Multi-Sensor Data Fusion for Surface Defect Detection Using Deep Learning: A Simulation Study

Abstract

Surface defect detection is a critical quality control task in precision manufacturing of optical components and thermal management systems. Conventional single-sensor inspection methods face significant challenges in detecting subtle defects beneath surface geometry effects, particularly on non-flat surfaces where shadowing, self-radiation, and complex emissivity distributions interfere with defect signatures. This study proposes a multi-sensor data fusion framework that integrates thermal imaging data with fringe projection profilometry (FPP) data for robust surface defect detection using a deep convolutional neural network. The framework is developed in the context of the 4D thermal imaging methodology established by Huang, Yang, and Zhu (2023) and the deep learning-enhanced optical metrology approach demonstrated by Huang, Tang, Liu, and Huang (2026). Specifically, a dual-branch 3D convolutional network processes co-registered thermal and depth image pairs, extracting both thermal anomaly features and geometric surface features for unified defect classification and localization. A simulation dataset modeling diverse defect types (cracks, pits, delamination, contamination) on optical component surfaces is constructed for training and evaluation. Simulation results demonstrate that the proposed fusion framework achieves a defect detection accuracy of 96.3% and a mean Intersection over Union (mIoU) of 81.7%, representing a 12.4% accuracy improvement and a 15.8% mIoU improvement over single-sensor baselines. The approach shows particularly strong performance on non-flat surfaces where single-sensor methods struggle. This work provides a data-driven pathway toward automated, reliable surface defect detection in precision manufacturing environments.