Lightweight fault diagnosis of automotive gearboxes via CWT-MobileNetV2 collaborative transfer learning

The difficulty of representing non-stationary fault signals in automotive transmission system health monitoring, the high deployment cost of heavyweight models under onboard resource constraints, and the limited availability of labeled samples under target operating conditions are main challenges for fault diagnosis of automotive gearboxes. To address them, this paper proposes a lightweight collaborative transfer learning framework, termed CWT-MobileNetV2-TL, for small-sample fault diagnosis. The framework integrates three coordinated components: time-frequency representation enhancement, lightweight backbone construction, and transfer adaptation optimization. Specifically, continuous wavelet transform (CWT) is employed to convert one-dimensional vibration signals into time-frequency images, while MobileNetV2 is adopted as a lightweight feature extraction backbone. The framework is combined with structural and training optimization strategies for small-sample transfer to improve transfer stability and generalization under limited-sample conditions. Experimental results on public gearbox datasets show that the proposed method achieves 99.52% accuracy on the same-condition diagnosis task in Dataset A. Under the complex operating conditions of Dataset B, the CWT-based representation improves diagnostic accuracy by 6.97 percentage points over STFT, reaching 99.66%. In the cross-condition small-sample transfer task from A to B, the accuracy reaches 92.29%. In the small-sample diagnosis task, the accuracy remains 92.63% when only 5% of the training data are used. In addition, the model contains only 2.21 million parameters, with a single-sample inference time of 15.17 ms on the same CPU platform. These results indicate that the proposed method achieves a favorable balance among diagnostic performance, data efficiency, and model complexity on public gearbox benchmarks, and provides a useful methodological reference for fault diagnosis of automotive transmission systems in onboard resource-constrained scenarios.

Authors

Institutions

Publication Details

Journal
Proceedings of the Institution of Mechanical Engineers Part D Journal of Automobile Engineering
Published
2026-11-16
DOI
https://doi.org/10.1177/09544070261502672
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

Lightweight fault diagnosis of automotive gearboxes via CWT-MobileNetV2 collaborative transfer learning

Zhang Jie, Lu Kaipeng, Yang Yu, Bohua Cai et al.
2 links
Proceedings of the Institution of Mechanical Engineers Part D Journal of Automobile Engineering
Machine Fault Diagnosis Techniques
article

Lightweight fault diagnosis of automotive gearboxes via CWT-MobileNetV2 collaborative transfer learning

Zhang Jie, Lu Kaipeng, Yang Yu, Bohua Cai, Zhu Zhifeng, Yao Yong, Yang Chen
article en

Abstract

The difficulty of representing non-stationary fault signals in automotive transmission system health monitoring, the high deployment cost of heavyweight models under onboard resource constraints, and the limited availability of labeled samples under target operating conditions are main challenges for fault diagnosis of automotive gearboxes. To address them, this paper proposes a lightweight collaborative transfer learning framework, termed CWT-MobileNetV2-TL, for small-sample fault diagnosis. The framework integrates three coordinated components: time-frequency representation enhancement, lightweight backbone construction, and transfer adaptation optimization. Specifically, continuous wavelet transform (CWT) is employed to convert one-dimensional vibration signals into time-frequency images, while MobileNetV2 is adopted as a lightweight feature extraction backbone. The framework is combined with structural and training optimization strategies for small-sample transfer to improve transfer stability and generalization under limited-sample conditions. Experimental results on public gearbox datasets show that the proposed method achieves 99.52% accuracy on the same-condition diagnosis task in Dataset A. Under the complex operating conditions of Dataset B, the CWT-based representation improves diagnostic accuracy by 6.97 percentage points over STFT, reaching 99.66%. In the cross-condition small-sample transfer task from A to B, the accuracy reaches 92.29%. In the small-sample diagnosis task, the accuracy remains 92.63% when only 5% of the training data are used. In addition, the model contains only 2.21 million parameters, with a single-sample inference time of 15.17 ms on the same CPU platform. These results indicate that the proposed method achieves a favorable balance among diagnostic performance, data efficiency, and model complexity on public gearbox benchmarks, and provides a useful methodological reference for fault diagnosis of automotive transmission systems in onboard resource-constrained scenarios.

Proceedings of the Institution of Mechanical Engineers Part D Journal of Automobile Engineering
Anhui University of Technology (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