Casting Defect Detection Based On Laser Ultrasound And Convolutional Neural Network

All kinds of metal castings,such as metal tools,are often used in power line construction,which is related to the safe operation of power system.When the equipment is damaged,it will not only affect the mechanical properties and insulation properties,but also bury hidden dangers to the safe operation of the power system.For the inspection of metal castings,the traditional manual inspection means may fail to detect minor defects.In response to these problems,this paper proposes the application of laser ultrasonic visualization testing technology to nondestructive testing of metal castings.The advantages of this technology are high accuracy and visual testing effect.The maximum amplitude image is obtained by laser ultrasonic visualization detector,and then artificial intelligence technology is used to identify and grade the image defects.Combining the above two methods,a defect detection system based on laser ultrasound and convolutional neural network is designed.The details are as follows:(1)Investigate the defect detection methods of metal castings and the current situation at home and abroad,design the overall framework and specific work flow of the detection system,and focus on the analysis of image acquisition of laser ultrasonic visualization detector and image classification technology of convolutional neural network.(2)The laser ultrasonic visualization instrument is used to scan the metal castings,and the ultrasonic energy data of the detection area is processed into the maximum amplitude map of the detection area according to the ultrasonic imaging technology.The characteristics of the maximum amplitude image are studied,and the appropriate image classification technology is selected according to this feature for defect identification and grading.(3)There are many methods of image classification in the field of artificial intelligence,each with its own characteristics.By comparing different methods,convolutional neural network is selected as the image classification method of this system.At the same time,based on the analysis of the obtained maximum amplitude image features,the convolutional neural network was optimized by changing the size of convolutional kernel,the number of convolutional layers,pooling method,activation function and loss function in the convolutional neural network,and a convolutional neural network model that can accurately identify defects was obtained.In addition,according to the relevant national standards,the identified defects are graded,and the defects are divided into uneven and damaged.(4)According to the needs of practical applications,a system to assist users in defect identification and grading is designed.In addition to the two core functions of defect identification and grading,the system is equipped with user management,parameter storage and result query functions.Finally,through experimental verification,getting a system which accuracy of defect identification reached 93.5%,and the accuracy of defect grading reached more than 95%.