YOLOv4 Object Detection Model for Nondestructive Radiographic Testing in Aviation Maintenance Tasks

Abstract

to initiate your request. See also AIAA Rights and Permissions www.aiaa.org/randp.

Nondestructive Testing of Metal Cracks: Contemporary Methods and Emerging Challenges

“…Chen et al [17] combined the YOLOv4 object inspection model with RT image-based NDT techniques to improve the inspection efficiency of aero structural and engine components. A dataset is preprocessed by marking and classifying defects in RT images, including maintenance record images of civil aircraft fuselages and engines.…”

“…The morphology and location information of cracks can be obtained by the leakage magnetic field. In the early days, a series of methods were explored based on fundamental physical principles, such as eddy current testing (ECT) [10][11][12], magnetic particle testing (MPT) [13,14], penetrant testing (PT), ultrasonic testing (UT), and radiographic testing (RT) [15][16][17], which have been extensively utilized for the detection of metal cracks, among other defects. These methods have been refined over time and honed to address specific challenges posed by different materials and operating environments.…”

Augmented Reality in Maintenance—History and Perspectives

“…Similarly, Zhi-Hao Chenÿ presented an automatic mechanism based on YOLOv4 to recognize potential cracks during nondestructive testing (NDT) of civic aeroengines. In their paper [77], training and testing images of task datasets were taken from the archives of the civic aircraft fuselage and engine repair records, theb various defects withn this dataset were categorized. A simple experiment compared different Faster R-CNN (Region-Based Convolutional Neural Network) and YOLOv4 framework models.…”