Research On Sensitive Feature Sets Of Total Focus Images Based On Ultrasonic Phased Array For Weld Defects

As an important processing technology,welding technology has been widely used in many industrial manufacturing fields,such as automobile manufacturing,shipbuilding,building construction,aerospace equipment and nuclear industry.Due to the influence of factors such as welding environment,welding method and structural stress,there may be five kinds of internal defects in the formed weld,such as inclusions,pores,cracks,incomplete penetration and incomplete fusion.These defects will greatly affect the performance of the welded structure,causing product damage and even casualties.Therefore,the nondestructive testing of the weld structure is an important guarantee to ensure the normal industrial production.With the development of science and technology,ultrasonic phased array detection technology has been more and more widely used in the engineering detection process because of its high detection sensitivity,fast detection speed,intuitive detection results and the ability to detect irregular surface of the workpiece.At the same time,the maturity of full-matrix-full-focus imaging technology greatly improves the imaging resolution and provides a new idea for the classification and recognition of weld defects.However,there are still many difficulties in identifying the weld defects using the full-focus atlas.For example,the full-focus atlas features of all kinds of defects are similar visually,and it is difficult to identify the defect types directly by artificial observation and there are many kinds of weld defects.By using convolutional neural network technology based on depth learning theory to identify the atlas will bring extremely high computational complexity,and has high requirements on computer hardware,the number and quality of training samples.Therefore,in this paper,the above background and the existing problems at this stage are studied as follows:(1)The development history,testing methods and working methods of ultrasonic phased array nondestructive testing technology are summarized.The principle of calculation and generation of full focus chromatogram is introduced,and the related diagram and formula are given.The test pieces with five kinds of defects including inclusion,porosity,crack,incomplete penetration and incomplete fusion were designed and made by the entrusted manufacturer according to the standard of Nondestructive Testing of Pressure Equipment NB/T47013-2015,and the defects of the test pieces were qualitative analyzed by the method of X-ray detection.On this basis,the rest of this paper is completed.(2)In the preprocessing of ultrasonic phased array full-focus atlas,this paper starts with the characteristic differences between target defects and reflected clutter,and integrates threshold method and region segmentation method to effectively filter the interference of reflected clutter on the basis of completely retaining the details of defects.The experimental comparison shows that the algorithm in this paper not only achieves good denoising effect,but also brings less computational complexity,so it is suitable for the preprocessing of full-focus atlas of ultrasonic phased array for welding seam.(3)In the aspect of feature extraction of ultrasonic phased array full-focus atlas,this paper combines the characteristics of weld defects and the generating principle of ultrasonic phased array full-focus atlas,extracts 70 defect features from three aspects of atlas color distribution,defect morphology and regional texture details and constructs the total feature set of weld defect atlas,aiming at quantifying all kinds of defect features from various angles.(4)In the aspect of defect sensitive identification feature set construction,in view of the characteristics of many kinds of weld defects and high dimension of feature set,this paper adopts feature sensitivity based on Euclidean distance and feature correlation based on Spearman coefficient as evaluation criteria,and visually displays the distribution of defects through t-sne feature visualization algorithm in MATLAB software platform,finds out the sensitive features of each kind of defects,summarizes the sensitive identification features of each kind of defects,and finally realizes the sensitive identification feature set construction of ultrasonic phased array inspection full focus atlas of all kinds of weld defects.Finally,based on the proposed defect sensitive identification feature set,a multi-layer perceptron model is built to classify and identify five types of weld defects,and the recognition accuracy rate reaches 80.7%,which proves that the ability of sensitive identification feature set to describe defects can accomplish the task of defect intelligent identification.