From Inspection Report to Risk Scores: A Data-Driven Automated Framework for Fine-Grained Health Assessment of Pressure Vessels

Abstract Health assessment of pressure vessels is essential for predictive maintenance and accident prevention. Existing methods face certain limitations: those relying solely on real-time monitoring often lack comprehensive inspection information, while those based on inspector expertise and discrete safety condition levels (1-5) suffer from limited granularity and objectivity. To overcome these challenges, this study proposes a data-driven framework for automated and fine-grained health assessment based on periodic measurement-derived inspection reports. Although underutilized in practice, such reports contain rich health-related information of equipment, including basic equipment information and records of surface and interior defects. A deep learning model has been developed to identify safety condition levels (3 or above vs. below 3) from raw inspection data with an impressive accuracy of 98%. The results highlight both the practical reliability and the academic value of using periodic measurement-derived inspection reports for health assessment. Building on this foundation, a nonlinear transformation inspired by credit scoring converts predicted probabilities into continuous health risk scores ranging from 1 to 100. Empirical analysis shows that this method enhances assessment granularity by nearly 20-fold compared with conventional safety condition levels. The proposed framework provides an automated, reliable, and individualized tool that reduces reliance on expert judgment. This will advance the research of effective utilization of measurement-derived information. It holds promise for improving risk prevention across a broad range of safety-critical equipment.