Recognition And Classification Of Ultrasonic Micro Wire Joint Image

In recent years,AI based on statistical learning has been developing rapidly,and the application of AI technology to industrial production has become an increasingly popular research direction.Compared with traditional welding technology,ultrasonic metal welding has many advantages,such as low energy consumption,fast speed,high precision and simple operation.Ultrasonic metal welding system is composed of multiple systems.During the welding process,the coordination of each system need to be realized,otherwise the welding quality of metal solder joints will be affected.Therefore,the issue aims at the automation of thick aluminum wire ultrasonic metal welding of the ultrasonic system instability caused by the effect of welding instability problems,explore the use of machine vision and statistical learning techniques,quality testing and research of a joint positioning method of ultrasonic metal solder joints.Aiming at the positioning of thick aluminum wire ultrasonic metal joints and extracting joint image,an algorithm which combines Canny edge histogram equalization and image morphology is proposed.In order to accelerate the algorithm,the source images are reduced by using gaussian pyramid before the joint image processing,and then the position coordinates of the obtained solder joints are amplified in equal proportion.Finally,in order to improve the robustness of the algorithm,an algorithm for joint position screening is proposed.The experimental results show that the algorithm has good recognition accuracy for single ultrasonic metal solder joint image and multi metal solder joint image.Aiming at the quality classification of thick aluminum wire ultrasonic metal joint,PCA is used as the preprocessing algorithm for extracting solder joint image,and SVM is used as the joint classification algorithm.Because the dimension of the joint image is large,PCA is used to reduce the dimension of the joint image.In order to determine the appropriate reduction dimension,PCA is used to reconstruct the joint image.At the same time,according to the dimension of a single sample and the size of the dataset,SVM is used as the classifier algorithm.In order to determine the kernel function used in the SVM,using the PCA again for the visualization of dataset,and finally determining the Gauss-RBF kernel as kernel function of SVM,which can project the PCA processed data into high dimensional and guarantee the sparsity of high dimensional data to create a linear separability of the dataset.In order to prevent the training model from overfitting,the Kcross validation method was used to train the SVM.According to the test result of test set,SVM has good effect on the quality inspection of the solder joint.