Lightweight Transformer Network TDSPnet Detects Defects in Battery Welds

Researchers have developed TDSPnet, a lightweight point cloud classification model based on a Transformer architecture designed to detect welding defects in power battery terminals. Created to address the limitations of existing weld defect detection methods, the network focuses on the welding quality of new energy vehicle battery terminals, which is critical for the safety and reliability of the entire vehicle.

At the core of the system is a dynamic similarity mechanism that enhances the accuracy and efficiency of weld defect identification. The researchers designed a dynamic similarity attention module to strengthen algorithmic responses to morphology-sensitive weld regions, linking attention weights directly with physical defect characteristics such as missing solder, grooves, local height discontinuities, and surface-continuity breaks.

The structural framework of TDSPnet comprises three main components: a feature embedding module, the dynamic similarity attention module, and a classification module. Together, these modules process battery weld point clouds to perform segment-level defect judgments during inspection.

In experimental evaluations using the BWS dataset, provided by Hexagon and a power battery manufacturer, TDSPnet achieved an overall classification accuracy of 0.9555. The model recorded a miss rate of 4.52 percent and a false alarm rate of 4.37 percent during testing on the battery weld point cloud data.

When evaluated against alternative point cloud networks including PointNet, PCT, and DGCNN, TDSPnet demonstrated superior performance in convergence speed, training stability, detection accuracy, and model-inference efficiency. According to the research team, the findings provide a new solution for defect detection in power battery terminal welds and indicate promising engineering application value.

Subject of Research: Deep learning for power battery weld defect detection

Article Title: Research on power battery welding defect detection based on deep learning models with transformer architecture

Article References: Tang, Z., Feng, S., Hao, J., & Sun, C. (2026). Research on power battery welding defect detection based on deep learning models with transformer architecture. Discover Applied Sciences. https://doi.org/10.1007/s42452-026-09788-3

Image Credits: AI Generated

DOI: 10.1007/s42452-026-09788-3

Keywords: battery manufacturing, weld defect detection, point clouds, transformer architecture, deep learning, power batteries, quality control, TDSPnet, Research, power, battery, welding