Quality Classification of Ultrasonically Welded Automotive Wire Harness Terminals by Ultrasonic Phased Array

Abstract

An ultrasonic nondestructive evaluation technique is proposed for ultrasonically welded joints of multi-strand copper cables in automobile wire harness terminals. The 32/128 ultrasonic phased array system is used to acquire the complete matrix data of the pulse-echo of the wire harness joints. The eigenvalues of the time, frequency, and time-frequency domains are extracted, and the wire harness joint quality is classified by machine learning. Firstly, 28 wire harness terminal joint samples were prepared 14 under different welding parameters; 14 were okay (OK), and were negative (NG). Then a linear array probe 5L32-0.6 × 10 is used to collect and preprocess the complete matrix data in these joints, and 11 200 echo signals are obtained. A principal component analysis algorithm was employed for data dimensionality reduction and denoising. Finally, machine learning algorithms were used to train and verify the model. The accuracy and performance of the traditional algorithms such as Logistic Regression (LR), K-Nearest Neighbor (KNN), Decision Tree (DT), Naive Bayes (NB), Support Vector Machine (SVM), and Neural Network (NN) were compared. The KNN and NN perform well in this study. In the test set, the accuracy of KNN and NN reached 90%. The study showed that echo features could effectively identify joint quality.