Machine Vision Based Snap-type Thermostat Defect Detection System

With the introduction of "Made in China 2025",intelligent manufacturing has become the development trend of China's industrial field.As a product manufacturing technology,the product defect detection technology has also developed rapidly,from simple manual detection to advanced machine vision inspection,production efficiency and The product quality is constantly improving,and the research of machine vision inspection technology is of great theoretical significance and practical value for industrial automation detection.This topic takes the snap-type thermostat as the research object.According to the actual detection requirements of the thermostat,the design of the thermostat product defect detection system is carried out,and the related algorithms of image processing and machine learning are deeply studied to realize the automation of the thermostat defect.Detection,the test results meet the actual detection needs.The types of defects detected by this subject include the mixing of different types of products,the size of the pins and insulation paper,the parallelism of the insulation paper,and the contact depth.The model identification of the snap-type thermostat mainly identifies whether the wrong model of the same product model or mixed with other models,in order to judge whether the product type of the snap-type thermostat is qualified.Aiming at the problems of metal surface reflection,unevenness and embossed characters causing the product model recognition rate to drop,the improved BP neural network recognition method is proposed to characterize the grayscale character and character block duty of character images.And the total length of the line segment between the eight feature points of the character as the neural network input vector,the improved network can better recognize similar characters,thereby improving the overall recognition rate.The system first performs gray level enhancement,filtering denoising,binarization and other pre-processing work on the product model area in the image,and then divides and normalizes the characters.Finally,the improved BP neural network training is used to realize character recognition.Field experiments show that the method achieves accurate identification of the model characters of the thermostat.The digital recognition rate in single characters can reach 96.7%,the letter recognition rate can reach 95.5%,and the overall recognition rate can reach 95.1%.The detection of the pin and insulation paper size of the snap-type thermostat mainly measures whether the two pins of the thermostat are flush and the size of the insulation paper on both sides is the same,so as to judge whether the size of the thermostat pin and the insulation paper are qualified..According to the different degree of reflection between the metal casing and the insulating paper of the snap-type thermostat,the illumination method combining the ring light source and the strip light source is adopted,and the pixel difference between the temperature controller and the background is used,and the linear scanning method is firstly used to find The feature points with sudden changes in pixel values are recorded,and the coordinates of these feature points are recorded.Then,the dimensions of the pin and the insulating paper are calculated by the coordinate values of these feature points.Experiments show that the accuracy of the dimensional detection can reach 0.0714 mm.The system uses Hough transform algorithm to detect the edge of the insulating paper edge,obtain the coordinates of the two ends of the line segment,and judge whether the insulating paper is parallel by the coordinate value of the end point of the line segment.The contact depth detection of the snap-type thermostat mainly measures whether the contact depth of the thermostat is less than 1 mm,so as to judge whether the contact depth of the thermostat is qualified.The contact size of the snap-type thermostat is small,and it is difficult to realize the problem of depth measurement.The line laser triangulation method is adopted to first collect the laser stripe image,and then the pre-processing of filtering,binarization and stripe segmentation of the stripe image is performed.Then,the center is discretely extracted from the fringes,and the discrete points are fitted by linear interpolation.Finally,the depth of the thermostat contacts is calculated by the triangular geometric relationship.The experimental results show that the maximum deviation between the contact depth value measured by the line laser triangulation and the laser scanner measurement value is 0.0241 mm,and the maximum relative error is 5.53%.This measurement method improves the measurement efficiency and saves the detection cost.It can meet the needs of real-time detection of contact depth of the snap-type thermostat.