Use of Machine Learning to Identify Adhesive Joints Defects by Using Ultrasonic Waves

Defects in adhesive joints are an important issue in the construction of space structures. In this paper, using lamp waves, suitable properties have been obtained to identify the size and position of the defects of the adhesive joints. Using finite element simulations, the effect of the defect on the propagation of the lamp waves has been investigated. Simulations have been performed for three different adhesive thicknesses, three different sizes of circular defects in 9 different positions, and the effect of each of them on the wave passing through the joint has been investigated. The signals obtained from the faulty connections were compared with the signal obtained from the healthy connection and the desired area was isolated from the total received signal for further analysis. Proper and correct separation of defects requires finding suitable characteristics for it. Therefore, 34 features were examined to differentiate and separate defects. Then, the neural network was used to provide the basis for creating appropriate patterns for the separation of defects. The percentage of correct detection of neural network for adhesive thickness separation was 93.8%, for defect area separation in terms of size 100% and for defect position separation in X and Y axes were 96.1 and 95.1%, respectively. The obtained results show the efficiency of the IDE method and the features selected to distinguish the defects of such connections.