AI-driven Prediction of Structural Steel Mechanical Properties Using Ultrasound and Eddy Current Testing

Reuse of structural steel can reduce embodied carbon but depends on reliable knowledge of the mechanical properties of reclaimed members. Conventional assessment relies on destructive tensile tests, which are costly, time-consuming, and difficult to scale. This study evaluates the use of ultrasound testing (UT), eddy current testing (ECT), and data-driven models to predict mechanical properties and classify the strength grade of salvaged structural steels. Specimens from reclaimed beam flanges and webs were tested in tension and measured using UT and ECT. Random Forest models were used for strength-grade classification and property prediction. The ECT-based classifier achieved an accuracy of 0.92, while the UT-based classifier reached about 0.82. The UT-based yield strength model achieved a mean absolute percentage error of 5.86%, demonstrating that yield strength can be estimated with reasonable accuracy without prior knowledge of steel grade.