چکیده مقاله
The ultrasonic non destructive evaluation NDE is a key technique for assessing internal material integrity without causing damage However, manual interpretation of ultrasonic A scan signals remains challenging due to their complex echo patterns and strong dependence on operator experience This study presents an interpretable, envelope based machine learning framework for automated ultrasonic defect detection and classification The approach combines physics based signal conditioning—such as pulse template removal, RMS normalization, and envelope extraction—with statistical and spectral feature analysis to generate reliable representations of ultrasonic responses The proposed framework was trained and validated on laboratory A scan data obtained from stainless steel calibration blocks containing both sound and artificially induced defects Results demonstrate that the extracted envelope features enable accurate and stable defect identification across multiple classification algorithms The workflow maintains low computational complexity while achieving strong generalization, making it well suited for practical deployment in real time or embedded ultrasonic inspection systems This research highlights the potential of signal driven, interpretable machine learning approaches as a scalable foundation for intelligent ultrasonic evaluation Future studies will expand the dataset and explore industrial materials and field conditions to further assess robustness and applicability
کلیدواژهها
نویسندگان
Sina Sodagar
Petroleum University of Technology
Abdollah Cheraghi
Petroleum University of Technology
شیوه ارجاع
Sodagar, Sina and Cheraghi, Abdollah,1404,Envelope-Based Machine Learning Framework for Automated Ultrasonic A-Scan Defect Detection and Classification,The First National Conference on Applied Technologies in Mechanical Engineering (ATME2025),Ahvaz
ارائهشده در
مجموعه مقالات اولین کنفرانس ملی فناوری های کاربردی در مهندسی مکانیک27 آبان 1404 · اهواز