CERTIFICATION AND TRUSTWORTHY DEPLOYMENT OF EXPLAINABLE AND PHYSICS-INFORMED ARTIFICIAL INTELLIGENCE FOR AEROSPACE NON-DESTRUCTIVE TESTING

Artificial intelligence (AI) is increasingly used in aerospace non-destructive testing (NDT) for composite materials, additively manufactured components, and automated inspection systems. Despite promising detection performance, the use of AI in certified aerospace operations remains limited. The main barrier is not accuracy, but the lack of assurance related to safety, traceability, uncertainty management, and regulatory compliance. This review examines explainable AI (XAI), physics-informed modelling, data provenance, and validation practices in the context of aerospace certification requirements. Unlike previous surveys that focus primarily on algorithm performance, this study analyses how AI-based NDT systems align with airworthiness standards and certification expectations. Based on a structured review of literature covering ultrasonic, radiographic, and optical inspection, the selected studies are categorised according to inspection modality, data source, explainability method, physics integration, and validation maturity. The findings reveal recurring trade-offs between model performance and interpretability. In particular, explanation techniques applied after model training are often insufficient to meet safety-critical assurance requirements. Physics-informed and hybrid approaches show greater potential for improving model reliability, robustness, and consistency with known physical behaviour. The review also identifies misalignment between generic AI development practices and aerospace NDT requirements, especially in relation to data traceability, probability-of-detection metrics, and certification documentation. Drawing on these insights, the paper proposes design and validation guidelines that link modelling choices, uncertainty reporting, and human oversight mechanisms to certification readiness. The goal is to support regulators, engineers, and operators in advancing the trustworthy deployment of AI-based aerospace NDT systems within certified operational environments.