Research On Intelligent Qualitative Classification System Of Ultrasound Phased Array Map For Weld Defects Of Aluminum Alloy With Heteromorphic Structure

As a common connection method between metal workpieces,welding is used in many fields.Its excellent quality directly affects the strength and service time of special-shaped bearing parts.In severe cases,it may lead to catastrophic accidents.With the advent of Industry4.0,the industry pays more attention to the quality of welding and the inspection rate.At present,the traditional method in the industry to qualitatively analyze the ultrasonic phased array pattern of internal defects in welds often relies on experienced employees to observe the characteristics of the pattern.This difference is magnified due to changes in energy when large-scale detection of defect maps is required.Due to the above reasons,the use of traditional methods to qualitatively qualitative ultrasonic phased array patterns had gradually been eliminated.In recent years,the trend of cross-discipline and multi-discipline integration had become unstoppable.In particular,the combination of artificial intelligence and various industries had achieved remarkable results,prompting people to introduce artificial intelligence into the qualitative recognition of the ultrasonic phased array map of special-shaped welds.Aiming at the current problem of difficult qualitative determination of small defects in welds,this thesis was based on the ultrasonic phased array inspection map of typical weld defects,and aimed to solve the problem of quality inspection of welding workpieces.The main work of the thesis includes:(1)In the process of constructing the feature set of ultrasonic phased array inspection image for identifying the types of internal defects in welds,in order to accurately extract feature sets that significantly respond to the types of weld defects.A salient feature evaluation method based on the fusion of multiple evaluation criteria was proposed,and the calculated defect image feature index was used as the evaluation object,and the gray image features that significantly respond to the weld defect type were screened and sorted.This method aimed to obtain a feature set that significantly responds to the types of internal defects in the weld,and to improve the accuracy of the qualitative results of the internal defects of the aluminum alloy weld with special-shaped structures.(2)In the process of constructing an intelligent network architecture for ultrasonic phased array inspection images to identify weld defect types,a stack sparse autoencoder fusion core extreme learning network model was proposed to improve the classification accuracy.Calculate the sensitivity of the defect type inside the weld,specifying it as the initial weight parameter of the stack sparse autoencoder fusion kernel extreme learning network,At the same time,the optimization goal of the stack sparse autoencoder fusion core extreme learning network was that,minimized the sample reconstruction error,minimized the sample difference within the class,and maximized the sample difference between classes.Finally,the intelligent qualitative classification of the ultrasonic phased array image of the internal defects of the special-shaped structure weld is realized.(3)In the process of designing database software for ultrasonic phased array inspection images for identifying defect types inside welds,for the convenience of use,the database software for analyzing typical defect maps of special-shaped structure welds combines stack sparse autoencoder fusion kernel extreme learning Network,mysql and pyqt5,the intelligent identification of the final defect on-site inspection map and the on-site rapid comparison function of the standard map are realized.(4)The classification accuracy of the method proposed in this thesis reaches 97.8%,which can be used for actual field detection.