Design Of On-line Inspection System For Surface Defects Of Cold Rolled Strip

Surface quality is an important index of strip steel production.Quickly and accurately detecting surface defects of strip steel has always been a concern of steel production enterprises.There are many problems of existing detection algorithm,such as high complexity,long calculation time,and false information like noise and false defects,due to the harsh environment and serious external interference.In view of the above problems,the main contents of this thesis are as follows:(1)The defect detection system for strip steel is designed in this research.According to the characteristics of the strip production site,the camera and light source are selected with appropriate parameters,and the design and construction of the hardware platform are completed.The software program is compiled,realizing the defect detection and classification algorithm of strip steel,as well as the human-machine interaction function of image display and storage.(2)The strip steel surface defects are rapidly detected and positioned.Firstly,the edge characteristics and causes of 5 typical defects studied in this thesis were researched,according to the different of pixel change rate complexity refers to defective and no defective images,preliminarily screens defect images.Then,using edge detection algorithm based on wavelet transform modulus maxima,accurately detected defect edge and suppressed noise.Finally,an adaptive double threshold edge connection algorithm with two different thresholds is proposed to improve the edge integrity of the detection results.(3)Features of strip steel surface defects are extracted and classified.Firstly,extracted 78 dimensional feature including morphological characteristics,gray feature,texture feature and projection feature.Then,according to the different characteristics of the rotation invariant and similar characteristics together amplification to remove redundancy.And through kernel principal component analysis selecting 34 dimensional feature according to the cumulative contribution rate was above 95%,and is input to classifier.Finally,the BP neural network is selected,,and the final classifier is got by using defect sample and setting reasonable parameters.(4)Application of defect detection system is researched.Firstly,the reasons for misreporting is analyzed,such as noise interference,uneven illumination,water stains and other false defects.The solution of the above three reasons are as follows,calculating the detected region gray level co-occurrence matrix,that is judged as noise which is less than the threshold,proposing the line fitting gray correction method to remove uneven illumination,and using the BP neural network classifier to separate stains and defects.The performance of this algorithm is verified by experiments.The algorithm of edge detection algorithm is better to precisely detect defect edge.The edge defect rate of 104 samples is 97.1%.The average classification rate of 5 kinds of defects that are investigated in 1305 images is 91.24%.The average detection algorithm time of 2049×256 images is 41.99ms at present,which meet the real-time requirements of strip production.