CN-115358964-B - Omnibearing defect detection method and system for industrial defects of valve oil seal

CN115358964BCN 115358964 BCN115358964 BCN 115358964BCN-115358964-B

Abstract

The invention belongs to the technical field of machine vision, relates to product surface defect detection, and discloses an omnibearing defect detection method and system aiming at industrial defects of valve oil seals. The method comprises the steps of collecting overlooking images of the valve oil seal, obtaining lip radius from a front end interface, obtaining radius data of a lower spring seat, carrying out area positioning and detection on a lip rubber head and the lower spring seat, collecting side view images of the valve oil seal, obtaining metal framework height from the front end interface, carrying out area positioning and detection on a metal framework, a lip bottom edge and a compression spring, collecting looking-up images of the valve oil seal, carrying out area positioning and detection on the lower spring seat, integrating obtained detection information, and judging qualified products and unqualified products. The invention has the capability of detecting the valve oil seals of different types, has good compatibility of a detection system, better algorithm robustness and meets the quality inspection requirement.

Inventors

CHEN MEIMEI
Zhang Genrui
WANG QINGFANG
XUAN YUBO
Cao Diejie
LI BOYANG
He Houliang
WANG YUE
LIU DONGXUE

Assignees

吉林大学

Dates

Publication Date: 20260508
Application Date: 20220629

Claims (4)

1. An all-round defect detection system to valve oil blanket industry defect, characterized in that, this system includes: the data acquisition unit acquires a top view image, a side view image and a bottom view image of the valve oil seal, and acquires lip radius data and lower spring seat radius data from a front end interface; The area positioning unit is used for positioning the lip rubber head and the lower spring seat area and positioning the metal framework, the lip bottom edge and the compression spring area according to the acquired data and images, wherein the lip rubber head and the lower spring seat area are used for performing contour detection on overlooking images, the outermost layer is externally connected with a rectangular positioning oil seal, and the lip rubber head roi and the overlooking lower spring seat roi are intercepted according to known parameter proportion; The positioning of the metal framework, the lip bottom edge and the compression spring area comprises the steps of obtaining the width of a valve oil seal, obtaining the height of the valve oil seal, and intercepting, wherein the specific steps are as follows: acquiring the width of the valve oil seal, namely intercepting the upper half part of the image along the 1/2 position of the side view image, and detecting the width value of the valve oil seal metal skeleton in the horizontal direction after pretreatment; The valve oil seal height obtaining method comprises the steps of performing double-threshold binarization operation on an image, performing N equal division on the binarized image along the horizontal direction, performing N equal division image phase or operation, and performing outer layer horizontal line identification to obtain the valve oil seal metal framework height in the vertical direction; intercepting, namely intercepting to obtain a compression spring roi, a metal framework roi and a lip bottom edge roi; The lower spring seat area positioning is that contour detection is carried out on the upward image, a rectangular positioning oil seal is externally connected with the outermost layer, and the upward lower spring seat roi is intercepted according to the known parameter proportion in the step 1; The detection unit is used for detecting the lip rubber head, the lower spring seat, the metal framework, the lip bottom edge and the compression spring according to the region positioning result, wherein: The lip rubber head detection and the lower spring seat detection comprise lip framework segmentation, burr identification, lip rubber head detection and lower spring seat detection, and specifically comprise the following steps: Performing contour detection on the acquired overlooking image of the valve oil seal, performing polygon fitting on the obtained contour, screening under certain conditions to obtain the skeleton contour of the lip rubber head image, and judging that the oil seal with the contour level not meeting the corresponding level is defective; lip rubber head detection, namely predicting lip rubber head roi through a twin network depth model, and outputting a qualified result; The lower spring seat detection, namely preprocessing a spring seat roi in a overlooking state, then carrying out convex hull identification, calculating the area and gray moment of the convex hull, and judging the convex hull which does not meet a threshold value as a glue dropping point, namely a defective product; The metal skeleton detection is to perform convex hull identification on a metal skeleton roi, calculate the area and gray moment of the convex hull, and judge the convex hull which does not meet the threshold value as a glue dropping point, namely as a defective product; lip bottom edge detection is to predict lip bottom edge roi through a twin network depth model; the detection of the compression spring comprises the following steps of accurate positioning, stripe band gap identification, stripe band amplitude distribution change identification, and the specific steps of: Performing convex hull identification on the compression spring roi, and fitting the horizontal line positions of the convex hull with zero-order moment dense distribution to obtain the accurate positions of the spring areas to be identified; the strip gap identification, namely carrying out horizontal opening operation on the precisely positioned image to obtain an image focusing on strip distance information only, and identifying the mutual distance of the strips, wherein if the distance exceeds a specified threshold value, the spring is judged to have a fracture phenomenon, namely a defective product; The strip band amplitude distribution change identification comprises binarizing a compression spring roi, performing left-right folding operation, and judging that the strip band area with overlarge area proves that the amplitude change is asymmetric and is defective, namely a defective product; The lower spring seat detection is to perform convex hull identification on the lower spring seat roi, calculate the area and gray moment of the convex hull, and judge the convex hull which does not meet the threshold value as a glue dropping point, namely as a defective product; and judging whether the product is qualified or not according to the detection result.
2. An omnibearing defect detection method aiming at the industrial defect of a valve oil seal, which adopts the omnibearing defect detection system aiming at the industrial defect of the valve oil seal according to claim 1, and is characterized in that the lip part of the valve oil seal, the surface of a metal framework and a lower spring seat are detected, and the method comprises the following steps: step 1, acquiring a overlooking image of an air valve oil seal, acquiring lip radius from a front end interface, acquiring radius data of a lower spring seat, and carrying out area positioning and detection on a lip rubber head and the lower spring seat; Step 2, acquiring a side view image of the valve oil seal, acquiring metal skeleton height and lip height data from a front end interface, and carrying out region positioning and detection on the metal skeleton, lip bottom edge and a compression spring; Step 3, acquiring a bottom view image of the valve oil seal, and carrying out area positioning and detection on the lower spring seat; And 4, integrating the detection information obtained in the steps 1, 2 and 3, and judging the qualified products and the unqualified products.
3. The omnibearing defect detection method for valve oil seal industrial defects according to claim 2, wherein, The training method of the network training classifier of the twin network depth model comprises the following steps: 3-2-1), dividing the pretreated roi data set of the lip rubber head into a positive example and a negative example, and dividing a training set, a verification set and a test set; 3-2-2) sending the data set into a vgg-based twin network depth model for training; 3-2-3) predicting the image to be detected by the trained model to obtain the similarity between the image to be detected and the qualified product image, and judging the image to be unqualified if the similarity is low.
4. The omnibearing defect detection method of the industrial defect of the valve oil seal according to claim 2, wherein the lip bottom edge detection in the step 2 is to predict a lip bottom edge roi through a twin network depth model, and the training method of a twin network training classifier is as follows: 6-1), dividing the pretreated roi data set of the oil seal shoulder into a positive example and a negative example, and dividing a training set, a verification set and a test set; 6-2) sending the data set into a vgg-based twin network depth model for training; And 6-3) predicting the image to be detected by the trained model in the step 6-2) to obtain the similarity between the image to be detected and the qualified product image, and judging the image to be unqualified if the similarity is low.

Description

Omnibearing defect detection method and system for industrial defects of valve oil seal Technical Field The invention belongs to the technical field of machine vision, relates to product surface defect detection, and discloses an omnibearing defect detection method and system aiming at industrial defects of valve oil seals. Background Defect detection is a very important part of industrial production, and quality control has a significant impact on the quality of the final product and thus on the reputation of the company in the market. In the traditional industrial production, the defect detection is often finished by using human eyes, which causes the defects that subjective influence is large, the human eyes can easily miss detection and miss detection in a large number of repeated works, human resource cost is high, and the working environment of quality inspection of workers is severe and causes great damage to the bodies of the workers. Therefore, the machine vision replaces human eye defect detection, so that the cost can be reduced, the precision can be improved, and the working environment of workers can be improved. The machine vision algorithm is developed based on complementary hardware detection light paths and a relatively stable darkroom working environment. The application of the traditional machine vision algorithm in defect detection is often based on feature extraction, extracted feature operators are often low-order and are easy to be influenced by environmental change and deformation, but the algorithm is flexible, the program speed is high, and the data set is not relied on. The deep learning achieves very good results on the extraction of image features, can ensure the capability of discriminating complex defects while achieving high precision, but needs to rely on enough dataset pictures as support. Therefore, the system algorithm combining the two can better solve the problem. In the existing literature for detecting visual defects of the valve oil seal, a deep convolutional neural network is used for detecting partial defects on the surface of the valve oil seal, classifying the defects, identifying the defects by using a target detection or classification model, and directly using edge extraction and pattern identification of traditional machine vision to detect single defects. The above visual defect identification method for the valve oil seal has two problems in practical industrial detection application: 1. the training data set covering all defects of the valve oil seal is difficult to manufacture; in practical problems, the number of defect types in a training sample is always small, and the work of acquiring a sufficient defect sample acquisition data set is particularly difficult in consideration of the practical production flow of the valve oil seal, so that the detection capability and detection precision of a depth model on defects are greatly limited. 2. The identification algorithm of single defect can not meet the requirement that the quality inspection requirement of the valve oil seal in actual production is zero defect, and for the full-flow algorithm which can be practically applied, the identification capability of multiple defects and the high-precision identification of complex defects are required, so that the effective classification of qualified products and unqualified products is realized. Aiming at the problems existing in practical application of the valve oil seal defect detection method, the invention provides an omnibearing defect detection method for detecting industrial defects of a valve oil seal, and various defects are effectively detected by combining a traditional method with deep learning. Disclosure of Invention Aiming at the technical problems in the related art, the invention provides an omnibearing defect detection method aiming at the industrial defects of the valve oil seal, and on the other hand, provides an omnibearing defect detection system aiming at the industrial defects of the valve oil seal. The present invention has been achieved in such a way that, An omnibearing defect detection method for detecting the industrial defects of a valve oil seal comprises the following steps of: step 1, acquiring a overlooking image of an air valve oil seal, acquiring lip radius from a front end interface, acquiring radius data of a lower spring seat, and carrying out area positioning and detection on a lip rubber head and the lower spring seat; Step 2, acquiring a side view image of the valve oil seal, acquiring metal skeleton height and lip height data from a front end interface, and carrying out region positioning and detection on the metal skeleton, lip bottom edge and a compression spring; Step 3, acquiring a bottom view image of the valve oil seal, and carrying out area positioning and detection on the lower spring seat; And 4, integrating the detection information obtained in the steps 1, 2 and 3, and judging the qualified products and the