Deep Learning for Non-Destructive Testing and Evaluation of Composite Materials: State of the Art, Challenges, and Perspectives

Composite materials are widely used in aerospace and other fields, but defects and damage arising from manufacturing and service pose threats to structural integrity. Traditional non-destructive testing and evaluation (NDT&E) methods exhibit limitations in interpretation efficiency, detection sensitivity, and multimodal fusion evaluation. Deep learning (DL) has shown great potential in this area through its automatic feature extraction and pattern recognition capabilities. This paper systematically reviews the state-of-the-art of DL-enabled NDT&E of composite materials. First, the evolution and technical characteristics of DL architectures are summarized chronologically. Secondly, organized by the NDT&E task flow, representative studies are elaborated from four aspects—data preprocessing, defect detection, defect evaluation, and human-in-the-loop. Current deficiencies and challenges are highlighted. Finally, based on the level of DL autonomy in NDT&E, we propose a five-stage progressive path: AI-Assistant, AI-Copilot, AI-Agent, AI-Autonomy, and AI-Complete (AI-ACAAC), aiming to provide a roadmap for intelligent development in composite inspection and evaluation.