Research On Metal Sheets Defect Detection Based On Visual Laser Ultrasound

The application of metal sheet covers an abundance of fields such as life and production.In the process,various defects will appear in the sheet,which will cause harm to the normal operation of equipment and personal safety in long-term use,so the defect detection of metal sheet is crucial.With the development of ultrasound technology,ultrasound phased array and laser ultrasound technology still has the problems of high requirements for the test object and low defect detection capability,while phased array laser ultrasound inspection technology has the advantages of low detection threshold,high defect detection capability and high detection accuracy.The paper constructs a phased-array laser ultrasonic visualization detection system,conducts research on key technologies such as system hardware selection and supporting software design,and processes the echo data into a maximum amplitude map of the detection area according to ultrasonic imaging technology to complete the visualization of the collected signal.In response to the issues of poor accuracy,low efficiency and susceptibility to human factors in defect detection,the automatic recognition of defects is achieved by means of image pre-processing and deep learning.The main work of this paper is as follows:(1)With ultrasonic inspection technology as the theoretical support,the design of a phased-array laser ultrasonic visual inspection system is given,focusing on software system design.The software system adopts the modular idea based on Lab VIEW platform and is divided into several modules according to the functions.Based on the detection system to detect defects in metal plates,by the laser source excites ultrasonic waves on the surface of the plate,the control software collects ultrasonic signal data by the human-computer interaction interface,the signal data is filtered,interpolated and envelope taken to finally obtain the maximum amplitude map for visualizing the defect image acquisition.(2)The maximum amplitude map has problems such as low signal-to-noise ratio and poor contrast,and image noise interference is reduced using image graying and bilateral filtering techniques.Comparing the effects of classical image enhancement algorithms and analyzing their limitations,an improved multi-scale Retinex algorithm is proposed to combine the subjective and objective evaluation results to verify that the improved algorithm can better retain the defect contour edge information and filter out the noise.(3)Build a convolutional neural network model,design the convolutional kernel size and overall architecture.The experimental data set was trained and tested,the defect recognition rate of the test set data was as high as 97.00%,and the noise misjudgment rate was as low as4.29%.To further verify the effectiveness of the model,two sets of experimental controls were established and the test samples were tested according to the specified conditions,the results show that the network model outperforms the traditional defect recognition model in terms of defect recognition rate,noise misjudgment rate,convergence time,loss value and other indicators,and can complete the metal sheet defect detection with high accuracy and efficiency.