Classification for Ultrasound Welding Joints Based on PCA and Improved Adaptive PSO-SVM

Ultrasonic welding is widely employed in joining metal material, where the quality of the welding joints directly affects the strength and durability of the components. We propose a visual classification method for classifying welding joints, enabling feedback control of ultrasonic welding parameters to improve joints quality. The Principal Component Analysis (PCA) is utilized to extract features from welding joints, followed by experimental testing to determine the optimal feature length for the classifier. Additionally, an adaptive particle swarm optimization (PSO) with a decay factor is used to optimize the support vector machine (SVM) for welding joints classification. Experimental results demonstrate that the Gaussian radial basis kernel function (RBF) reaches 86% accuracy on the training set and 85% accuracy on the test set, which achieves better classification performance compared to linear kernel function and Multi-Kernel Learning Method (MKL). Compared to the original SVM, our proposed method achieves superior classification accuracy and reduces computational time.