Optimal Transducer Placement for Deep Learning-Based Non-Destructive Evaluation

Abstract

In this study, the Convolution Neural Network (CNN) algorithm is applied for non-destructive evaluation of aluminum panels. A method of classifying the locations of defects is proposed by exciting an aluminum panel to generate ultrasonic Lamb waves, measuring data with a sensor array, and then deep learning the characteristics of 2D imaged, reflected waves from defects. For the purpose of a better performance, the optimal excitation location and sensor locations are investigated. To ensure the robustness of the training model and extract the feature effectively, experimental data are collected by slightly changing the excitation frequency and shifting the location of the defect. The high classification accuracy for each defect location can be achieved. It is found that the proposed algorithm is also successfully applied even when a bar is attached to the panel.