Application of Artificial Intelligence (AI) for Smart Automated Radiographic Testing (RT) Welding Anomaly Evaluation and Image Fingerprint Analysis
This paper outlines the development of Artificial Intelligence (AI) and Deep Learning (DL) algorithms in real-time RT image interpretation of welding anomalies and duplication detection, specifically focusing on weld joints in oil and gas construction and fabrication. Typically, weld joints for piping, pipelines, pressure vessels, and storage tanks are examined to ensure their quality and integrity. Radiography testing (RT) is a commonly used inspection method per or under welding codes and standards. RT result assessment relies on qualified radiographic interpreters (RI) and third-party inspectors to ensure weld quality and assurance. However, the current practice heavily depends on human skills and experience, which can lead to errors and inconsistencies in RT film interpretation, ultimately affecting project quality, timelines, and equipment reliability.
To address these issues, PETRONAS has developed algorithms that harnessed the power of AI and DL applications to enhance film image interpretation and streamline the review process. The algorithms are currently capable of detecting 11 common weld anomalies, achieving an impressive 91.0% overall defect detection accuracy and 84.7% classification accuracy. In addition, the model is also capable to analyze weld defect according to specific acceptance criteria such as API 1104, ASME B31.3, and ASME Section VIII codes and standards. Another pain point faced by industry is RT sub-standard practice performed by RT contractors during construction stage. It requires a lengthy review and assurance process to detect the frauds. PETRONAS has developed another model algorithm that enables the comparison of RT film images and weld profile to authenticate RT image fingerprints for fraud prevention by utilizing the AI. The model algorithm is developed based on Scale-Invariant Feature Transform (SIFT) to identify key points in the images.
Based on testing carried out by the PETRONAS team, the model can authenticate and detect radiographic test (RT) image fingerprints as low as 1% image similarity up to 100%. The success of both AI and DL model algorithms are due to extensive collaborative effort between PETRONAS and Universiti Teknologi Petronas (UTP), Malaysia.