AI-driven defect recognition and quality monitoring for additively manufactured sports protective equipment using nondestructive evaluation

Additive manufacturing (AM) enables the production of lightweight, complex, and customized sports protective equipment (SPE). However, conventional non-destructive evaluation (NDE) techniques face limitations in multimodal defect detection, defect characterization, quality assessment, and adaptive decision-making. This study proposes a Convolutional Attention Transformer Reinforcement Network (CA-TRN) for intelligent defect detection and quality inspection of AM-manufactured SPE. The proposed framework integrates MobileNetV3 for spatial feature extraction, Temporal Convolutional Networks (TCN) for temporal signal analysis, Convolutional Block Attention Module (CBAM) for enhanced feature representation, and Real-Time Detection Transformer (RT-DETR) for defect detection and localization. Deep reinforcement learning (DRL) supports defect severity assessment, quality evaluation, and adaptive manufacturing control, while Shapley Additive Explanations (SHAP) improves model interpretability. Multimodal NDE inputs, including visible-light images, X-ray computed tomography scans, sensor signals, laser intensity, melt-pool measurements, and layer-temperature data, enable comprehensive defect characterization. Experimental results demonstrate an accuracy of 96.2%, highlighting the effectiveness of CA-TRN in reliable, real-time, and explainable quality inspection. The proposed framework offers a promising approach to improving manufacturing quality, structural reliability, and safety in customized sports protective equipment.