AI-driven non-destructive inspection in smart manufacturing: A review of techniques for structural integrity and process quality enhancement

Non-Destructive Evaluation (NDE) is essential for ensuring product quality, structural integrity, and process reliability in modern manufacturing without disrupting production. However, conventional NDE methods often involve operator-dependent interpretation, large and heterogeneous datasets, and limited integration with automated manufacturing systems. Machine Learning (ML) has increasingly transformed NDE into an intelligent, data-driven approach by enabling automated defect detection, classification, segmentation, characterization, and real-time decision-making. This paper presents a comprehensive review of ML applications across major NDE modalities, including Ultrasonic Testing (UT), Phased Array Ultrasonic Testing (PAUT), Radiographic Testing (RT), Infrared Thermography (IRT), Acoustic Emission (AE), Eddy Current Testing (ECT), and vision-based inspection, with emphasis on manufacturing quality assurance. Classical ML techniques, including Support Vector Machines, Random Forests, k-Nearest Neighbors, and Gaussian Mixture Models, are reviewed alongside deep-learning architectures such as Convolutional Neural Networks, U-Net, CNN–LSTM, and transformer-based models. The review critically examines challenges related to limited and imbalanced datasets, domain generalization, interpretability, computational requirements, uncertainty, and standardization. Emerging directions, including multimodal data fusion, physics-informed ML, digital twins, edge AI, autonomous inspection, and explainable AI, are discussed as key enablers for reliable and adaptive NDE systems. Overall, ML-driven NDE provides a pathway toward intelligent, predictive, and increasingly autonomous quality assurance aligned with Industry 4.0 and sustainable manufacturing.