Research On Surface Defect Detection System Of Magnesium Alloy Sheet Based On Computer Vision

Magnesium is one of the lightest structural metal materials.Owing to magnesium alloy characters of light weight,good rigidity,low density,and heat dissipation,it is widely used in various fields,such as aviation,transportation,and chemical.During the rolling process of magnesium alloy sheet,due to the mechanical and control accuracy,these defects are easily appeared such as the edge cracking,ripple and fold.These problems are not solved in time,that seriously affects the quality of magnesium alloy sheet.Therefore,how to detect efficiently and quickly the magnesium alloy strip surface defection is the key to its manufacture.The traditional manual detection method has the problems of low efficiency,high false-positive rate and human resource cost.Using these existed detection techniques,such as the eddy current testing,the infrared testing and the magnetic flux leakage detection,the detection rate and defect type identification are extremely limited because of the principle limitations of the traditional techniques.With these technologies development of computer,automation,artificial intelligence and image recognition,the surface defect detection technology based on computer vision recently has been the study emphasis in the production process of the magnesium alloy strip.The magnesium strip surface defect recognition system is developed in the study,which real-time captures the strip surface images,automatically detects the defects by the computer vision technology and classifies these defects by the Bayes Classifier.These functions lay the foundation for the unattended operation and automation in the next step produce process.The main achievements are as following:(1)According to the experimental environment,the actual condition of production line and the project budget,the hardware and software design of this system is decided,such as the choice of LED light source,lighting scheme,and the configuration of operating system,etc.(2)Because of the strip surface characteristics including low contrast and strong reflection,some image preprocessing methods such as the histogram equalization and the median filtering are chosen.(3)Through analyzing the five common defect features of the magnesium alloy sheet,nine eigenvalues of geometric and texture features,are selected as the eigenvectors.(4)On the basis of comparing the characteristics of different kinds of defect classifiers,considering the detection speed requirement in the actual production,Bayes Classifier which can quickly deal with multi-class problem has been chosen to judge the defect classification.(5)Through the research of various software platforms,the surface defect real-time detection system of magnesium alloy sheet are realized.During the real-time detect process,the defect recognition rate is 83.6% usingthe system and 16 milliseconds is taken to identify one sample,which meet the actual needs of the industrial production.