Forecasting Defect Classification in Pressure Vessel Welds with Nondestructive Testing Indicators

Abstract

The structural integrity of pressure vessels is critical for maintaining safety and operational efficiency across various industrial sectors, including chemical processing, energy generation, and aerospace. Welding processes, while essential for the fabrication of these vessels, inherently introduce anomalies that can compromise structural stability. This paper presents a comprehensive investigation into predicting defect classification from nondestructive testing indicators using advanced experimental analysis in pressure vessel welds. By employing a rigorous methodology that integrates ultrasonic testing signal acquisition with machine learning classification algorithms, this study aims to transition defect evaluation from subjective human interpretation to objective, data-driven analysis. Experimental specimens containing artificially induced, well-characterized defects such as porosity, slag inclusion, lack of fusion, and cracks were subjected to exhaustive automated scanning. The extracted nondestructive testing indicators, encompassing time-domain, frequency-domain, and time-frequency domain features, were analyzed to determine their predictive power. The results demonstrate that specific combinations of signal features significantly enhance the discriminatory capability of classification models, leading to high diagnostic accuracy. Furthermore, this research elucidates the physical correlations between defect morphology and acoustic wave propagation characteristics, providing a robust theoretical foundation for the observed statistical trends. The findings offer profound implications for automated quality assurance systems, promising improved reliability and reduced downtime in industrial applications.