber robust kernel and Sage–Husa exponentially weighted noise estimation into the iterated extended Kalman filter, with the fault size modeled as a Wiener-process-driven dynamic evolution state, thereby achieving robust physical parameter inversion and joint tracking of fault degradation trends under time-varying noise. At the feature extraction level, a dual-branch architecture comprising a Mamba global branch and a local window attention branch is constructed, and a triple physics–data coupling mechanism—physics-consistency regularization, uncertainty-guided feature attention modulation, and physics-conditional selective scan modulation—is designed to propagate the estimated physical parameters and their uncertainties into the feature-learning process. Experimental results on four cross-domain diagnosis tasks demonstrate that the proposed model achieves an average diagnosis accuracy of 96.44%, outperforming the standard Transformer by 12.29 percentage points, while reducing the parameter estimation RMSE by 47.2% relative to the standard IEKF and the inference latency by approximately 62% relative to the standard Transformer. These results verify that bidirectional physics–data interaction with quantified uncertainty propagation provides an accurate, robust, and real-time solution for industrial online condition monitoring of rolling bearings.

Authors

Institutions

Publication Details

Journal
Sensors
Published
2026-11-18
DOI
https://doi.org/10.3390/s26227285
Primary Topic
Machine Fault Diagnosis Techniques
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Multi-Scale Mamba–Transformer Model with Physical Parameter Interaction for Cross-Domain Fault Diagnosis of Rolling Bearings

Jiachun Li, 唐云, Qi Tang, Lei Lin et al.
6 links
Sensors
Machine Fault Diagnosis Techniques
article

A Multi-Scale Mamba–Transformer Model with Physical Parameter Interaction for Cross-Domain Fault Diagnosis of Rolling Bearings

Jiachun Li, 唐云, Qi Tang, Lei Lin, Xianfu Zeng
article en

Abstract

Rolling bearings, as the core load-bearing and transmission components of rotating machinery, directly determine the reliability and safety of equipment operation; however, their cross-domain fault diagnosis is challenged by the scarcity of labeled fault samples, time-varying noise statistics accompanied by impulsive outliers, and the disconnection between physical prior knowledge and data-driven feature extraction. To address these challenges, this paper proposes a multi-scale Mamba–Transformer model with physical parameter interaction (MSMT-PPI). At the physical parameter perception level, a Huber–Sage adaptive robust iterated extended Kalman filter (HS-IEKF) is developed by integrating the Huber robust kernel and Sage–Husa exponentially weighted noise estimation into the iterated extended Kalman filter, with the fault size modeled as a Wiener-process-driven dynamic evolution state, thereby achieving robust physical parameter inversion and joint tracking of fault degradation trends under time-varying noise. At the feature extraction level, a dual-branch architecture comprising a Mamba global branch and a local window attention branch is constructed, and a triple physics–data coupling mechanism—physics-consistency regularization, uncertainty-guided feature attention modulation, and physics-conditional selective scan modulation—is designed to propagate the estimated physical parameters and their uncertainties into the feature-learning process. Experimental results on four cross-domain diagnosis tasks demonstrate that the proposed model achieves an average diagnosis accuracy of 96.44%, outperforming the standard Transformer by 12.29 percentage points, while reducing the parameter estimation RMSE by 47.2% relative to the standard IEKF and the inference latency by approximately 62% relative to the standard Transformer. These results verify that bidirectional physics–data interaction with quantified uncertainty propagation provides an accurate, robust, and real-time solution for industrial online condition monitoring of rolling bearings.

SensorsVol. 26(22)
Guizhou University (CN), Dalian University of Technology (CN)
Guizhou Science and Technology Department
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