Research On Visual Inspection Method Of Surface Crack Defects Of Powder Metallurgy Part

Powder metallurgy parts have the advantages of high precision,high energy saving and good stability,and have been widely used in industrial production.But in the preparation of powder metallurgy parts will produce crack due to various reasons,the need for effective control.Traditionally,manual inspection of crack defects is slow and costly.Visual defect detection method testing speed,high precision,good stability,can solve the problem of crack defects of powder metallurgy parts batch testing.In this paper,a systematic study is carried out,and the work of visual imaging,image analysis and processing,training of detection model and software development are completed,which can realize automatic online detection.First of all,through the analysis of the characteristics of surface crack defects of a powder metallurgy parts,test system overall scheme is designed.In terms of hardware,an image acquisition platform is designed and built to ensure that the crack defects can be stably imaged.In terms of software algorithm,a detection algorithm combining traditional image processing and deep learning is designed for defect recognition.Secondly,in order to improve the speed and accuracy of the detection algorithm,a preprocessing algorithm is proposed.Aiming at the problem of large amount of calculation caused by the large proportion of image background,an improved Canny algorithm and improved SSDA matching algorithm are proposed,combined with the image pyramid template matching search strategy,which can achieve the effect of fast and accurate location of parts,eliminate the background interference,and reduce the amount of calculation.An improved adaptive logarithmic transformation algorithm is proposed to solve the problem that crack and part surface contrast have weak influence on the detection results.The algorithm can achieve a high contrast effect between crack and part surface,and complete the image preprocessing of parts.Then,an improved VGG19 network detection model is proposed.Aiming at the problem that the traditional VGG19 model has low accuracy in crack detection of parts,it has been improved in many aspects.Firstly,the BN layer and random inactivation layer are introduced,and the number of network layers is added and deleted to get a new network structure.Then,the image preprocessing and enhancement of the dataset are carried out.Finally,transfer learning is carried out.Design experiment training and compare the advantages and disadvantages of VGG19 migration model and the improved VGG19 migration model,it is concluded that the improved network model training effect is better.Finally,in order to verify the effectiveness of the detection algorithm proposed in this paper,the detection software and the algorithm are implemented by programming,and the actual defect samples are used for experimental verification.The detection accuracy of 390 test images is 98.97%,and the average detection time of each image is about 0.8s.The experimental data show that the detection algorithm proposed in this paper is better than the traditional algorithm and the unimproved deep learning algorithm,and meets the indicators of the research subject and meets the production needs of enterprises.