Research On Visual Detection Technology Of Typical Powder Metallurgy Products

Powder metallurgy uses metal powders or mixed metal and non-metal powders as raw materials to produce metallic materials,composite materials and powder metallurgical products through pressing and sintering processes.This technology involves several industrial disciplines,such as metallurgy,mechanics and materials,and has become one of the effective solutions to new material problems.Powder metallurgical workpieces are widely used in the automotive industry,aerospace,energy and environmental protection,and have become an irreplaceable part of industrial production and manufacturing.Therefore,the importance of inspection research for powder metallurgy products is self-evident.Traditional inspection techniques for powder metallurgy products include manual inspection,ultrasonic inspection,eddy current inspection and magnetic leakage detection,etc.Although certain results have been achieved,they still cannot avoid the disadvantages of low efficiency,high cost and damage to products.In response to the current problems,this paper proposes a machine vision-based method combined with deep learning for detecting defects in powder metallurgy products,using a typical powder metallurgy product gear as the inspection object,and accomplishing the following research tasks:(1)According to the inspection demand of powder metallurgical products and the design process of the inspection system,the design scheme of the inspection system is completed.The whole system includes several parts of control system,mechanical system,software system and vision system,while the hardware parts of the vision system such as light source,camera and lens are selected and analyzed,and the calibration of the system is completed.(2)The image pre-processing operation was performed on the images captured by the detection system,including grayscale processing of the color images,followed by filtering of the grayscale images,followed by automatic thresholding of the images to divide them into two parts,and morphological open operation of the images for problems such as highlight noise and edge burr in the binary images after thresholding,and then the contour screening extraction algorithm,which not only excludes the influence of redundant clutter contours,but also successfully extracts the edge contours of defective regions.(3)The size of the standard gear and its error range are determined by particle analysis method measurement.Broken teeth and stain defects can be judged directly by area comparison,while the scratch area belongs to the highlight area,which cannot be detected by binarization and edge extraction,so this paper designs a grayscale operation method for scratch defects,analyzes the difference between the grayscale value of the defect area and the background by grayscale histogram,and then performs subtraction operation to extract the After that,the noise effect is removed by filtering and morphological processing,and finally the measurement is completed by particle analysis.(4)The study analyzed the basic structure and training process of convolutional neural network,produced a defect dataset and performed noise and geometry expansion for the types of defects in powder metallurgical gears;analyzed the structure and characteristics of two typical convolutional neural classification detection networks,VGG16 and Inception V4,and conducted training comparison using the dataset images,and the results showed that the Inception V4 model The training results show that the Inception V4 model outperforms the VGG16 model.(5)Using LabVIEW as the development environment and TensorFlow as the deep learning framework,the software system design was completed,including user login module,image acquisition module,image processing module,detection and recognition classification module,sorting module and data management module.Finally,the system is tested for performance.The experiments show that the system achieves 96%accuracy,1.67%false detection rate and 2.33%leakage rate for defect detection of powder metallurgy gears,which is in line with the expected set target.