Automatic defect detection in radiographic images of casting components: a comprehensive review

Casting is a critical manufacturing process widely employed in industries such as nuclear, automotive, petrochemical and aerospace. Ensuring the structural integrity and reliability of casting products is paramount, as undetected defects may lead to catastrophic failures in service. Radiographic testing (RT) is one of the most extensively applied non-destructive testing (NDT) techniques for casting inspection, due to its ability to reveal internal flaws such as porosity, shrinkage cavities, cracks and inclusions. However, manual interpretation of radiographic images requires a lot of time, is subjective and is prone to human error, especially in complex castings with overlapping defect patterns. With recent advances in digital image processing, machine learning (ML) and deep learning (DL), automatic defect detection in radiographic images has emerged as a powerful alternative to traditional film interpretation. This review provides a comprehensive overview of state-of-the-art research in the field of automatic defect recognition for casting products. It examines classical machine learning approaches and modern deep learning frameworks, highlighting their respective advantages, limitations and industrial applications. Benchmark datasets, international standards and performance evaluation metrics are also presented to evaluate the effectiveness of automated defect detection systems. Finally, challenges such as limited labelled datasets, noise artefacts and regulatory acceptance are discussed and future research directions are proposed. This review consolidates current advances while identifying gaps and opportunities for future developments in automatic defect detection during radiographic testing of casting products.