Related Experiment Videos
A Span-Prior-Guided Explainable Multimodal Neural Network Method for Final-State Quality Inspection of Hairpin
Xiaopeng Chang1, Bangcheng Zhang1,2, Zhi Gao1
1School of Mechatronic Engineering, Changchun University of Technology, Changchun 130012, China.
Sensors (Basel, Switzerland)
|August 13, 2026
Summary
This study introduces SPIMA-Net, a multimodal neural network for inspecting hairpin winding quality. It effectively integrates visual data with geometric priors, achieving high accuracy and interpretability in industrial applications.
Area of Science:
- Industrial Automation
- Computer Vision
- Machine Learning
Background:
- Current hairpin winding quality inspection lacks multimodal methods integrating geometric constraints and interpretability.
- Existing methods struggle with combining visual data and mechanical properties for accurate defect detection.
Purpose of the Study:
- To develop a novel multimodal neural network for final-state quality inspection of three-dimensional stamped hairpin windings.
- To enhance inspection accuracy and provide engineering interpretability by integrating mechanical geometric constraints and prior-guided fusion.
Main Methods:
- Proposed SPIMA-Net, a span-prior-guided explainable multimodal neural network.
- Utilized final-state images and geometric priors (span measurements, model-type) as inputs.
- Implemented a visual branch, a span branch, and a span-prior-assisted gating mechanism for feature fusion.
Main Results:
- SPIMA-Net achieved 98.14% accuracy, 96.55% F1-score, and 0.9983 AUC on the test set.
- Demonstrated significant improvements in nonconforming-class F1-score compared to existing models.
- Interpretability analysis identified span openings, end profiles, and local abnormal regions as key focus areas.
Conclusions:
- The proposed neuro-mechanical fusion approach offers high discriminative performance for hairpin winding quality inspection.
- SPIMA-Net provides valuable engineering interpretability, identifying key geometric factors affecting quality.
- The method is effective for the investigated industrial dataset, advancing automated quality control.