비파괴검사 데이터의 머신러닝을 이용한 알루미늄 합금의 강도 추정

초록

The increasing demand for lightweight aluminum alloys in advanced industrial products has necessitated the development of monitoring techniques for material strength using non-destructive testing (NDT) methods to ensure quality control. In this study, a machine learning model was developed to estimate the strength of aluminum alloys using NDT parameters known to be related to material strength, such as ultrasonic longitudinal/transverse velocity, attenuation coefficient, nonlinearity parameter, and eddy current electrical conductivity. The training data set consisted of NDT parameters obtained from more than 400 specimens with diverse strength distributions, along with tensile test data. A dedicated automated measurement system was employed to enhance the reliability of NDT parameter measurements. The estimated strength achieved more than 90% accuracy when a +/- 20 MPa interval accuracy was applied to the ground-truth strength. As data accumulation continues in the future, the performance of the proposed model is expected to improve further. Considering that destructive tensile tests possess an uncertainty of approximately 10%, the proposed technique offers an alternative to destructive testing, thereby establishing itself as a promising technology.

키워드