CN-121982026-A - MIM gear tooth profile precision detection method based on machine vision

CN121982026ACN 121982026 ACN121982026 ACN 121982026ACN-121982026-A

Abstract

The invention relates to the technical field of image data processing, in particular to a machine vision-based MIM gear tooth profile accuracy detection method which comprises the steps of obtaining an MIM gear image, extracting a plurality of profiles in the MIM gear image, calculating curvature dispersion degree, homodromous degree and adjustment factors of all the profiles, adjusting an attention weight matrix in an edge detection model based on the adjustment factors, extracting a target profile of the MIM gear image by using the adjusted edge detection model, and judging the MIM gear tooth profile accuracy based on deviation of the target profile and a theoretical profile. The invention solves the problem of low stability of the detection result.

Inventors

HU JIANBIN
LIU XIAOJING
LIANG JUNHUI
GUO XIANLIANG
LUO PENGFEI

Assignees

连云港富驰智造科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (9)

1. The method for detecting the tooth profile precision of the MIM gear based on machine vision is characterized by comprising the following steps of: acquiring an MIM gear image, and extracting a plurality of contours in the MIM gear image; Calculating curvature dispersion of each contour, wherein the curvature dispersion represents the dispersion degree of each contour, calculating the homodromous degree of each contour, wherein the homodromous degree is positively correlated with the curvature dispersion degree and the cosine average value of the difference value between gradient direction angles of any two adjacent pixel points in each contour, and calculating the regulating factor of each contour, namely Adjustment factors for individual profiles The method comprises the following steps: , Is the first The degree of curvature dispersion of the individual profiles, Is the first The degree of co-directional of the individual profiles, Is the first The proportion of pixels in each contour to pixels in the MIM gear image, Is a non-zero constant; and adjusting the attention weight matrix in the edge detection model based on the adjustment factor, extracting the target contour of the MIM gear image by using the adjusted edge detection model, and judging the tooth profile accuracy of the MIM gear based on the deviation of the target contour and the theoretical contour.
2. The machine vision-based MIM gear tooth profile accuracy detection method of claim 1, wherein the first step is that Curvature dispersion of individual contours The method comprises the following steps: , 、、 Respectively the first In the first outline First, second First, second The local curvature of the individual pixel points, Is the first The total number of pixel points in the profile, Is the first The mean value of the local curvature of all pixel points in the profile, The super parameters are preset.
3. The machine vision-based MIM gear tooth profile accuracy detection method according to claim 2 is characterized in that the local curvature is specifically that a target pixel point in a profile is taken as a center, a local pixel section comprising five pixel points is formed by the target pixel point and the nearest four adjacent pixel points, arc fitting is conducted on the local pixel section by means of a least square method, and the curvature of the fitted arc is taken as the local curvature of the target pixel point.
4. The machine vision-based MIM gear tooth profile accuracy detection method of claim 1, wherein the first step is that Degree of co-directional of individual contours The method comprises the following steps: , Is the first The degree of curvature dispersion of the individual profiles, Is the first The total number of pixel points in the profile, Is the first In the first outline The average value of the difference value between each pixel point and the gradient direction angle of the nearest neighbor pixel point, wherein the nearest neighbor pixel point is a four-neighborhood or eight-neighborhood pixel point.
5. The machine vision-based MIM gear tooth profile accuracy detection method of claim 1, wherein attention weight matrix in an edge detection model is adjusted based on the adjustment factor, specifically, a modulation matrix is obtained, and the modulation matrix and an original attention weight matrix are multiplied element by element to obtain an adjusted attention weight matrix; the elements in the modulation matrix are obtained by bilinear interpolation of the adjustment factors of the contours to which the elements belong, and meanwhile, the value of the elements which do not belong to the contours is 1.
6. The machine vision-based MIM gear tooth profile accuracy detecting method of claim 1, wherein a CCD camera is used to obtain MIM gear images.
7. The machine vision-based MIM gear tooth profile accuracy detecting method of claim 1, wherein the MIM gear image is subjected to denoising and graying.
8. The machine vision-based MIM gear tooth profile accuracy detecting method of claim 1, wherein the edge detecting model is a transducer-based edge detecting model.
9. The machine vision-based MIM gear tooth profile accuracy detection method of claim 1 is characterized by judging the precision of the MIM gear tooth profile, specifically, judging that the precision of the MIM gear tooth profile is normal if the deviation of a target profile and a theoretical profile is smaller than a preset threshold.

Description

MIM gear tooth profile precision detection method based on machine vision Technical Field The invention relates to the technical field of image data processing. More specifically, the invention relates to a machine vision-based MIM gear tooth profile accuracy detection method. Background The metal injection molding technology is a precise manufacturing technology combining powder metallurgy and plastic injection molding technology, and is mostly applied to mass production of microminiature and complex-structure parts. MIM gears are applied to the fields of consumer electronics, automobile transmission mechanisms and precise instruments because of the advantages of high forming precision, suitability for complex tooth-shaped structures, high material utilization rate and the like. However, in the actual production process, the MIM gear is easily affected by factors such as uneven shrinkage of materials, release of internal stress, and slight deviation of a mold in degreasing, sintering and cooling stages, so that the profile of the tooth profile is slightly deformed, partially collapsed, or edge burrs are caused. Such defects tend to be small in size but directly affect meshing accuracy, transmission stability and service life, so that high-accuracy, non-contact detection of tooth profile accuracy is required. With the development of machine vision technology, a tooth profile detection method based on image processing is becoming a mainstream scheme. The gear image is obtained through an industrial camera, and the tooth-shaped edge is extracted, fitted and compared and analyzed, so that the tooth-shaped error can be rapidly evaluated. However, the MIM gear generally has the characteristics of small size, complex tooth shape, dense edge details, and the like, and is easily affected by factors such as uneven illumination, surface reflection, background interference, and the like in an actual collection process, so that noise, fracture or false contours exist in an edge area. In recent years, an edge detection model based on a transducer structure has a global modeling capability, and therefore, the edge detection model has a better feature extraction capability in a complex scene, and is gradually applied to the field of industrial visual detection. However, in the MIM gear profile detection scenario, the model needs to accurately identify a small and continuous real tooth edge, while suppressing a small burr and noise edge, which puts higher demands on feature expression and weight distribution. The existing edge detection model based on the Transformer generally relies on global correlation among features to perform unified modeling when calculating an attention weight matrix, and lacks a special constraint mechanism for geometrical continuity and contour quality difference. In the MIM gear detection scene, the real tooth profile generally has stronger curvature continuity and direction consistency, and the false edge generated by noise or reflection often presents irregular fluctuation, but the existing attention weight distribution mode is difficult to directly distinguish the two types of characteristics, and the high-contrast but unstable-form area is easily given too high weight, so that false detection or edge fracture is caused, and the problem of low stability of a detection result is caused. Disclosure of Invention In order to solve the problem of low stability of the detection result proposed in the background art, the invention provides the following scheme. The invention provides a machine vision-based MIM gear tooth profile precision detection method which comprises the steps of obtaining an MIM gear image, extracting a plurality of profiles in the MIM gear image, calculating curvature dispersion of each profile, calculating the degree of homodromous of each profile, wherein the degree of homodromous is positively correlated with the curvature dispersion and cosine average value of difference value between gradient direction angles of any two adjacent pixel points in each profile, and calculating adjustment factors of each profile, namelyAdjustment factors for individual profilesThe method comprises the following steps:, Is the first The degree of curvature dispersion of the individual profiles,Is the firstThe degree of co-directional of the individual profiles,Is the firstThe proportion of pixels in each contour to pixels in the MIM gear image,And adjusting the attention weight matrix in the edge detection model based on the adjustment factor, extracting the target contour of the MIM gear image by using the adjusted edge detection model, and judging the tooth profile contour precision of the MIM gear based on the deviation of the target contour and the theoretical contour. According to the technical scheme, the attention weight matrix in the edge detection model is adjusted in a targeted mode by utilizing the adjusting factors, and the weight distribution strategy is optimized on the premise th