[go: up one dir, main page]

CN110927171A - A method for detecting defects on the chamfered surface of bearing rollers based on machine vision - Google Patents

A method for detecting defects on the chamfered surface of bearing rollers based on machine vision Download PDF

Info

Publication number
CN110927171A
CN110927171A CN201911250746.7A CN201911250746A CN110927171A CN 110927171 A CN110927171 A CN 110927171A CN 201911250746 A CN201911250746 A CN 201911250746A CN 110927171 A CN110927171 A CN 110927171A
Authority
CN
China
Prior art keywords
defect
area
image
bearing roller
sample
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201911250746.7A
Other languages
Chinese (zh)
Inventor
杜劲松
白珈郡
李兴强
李祥
杨旭
崔维华
张清石
畅申
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenyang Institute of Automation of CAS
Original Assignee
Shenyang Institute of Automation of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenyang Institute of Automation of CAS filed Critical Shenyang Institute of Automation of CAS
Priority to CN201911250746.7A priority Critical patent/CN110927171A/en
Publication of CN110927171A publication Critical patent/CN110927171A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8806—Specially adapted optical and illumination features
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8854—Grading and classifying of flaws
    • G01N2021/8861—Determining coordinates of flaws
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G06T2207/30164—Workpiece; Machine component

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Pathology (AREA)
  • Artificial Intelligence (AREA)
  • Immunology (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Signal Processing (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

本发明公开了一种基于机器视觉的轴承滚子倒角面缺陷检测方法，针对轴承滚子在生产和运输过程中出现的倒角面缺陷问题，采集轴承滚子倒角面图像，利用基于深度卷积神经网络的深度学习算法建立倒角面缺件检测模型，实现倒角面缺陷的快速检测和位置定位。本发明运用深度学习模型和机器视觉算法对轴承滚子倒角面缺陷进行实时检测，具有快速检测、定位精准、识别准确度高的特点，能够代替传统人工检测出方法，满足滚子缺陷检测自动化的需求。

Figure 201911250746

The invention discloses a method for detecting chamfered surface defects of bearing rollers based on machine vision. Aiming at the chamfered surface defects of bearing rollers during production and transportation, images of the chamfered surfaces of bearing rollers are collected, and depth-based chamfered surface images are collected. The deep learning algorithm of the convolutional neural network establishes the detection model of the chamfered surface missing parts, and realizes the rapid detection and location of the chamfered surface defects. The invention uses the deep learning model and the machine vision algorithm to detect the defects of the bearing roller chamfer surface in real time, has the characteristics of rapid detection, accurate positioning and high recognition accuracy, can replace the traditional manual detection method, and satisfies the automatic detection of roller defects. demand.

Figure 201911250746

Description

Bearing roller chamfer surface defect detection method based on machine vision
Technical Field
The invention relates to the field of image detection, in particular to a bearing roller defect detection method based on machine vision.
Background
Bearing rollers are core components in bearings, and the defect of breakage of the rollers can cause accelerated wear and aging of the rollers and even operation failure of the bearings. Thus, the rollers need to be screened prior to assembly to remove defective rollers. The traditional screening method depends on manual detection, is complex in work, high in requirement on experience of detection personnel, easy to be influenced by environment, fatigue state of the detection personnel and the like, poor in detection efficiency and reliability, easy to cause defect omission detection and false detection, and only can qualitatively judge the defects and cannot quantitatively evaluate the defects. Therefore, the automatic bearing roller defect detection method has high application value and meets the practical requirements of intellectualization, high efficiency, high accuracy and good stability of roller defect detection.
Disclosure of Invention
The invention discloses a bearing roller chamfer surface defect detection method based on machine vision.
The technical scheme adopted by the invention for realizing the purpose is as follows:
a bearing roller chamfer surface defect detection method based on machine vision comprises the following steps:
1) establishing a machine vision acquisition system, acquiring a bearing roller image sample containing the chamfer surface defect, calibrating the defect position, and establishing a bearing roller chamfer surface defect database;
2) establishing a deep learning algorithm target detection model based on a deep convolutional neural network, and training the detection model according to image samples in a bearing roller chamfer surface defect database to obtain a network model suitable for bearing roller chamfer surface defect detection;
3) acquiring a roller image to be detected by using a visual acquisition system, extracting a roller chamfer surface outline by using an edge detection algorithm based on a metering model, judging whether a defect exists or not by using a trained defect detection network model, if the defect exists, positioning a defect area, and screening an area within a chamfer surface outline or intersected with the outline as a chamfer surface defect area;
4) and rechecking the defect area, extracting the area and gray value characteristics of the defect area by using a Blob analysis algorithm, judging whether the defect area is a false detection area, if not, feeding back the type and the position of the defect to a detection system database, and if the defect area is the false detection area, feeding back a defect-free result.
The machine vision acquisition system includes: the device comprises an area array camera, an outer-ring lens, a ball integral light source and a moving shaft, wherein the area array camera is connected with the outer-ring lens and is arranged vertically above a bearing roller, the ball integral light source is connected with the moving shaft, the central axes of the area array camera, the outer-ring lens and the ball integral light source are coincident with the central axis of the bearing roller, the moving shaft is controlled to drive the ball integral light source, the bearing roller chamfering surface is positioned in the irradiation area of the ball integral light source, and the image of the bearing roller chamfering surface is acquired through the area array camera.
The bearing roller chamfer surface defect database is established by the following steps:
1) acquiring bearing roller chamfer surface images including a defect-free image and a defective image by a machine vision acquisition system;
2) graying the image, adopting a threshold segmentation algorithm to segment and extract a to-be-detected region from the image, zooming the segmented to-be-detected region image, and storing the zoomed to-be-detected region image as an image sample;
3) selecting a zoom chamfer image sample containing a defect, marking the defect area by using a rectangular frame, and obtaining coordinate values a of four corners of the defect areaij,bij,cij,dijAnd i is 1,2,3 …, j is 1,2,3 …, i represents the serial number of the image sample, j represents the serial number of the labeling frame in the same image sample, coordinate value data of the four corners of all the labeling frames are stored, each group of coordinate values corresponds to one defect area sample, and a bearing roller chamfer surface defect sample database is established.
The step 2) is as follows:
the network model adopts a CNN model and consists of an input layer, a convolution layer, a pooling layer, a full-link layer and an output layer, and the process of training the network model is as follows:
1) scaling the marked defect image sample, and inputting the scaled defect image sample into the network model through the input layer;
2) performing convolution operation on the convolution layer to extract a characteristic diagram of a defect image sample, realizing nonlinear operation through an activation function to enhance the image characteristic fitting capability of the network, and obtaining the characteristic diagram of the defect image sample through calculation:
Figure BDA0002308971400000031
Figure BDA0002308971400000032
a characteristic diagram representing the output of the jth neural node in the ith layer of convolutional network, f is an activation function, MjFor the jth node feature map set,
Figure BDA0002308971400000033
is the ith characteristic diagram of the l-1 st layer,
Figure BDA0002308971400000034
the values of the convolution kernel weights are used as the values of the convolution kernel weights,
Figure BDA0002308971400000035
is a bias term;
3) inputting a sample characteristic diagram extracted from the convolutional layer into a pooling layer, establishing a sliding window, sliding on the sample characteristic diagram for average pooling, calculating a characteristic mean value of each window, inputting the sample characteristic diagram into a full-connection layer, refitting the sample characteristic diagram into a global characteristic diagram, and obtaining the global characteristic diagramGet the characteristic yj：yj＝f(∑iWij*xi+bj) F is an activation function, WijRepresenting respective neuron node weight values, xiAs input features, bjIs a bias value;
4) the network model inputs m defect samples, and the network total loss function can be expressed as:
Figure BDA0002308971400000036
w represents the weight of the network model, b represents the bias value of the network model, J (W, b; x)(i),y(i)) Representing the i-th defect sample loss function, with sample λ representing the network weight decay term, nlRepresenting the number of layers of the network, SlNumber of ganglion points, x, representing the l-th neural network(i)Predicted defect characteristics for the ith defect sample, y(i)The real defect characteristics of the ith defect sample. Through iterative training learning, the weight parameters of the network are corrected, the loss function is reduced, the network model is updated until the training is finished, and the defect detection model based on the deep convolutional neural network is obtained.
The step 3) is as follows:
the method comprises the steps of obtaining a roller chamfer surface and a partial outer diameter surface image in an image sample of a region to be detected, seeking a segmentation threshold value of the chamfer surface and the outer diameter surface by adopting an Otsu algorithm, segmenting a chamfer surface image region, then carrying out closed operation on the image region, taking a minimum inscribed circle to obtain a crude extraction contour of the chamfer surface, adding the crude extraction contour into a template, creating a metering model template, creating a plurality of adjacent quasi-rectangular regions with the height of h at intervals of l according to the initial contour position of the template, enabling the centers of the quasi-rectangular regions to be located on the initial contour and perpendicular to the initial contour, then determining edge boundary positions in the quasi-rectangular detection regions by using an RANSAC algorithm in each quasi-rectangular region, and finally connecting and fitting boundary points in all detected detection regions to obtain the accurate contour of the chamfer surface.
The step 4) is as follows:
taking a plurality of detection areas in the defect area in a fixed-scale sliding window mode, and dividing each detection area into a plurality of detection areasInputting the area image into the network model convolution layer to extract a characteristic diagram, inputting the characteristic diagram into the full-connection layer to obtain an output value y-f (sigma)kW*xk+ b), calculating the output value of the whole full-connection layer by adopting a Softmax function, wherein each defect classification corresponds to one node, and outputting the probability of the defect classification
Figure BDA0002308971400000041
ykRepresenting the probability, x, of a network model determining each defect class k corresponding to a certain detection areakInput value, x, representing the node corresponding to the k-th defectjThe input value of the corresponding node of the jth defect is n, the number of categories of defect classification is n, the weight value of the network model is W, the bias item of the network model is b, and the category with the highest probability and the probability exceeding the set threshold is the judged defect classification.
The invention has the following beneficial effects and advantages:
1. the invention replaces the traditional manual mode to detect the defects of the chamfer surface of the bearing roller, has high detection efficiency and strong stability, and effectively removes the conditions of defect omission and false detection caused by factors such as working environment, experience deviation of detection personnel, fatigue state and the like.
2. The invention improves the original vision acquisition system, adopts the lens outside the ring and the spherical integral light source, can effectively improve the acquisition integrity and the presentation effect of the bearing roller chamfer surface image, and improves the chamfer surface defect detection effect.
3. According to the invention, the intelligent prediction judgment is carried out on whether the roller chamfer surface has defects or not by adopting the deep convolution network model, so that the problems of insufficient defect rule description, difficult feature extraction, poor robustness and the like of the traditional algorithm are effectively solved, and the accuracy of defect detection is improved.
Drawings
FIG. 1 is a schematic diagram of a bearing roller defect detection system based on machine vision according to the present invention;
FIG. 2 is a structural diagram of a bearing roller defect detection vision acquisition system based on machine vision in the invention;
the system comprises an area-array camera 1, a 360-degree annular outer lens 2, a spherical integral light source 3, a motion shaft 4, a fixing frame 5 and a bearing roller 6, wherein the area-array camera is a three-dimensional spherical integral light source;
FIG. 3 is a flow chart of a bearing roller defect detection method based on machine vision according to the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples.
As shown in fig. 1, a method for detecting defects of a bearing roller chamfer surface based on machine learning includes the following steps:
1) establishing a machine vision acquisition system, acquiring a large number of bearing roller image samples containing the chamfer surface defects, manually calibrating the positions of the defects, and establishing a chamfer surface defect database;
2) establishing a deep learning algorithm target detection model based on a deep convolutional neural network, and training and optimizing the detection model by using defect samples in a defect database to obtain a network model suitable for detecting the defects of the bearing roller chamfer surface;
3) and (3) acquiring an image of the roller to be detected by using a visual acquisition system, and extracting the outline of the roller chamfer surface by using an edge detection algorithm based on a metering model. Judging whether a defect exists or not by using the trained and optimized defect detection network model, positioning a defect area, and screening an area within or intersected with the contour line of the chamfer surface as the chamfer surface defect area;
4) and rechecking the defect area, extracting the characteristics of the area, the gray value and the like of the defect by using a Blob analysis algorithm, judging whether the defect area is a false detection area or not, and storing the confirmed defect detection information.
As shown in fig. 2, is a machine vision acquisition system. The visual acquisition system consists of an area-array camera, a 360-degree lens outside a ring, a spherical integral light source and a motion shaft. The camera is connected with the lens outside the ring and is arranged above the roller vertically, and the spherical integral light source is connected with the moving shaft, so that the central axes of the camera, the lens and the light source are coincided with the central axis of the roller. And controlling the motion shaft to drive the ball integral light source, so that the bearing roller chamfer surface is positioned in a light source irradiation area, and acquiring a chamfer surface image through a camera.
Fig. 3 is a flowchart of a bearing roller defect detection method based on machine vision according to the present invention.
Establishing a bearing roller chamfer surface defect database, wherein the specific implementation process comprises the following steps:
1) acquiring a large number of roller chamfer surface images by a vision acquisition system, wherein the images need to comprise a defect-free image sample and a defect-carrying image sample;
2) graying the image sample, selecting a proper grayscale threshold by adopting a threshold segmentation algorithm, and segmenting and extracting the region to be detected from the image background. And scaling the segmented to-be-detected region image into 256 × 256, and storing as an image sample.
3) Selecting a zoom chamfer image sample containing a defect, manually marking the defect area by using a rectangular frame, and obtaining coordinate values a of four corners of the defect areaij,bij,cij,dijI is 1,2,3 … n, j is 1,2,3 … n, i indicates the number of the image sample, and j indicates the number of the label box in the same image sample. And storing coordinate value data of four corners of all the marking frames, wherein each group of coordinate values corresponds to one defect area sample, and establishing a bearing roller chamfer surface defect sample database.
And training a defect detection network model by using a defect sample database. The network model adopts a CNN model and consists of an input layer, a convolution layer, a pooling layer, a full-link layer and an output layer, and the specific process of training the network model comprises the following steps:
1) uniformly scaling the defect samples into 32 × 32 sizes, and inputting the defect samples into a network through an input layer;
2) convolution operation is carried out through the convolutional layer to extract the characteristics of the sample, nonlinear operation is realized through an activation function to enhance the image characteristic fitting capacity of the network, and a sample characteristic diagram is obtained through calculation:
Figure BDA0002308971400000061
Figure BDA0002308971400000062
Figure BDA0002308971400000063
representing the l-th convolutional network inputThe j th feature graph is shown, f is an activation function, MjFor the set of feature maps, W is the convolution kernel value,
Figure BDA0002308971400000064
is the ith feature map of layer l-1, and b is the bias term.
3) Inputting the characteristic graph extracted from the convolutional layer into a pooling layer, establishing a sliding window in a scale of 2 x 2, sliding on the characteristic graph for average pooling, solving a characteristic mean value of each window, and reducing the calculated amount of characteristic values. Then inputting the feature graph into a full-connection layer, refitting the feature graph into a global feature graph, and obtaining features: y isj＝f(∑iWij*xi+bj) F is an activation function, WijRepresenting respective neuron node weight values, xiAs input features, bjIs an offset value.
4) The network model inputs m defect samples, and the network total loss function can be expressed as:
Figure BDA0002308971400000071
w represents the weight of the network model, b represents the bias value of the network model, J (W, b; x)(i),y(i)) Representing the i-th defect sample loss function, with sample λ representing the network weight decay term, nlRepresenting the number of layers of the network, SlRepresenting the number of ganglion points of the l-th neural network. And continuously correcting the weight parameters of the network through continuous iterative training and learning to ensure that the loss function is smaller and smaller, and updating the optimized network model until the training is finished. And finally obtaining a defect detection model based on the deep convolutional neural network.
And extracting the outline of the chamfer surface by adopting an edge detection algorithm based on a metering model. The image sample of the region to be detected comprises a roller chamfer surface image and a partial outer diameter surface image, the segmentation threshold values of the chamfer surface and the outer diameter surface are searched by adopting an Otsu algorithm, a chamfer surface image region is segmented, then the image region is subjected to closed operation, the minimum inscribed circle is taken, and the crude extraction contour of the chamfer surface is obtained. And adding the crude extraction contour into the template, creating a metering model template, and modifying the control measurement parameters of the metering model template according to requirements. And creating a plurality of adjacent rectangle-like areas with the height h at intervals of l according to the initial contour position of the template, wherein the centers of the rectangle-like areas are positioned on the initial contour and are perpendicular to the initial contour line. Then, in each small rectangular area, the RANSAC algorithm is used to determine the edge boundary position in the small rectangular detection area. And finally, connecting and fitting the detected boundary points in all the detection areas to obtain the accurate profile of the chamfer surface.
And detecting the defects by using the defect detection network model after training optimization. Taking a plurality of small detection areas in an image area to be detected in a fixed-scale sliding window mode, inputting each area image into a network model convolution layer to extract a characteristic diagram, and then inputting the characteristic diagram into a full-connection layer to obtain an output value y-f (sigma)iW*xi+ b), calculating the probability of outputting the defect category by adopting a Softmax function to the output value of the whole full-connection layer
Figure BDA0002308971400000072
Figure BDA0002308971400000073
yiRepresenting the probability, x, that the network model determines each small image area for each defect class ikAnd (4) representing the input value of the node corresponding to the kth defect, wherein the class with the highest probability and the probability exceeding a set threshold is the judged defect type.

Claims (6)

1.一种基于机器视觉的轴承滚子倒角面缺陷检测方法，其特征在于，包括以下步骤：1. a bearing roller chamfered surface defect detection method based on machine vision, is characterized in that, comprises the following steps: 1)建立机器视觉采集系统，采集含倒角面缺陷的轴承滚子图像样本，并对缺陷位置进行标定，建立轴承滚子倒角面缺陷数据库；1) Establish a machine vision acquisition system, collect image samples of bearing rollers with chamfer surface defects, calibrate the defect positions, and establish a bearing roller chamfer surface defect database; 2)建立基于深度卷积神经网络的深度学习算法目标检测模型，根据轴承滚子倒角面缺陷数据库中的图像样本对检测模型进行训练，得到适用于轴承滚子倒角面缺陷检测的网络模型；2) Establish a deep learning algorithm target detection model based on deep convolutional neural network, train the detection model according to the image samples in the bearing roller chamfer surface defect database, and obtain a network model suitable for bearing roller chamfer surface defect detection ; 3)使用视觉采集系统采集待检测滚子图像，利用基于计量模型的边缘检测算法提取滚子倒角面轮廓，利用训练后的缺陷检测网络模型，判断是否存在缺陷，若模型判定存在缺陷，对缺陷区域进行定位，筛选在倒角面轮廓线以内或者与轮廓线相交的区域为倒角面缺陷区域；3) Use the visual acquisition system to collect the image of the roller to be detected, use the edge detection algorithm based on the metrology model to extract the contour of the chamfered surface of the roller, and use the trained defect detection network model to judge whether there is a defect. The defect area is located, and the area within or intersecting the contour line of the chamfered surface is selected as the chamfered surface defect area; 4)对缺陷区域进行复检，利用Blob分析算法提取缺陷区域的面积、灰度值特征，判断是否为误检区域，若不是误检区域，则将缺陷的类别和位置反馈给检测系统数据库，若是误检区域，则返回无缺陷结果。4) Re-inspect the defect area, use the Blob analysis algorithm to extract the area and gray value characteristics of the defect area, and judge whether it is a false detection area. In the case of a false detection area, no defect result is returned. 2.根据权利要求1所述的一种基于机器视觉的轴承滚子倒角面缺陷检测方法，其特征在于，所述机器视觉采集系统包括：面阵相机、环外侧镜头、球积分光源和运动轴，其中，面阵相机与环外侧镜头连接，安装于轴承滚子垂直上方，球积分光源与运动轴连接，面阵相机、环外侧镜头和球积分光源的中轴线与轴承滚子中轴线重合，控制运动轴带动球积分光源，使轴承滚子倒角面处于球积分光源照射区域内，通过面阵相机采集轴承滚子倒角面图像。2. A method for detecting defects on a bearing roller chamfer surface based on machine vision according to claim 1, wherein the machine vision acquisition system comprises: an area scan camera, an outer ring lens, a spherical integral light source and a motion Shaft, in which the area scan camera is connected to the outer lens of the ring and is installed vertically above the bearing roller, the spherical integral light source is connected to the motion axis, and the central axis of the area scan camera, the outer lens of the ring and the spherical integral light source coincides with the central axis of the bearing roller , control the motion axis to drive the spherical integral light source, so that the chamfered surface of the bearing roller is in the irradiation area of the spherical integral light source, and the image of the bearing roller chamfered surface is collected by the area array camera. 3.根据权利要求1所述的一种基于机器视觉的轴承滚子倒角面缺陷检测方法，其特征在于，所述轴承滚子倒角面缺陷数据库，其建立过程为：3. a kind of bearing roller chamfer surface defect detection method based on machine vision according to claim 1, is characterized in that, described bearing roller chamfer surface defect database, its establishment process is: 1)通过机器视觉采集系统采集轴承滚子倒角面图像，包括无缺陷图像和带缺陷图像；1) Collect images of bearing roller chamfer surface through machine vision acquisition system, including non-defective images and defective images; 2)将图像灰度化，采用阈值分割算法，从图像中分割提取待检测区域，将分割出的待检测区域图像进行缩放，并保存为图像样本；2) grayscale the image, adopt a threshold segmentation algorithm, segment and extract the area to be detected from the image, scale the segmented image of the area to be detected, and save it as an image sample; 3)选择含有缺陷的缩放倒角面图像样本，用矩形框标注缺陷区域，获得缺陷区域四个角的坐标值aij,bij,cij,dij,i＝1,2,3…,j＝1,2,3…,i表示图像样本的序号，j表示同一个图像样本中的标注框序号，保存所有标注框四个角的坐标值数据，每一组坐标值对应一个缺陷区域样本，建立轴承滚子倒角面缺陷样本数据库。3) Select the image sample of the scaled chamfered surface with defects, mark the defect area with a rectangular frame, and obtain the coordinate values of the four corners of the defect area a ij , b ij , c ij , d ij , i=1, 2, 3..., j=1,2,3...,i represents the serial number of the image sample, j represents the serial number of the labeling frame in the same image sample, saves the coordinate value data of all the four corners of the labeling frame, and each set of coordinate values corresponds to a defect area sample , to establish a sample database of bearing roller chamfer surface defects. 4.根据权利要求1所述的一种基于机器视觉的轴承滚子倒角面缺陷检测方法，其特征在于，步骤2)为：4. a kind of bearing roller chamfer surface defect detection method based on machine vision according to claim 1 is characterized in that, step 2) is: 网络模型采用CNN模型，由输入层、卷积层、池化层、全连接层和输出层组成，训练网络模型的过程为：The network model adopts the CNN model, which consists of an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer. The process of training the network model is as follows: 1)将标注后的缺陷图像样本进行缩放，并由输入层输入网络模型；1) The labeled defect image samples are scaled and input to the network model by the input layer; 2)通过卷积层进行卷积运算提取缺陷图像样本的特征图，通过激活函数实现非线性运算来增强网络的图像特征拟合能力，计算获得缺陷图像样本特征图为：
Figure FDA0002308971390000021
Figure FDA0002308971390000022
表示第l层卷积网络中第j个神经节点输出的特征图，f为激活函数，Mj为第j个节点特征图集合，
Figure FDA0002308971390000023
是第l-1层的第i个特征图，
Figure FDA0002308971390000024
为卷积核权值，
Figure FDA0002308971390000025
为偏置项；2) The feature map of the defect image sample is extracted by convolution operation through the convolution layer, and the nonlinear operation is realized by the activation function to enhance the image feature fitting ability of the network, and the feature map of the defect image sample obtained by calculation is:
Figure FDA0002308971390000021
Figure FDA0002308971390000022
Represents the feature map output by the jth neural node in the lth layer convolutional network, f is the activation function, Mj is the jth node feature map set,
Figure FDA0002308971390000023
is the ith feature map of the l-1th layer,
Figure FDA0002308971390000024
is the convolution kernel weight,
Figure FDA0002308971390000025
is the bias term; 3)将卷积层提取的样本特征图输入池化层中，建立滑动窗口，在样本特征图上滑动进行平均池化，求取每个窗口的特征均值，然后将样本特征图输入全连接层，重新拟合变成一个全局特征图，获得特征yj：yj＝f(∑iWij*xi+bj)，f为激活函数，Wij表示相应神经元节点权重值，xi为输入特征，bj为偏置值；3) Input the sample feature map extracted by the convolutional layer into the pooling layer, establish a sliding window, slide on the sample feature map for average pooling, obtain the feature mean of each window, and then input the sample feature map into the fully connected layer , re-fit into a global feature map, and obtain the feature y j : y j =f(∑ i W ij *x i +b j ), f is the activation function, W ij represents the weight value of the corresponding neuron node, x i is the input feature, b j is the bias value; 4)网络模型输入m个缺陷样本,网络总体损失函数可以表示为：
Figure FDA0002308971390000031
W代表网络模型的权重，b表示网络模型的偏置值，J(W,b；x(i),y(i))代表第i个缺陷样本损失函数，样本λ表示网络权重衰减项，nl代表网络的层数，Sl代表第l层神经网络的神经节点数，x(i)为第i个缺陷样本的预测缺陷特征，y(i)为第i个缺陷样本的真实缺陷特征。通过迭代训练学习，从而修正网络的权重参数，减小损失函数，更新网络模型，直至训练结束，获得基于深度卷积神经网络的缺陷检测模型。4) The network model inputs m defect samples, and the overall loss function of the network can be expressed as:
Figure FDA0002308971390000031
W represents the weight of the network model, b represents the bias value of the network model, J(W,b; x (i) , y (i) ) represents the ith defect sample loss function, sample λ represents the network weight decay term, n l represents the number of layers of the network, S l represents the number of neural nodes of the lth layer of neural network, x (i) is the predicted defect feature of the ith defect sample, y (i) is the real defect feature of the ith defect sample. Through iterative training and learning, the weight parameters of the network are corrected, the loss function is reduced, and the network model is updated until the end of the training, and a defect detection model based on a deep convolutional neural network is obtained. 5.根据权利要求1所述的一种基于机器视觉的轴承滚子倒角面缺陷检测方法，其特征在于，步骤3)为：5. a kind of bearing roller chamfer surface defect detection method based on machine vision according to claim 1, is characterized in that, step 3) is: 待检测区域图像样本中包含滚子倒角面和部分外径面图像，采用Otsu算法寻求倒角面和外径面的分割阈值，分割出倒角面图像区域，然后对图像区域进行闭运算，取最小内接圆，得到倒角面的粗提取轮廓，将粗提取轮廓加入到模板中，创建计量模型模板，根据模板初始轮廓位置以间隔l创建多个相邻的高度为h的类矩形区域，使类矩形区域的中心位于初始轮廓上，而且垂直于初始轮廓线，然后在每个类矩形区域中，使用RANSAC算法确定类矩形检测区域中的边缘分界位置，最后将检测到的所有检测区域中的分界点进行连接和拟合，得到倒角面的准确轮廓。The image sample of the area to be detected contains the image of the roller chamfer surface and part of the outer diameter surface. The Otsu algorithm is used to find the segmentation threshold of the chamfer surface and the outer diameter surface, and the image area of the chamfer surface is segmented, and then the image area is closed. Take the smallest inscribed circle to obtain the rough extraction outline of the chamfered surface, add the rough extraction outline to the template, create a measurement model template, and create multiple adjacent rectangle-like regions with a height of h at intervals of l according to the initial outline position of the template , so that the center of the rectangle-like area is located on the initial contour and perpendicular to the initial contour line, and then in each rectangle-like area, the RANSAC algorithm is used to determine the edge boundary position in the rectangle-like detection area, and finally all detected areas are detected. Connect and fit the demarcation points in the chamfered surface to obtain the accurate contour of the chamfered surface. 6.根据权利要求1所述的一种基于机器视觉的轴承滚子倒角面缺陷检测方法，其特征在于，步骤4)为：6. a kind of bearing roller chamfer surface defect detection method based on machine vision according to claim 1, is characterized in that, step 4) is: 在缺陷区域以固定尺度滑动窗的方式取若干个检测区域，将每个区域图像输入网络模型卷积层提取特征图，然后输入全连接层，得到输出值y＝f(∑kW*xk+b)，采用Softmax函数对整个全连接层输出值进行计算，每个缺陷分类对应一个节点，输出缺陷类别的概率
Figure FDA0002308971390000041
yk表示网络模型判定某个检测区域对应每种缺陷类别k的概率，xk表示第k种缺陷对应节点的输入值，xj为第j种缺陷对应节点的输入值，n表示缺陷分类的类别数，W为网络模型权重值，b为网络模型偏置项，概率最大并且概率超过设定阈值的类别就是判定的缺陷分类。In the defect area, several detection areas are taken in the form of fixed-scale sliding windows, and the image of each area is input into the convolutional layer of the network model to extract the feature map, and then input into the fully connected layer to obtain the output value y=f(∑ k W*x k +b), use the Softmax function to calculate the output value of the entire fully connected layer, each defect classification corresponds to a node, and output the probability of the defect category
Figure FDA0002308971390000041
y k represents the probability that the network model determines that a detection area corresponds to each defect category k, x k represents the input value of the node corresponding to the kth defect, x j is the input value of the node corresponding to the jth defect, and n represents the defect classification. The number of categories, W is the weight value of the network model, b is the bias term of the network model, and the category with the highest probability and the probability exceeding the set threshold is the determined defect classification.
CN201911250746.7A 2019-12-09 2019-12-09 A method for detecting defects on the chamfered surface of bearing rollers based on machine vision Pending CN110927171A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911250746.7A CN110927171A (en) 2019-12-09 2019-12-09 A method for detecting defects on the chamfered surface of bearing rollers based on machine vision

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911250746.7A CN110927171A (en) 2019-12-09 2019-12-09 A method for detecting defects on the chamfered surface of bearing rollers based on machine vision

Publications (1)

Publication Number Publication Date
CN110927171A true CN110927171A (en) 2020-03-27

Family

ID=69858438

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911250746.7A Pending CN110927171A (en) 2019-12-09 2019-12-09 A method for detecting defects on the chamfered surface of bearing rollers based on machine vision

Country Status (1)

Country Link
CN (1) CN110927171A (en)

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111929314A (en) * 2020-08-26 2020-11-13 湖北汽车工业学院 Wheel hub weld visual detection method and detection system
CN112129783A (en) * 2020-09-22 2020-12-25 广州番禺职业技术学院 Defect detection device and detection method at the bottom of transparent medicine bottle based on deep learning
CN112304960A (en) * 2020-12-30 2021-02-02 中国人民解放军国防科技大学 High-resolution image object surface defect detection method based on deep learning
CN112541910A (en) * 2020-12-23 2021-03-23 中国工程物理研究院机械制造工艺研究所 End face gap detection method, device, equipment and medium based on deep learning
CN113298857A (en) * 2021-05-20 2021-08-24 聚时科技(上海)有限公司 Bearing defect detection method based on neural network fusion strategy
CN113376172A (en) * 2021-07-05 2021-09-10 四川大学 Welding seam defect detection system based on vision and eddy current and detection method thereof
CN113466235A (en) * 2021-03-19 2021-10-01 江苏立讯机器人有限公司 Visual inspection module, defect inspection workstation and defect inspection method
CN113744297A (en) * 2021-08-31 2021-12-03 西部超导材料科技股份有限公司 A kind of MRI superconducting wire braided layer surface defect detection method and detection device
CN114140400A (en) * 2021-11-16 2022-03-04 湖北中烟工业有限责任公司 Method for detecting cigarette packet seal defects based on RANSAC and CNN algorithms
CN114397306A (en) * 2022-03-25 2022-04-26 南方电网数字电网研究院有限公司 Power grid grading ring hypercomplex category defect multi-stage model joint detection method
CN114654315A (en) * 2022-02-17 2022-06-24 杭州深度视觉科技有限公司 Machine vision detection system and method for poor grinding of tapered roller base surface
CN115351598A (en) * 2022-10-17 2022-11-18 南通钜德智能科技有限公司 Numerical control machine tool bearing detection method
CN115494659A (en) * 2022-08-10 2022-12-20 北京兆维电子(集团)有限责任公司 Liquid crystal panel detection method and system
CN115494065A (en) * 2022-04-28 2022-12-20 浙江大学台州研究院 A defect judgment method and device for pipe flaring based on positive light image
WO2023036557A1 (en) * 2021-09-08 2023-03-16 Aktiebolaget Skf Device for predicting the evolution of a defect of a bearing, associated system and method
CN119205721A (en) * 2024-10-31 2024-12-27 温州科技职业学院 Laser stripe detection method and system for steel components based on artificial intelligence algorithm

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101526484A (en) * 2009-04-13 2009-09-09 江南大学 Bearing defect detecting technique based on embedded-type machine vision
US20150049970A1 (en) * 2012-10-15 2015-02-19 Nanolab, Inc. Sensor for wear measurement, method for making same, and method for operating same
CN104502356A (en) * 2015-01-05 2015-04-08 江苏大学 Automatic detection method for defects of inner surface of sliding bearing on basis of computer vision
CN105118044A (en) * 2015-06-16 2015-12-02 华南理工大学 Method for automatically detecting defects of wheel-shaped cast product
CN107749067A (en) * 2017-09-13 2018-03-02 华侨大学 Fire hazard smoke detecting method based on kinetic characteristic and convolutional neural networks
CN109187581A (en) * 2018-07-12 2019-01-11 中国科学院自动化研究所 The bearing finished products plate defects detection method of view-based access control model
CN109557110A (en) * 2019-01-11 2019-04-02 新昌浙江工业大学科学技术研究院 The full surface blemish detection device of bearing ring and method based on machine vision

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101526484A (en) * 2009-04-13 2009-09-09 江南大学 Bearing defect detecting technique based on embedded-type machine vision
US20150049970A1 (en) * 2012-10-15 2015-02-19 Nanolab, Inc. Sensor for wear measurement, method for making same, and method for operating same
CN104502356A (en) * 2015-01-05 2015-04-08 江苏大学 Automatic detection method for defects of inner surface of sliding bearing on basis of computer vision
CN105118044A (en) * 2015-06-16 2015-12-02 华南理工大学 Method for automatically detecting defects of wheel-shaped cast product
CN107749067A (en) * 2017-09-13 2018-03-02 华侨大学 Fire hazard smoke detecting method based on kinetic characteristic and convolutional neural networks
CN109187581A (en) * 2018-07-12 2019-01-11 中国科学院自动化研究所 The bearing finished products plate defects detection method of view-based access control model
CN109557110A (en) * 2019-01-11 2019-04-02 新昌浙江工业大学科学技术研究院 The full surface blemish detection device of bearing ring and method based on machine vision

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
乔丽: "基于CNN的工件缺陷检测方法研究及系统设计", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *

Cited By (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111929314A (en) * 2020-08-26 2020-11-13 湖北汽车工业学院 Wheel hub weld visual detection method and detection system
CN112129783A (en) * 2020-09-22 2020-12-25 广州番禺职业技术学院 Defect detection device and detection method at the bottom of transparent medicine bottle based on deep learning
CN112129783B (en) * 2020-09-22 2024-03-29 广州番禺职业技术学院 Transparent medicine bottle bottom defect detection method based on deep learning
CN112541910B (en) * 2020-12-23 2023-07-11 中国工程物理研究院机械制造工艺研究所 End face gap detection method, device, equipment and medium based on deep learning
CN112541910A (en) * 2020-12-23 2021-03-23 中国工程物理研究院机械制造工艺研究所 End face gap detection method, device, equipment and medium based on deep learning
CN112304960A (en) * 2020-12-30 2021-02-02 中国人民解放军国防科技大学 High-resolution image object surface defect detection method based on deep learning
CN113466235A (en) * 2021-03-19 2021-10-01 江苏立讯机器人有限公司 Visual inspection module, defect inspection workstation and defect inspection method
CN113298857A (en) * 2021-05-20 2021-08-24 聚时科技(上海)有限公司 Bearing defect detection method based on neural network fusion strategy
CN113376172A (en) * 2021-07-05 2021-09-10 四川大学 Welding seam defect detection system based on vision and eddy current and detection method thereof
CN113744297B (en) * 2021-08-31 2024-12-03 西部超导材料科技股份有限公司 A method and device for detecting surface defects of MRI superconducting wire braiding layer
CN113744297A (en) * 2021-08-31 2021-12-03 西部超导材料科技股份有限公司 A kind of MRI superconducting wire braided layer surface defect detection method and detection device
WO2023036557A1 (en) * 2021-09-08 2023-03-16 Aktiebolaget Skf Device for predicting the evolution of a defect of a bearing, associated system and method
CN114140400B (en) * 2021-11-16 2024-04-12 湖北中烟工业有限责任公司 Method for detecting cigarette packet label defect based on RANSAC and CNN algorithm
CN114140400A (en) * 2021-11-16 2022-03-04 湖北中烟工业有限责任公司 Method for detecting cigarette packet seal defects based on RANSAC and CNN algorithms
CN114654315A (en) * 2022-02-17 2022-06-24 杭州深度视觉科技有限公司 Machine vision detection system and method for poor grinding of tapered roller base surface
CN114397306A (en) * 2022-03-25 2022-04-26 南方电网数字电网研究院有限公司 Power grid grading ring hypercomplex category defect multi-stage model joint detection method
CN115494065A (en) * 2022-04-28 2022-12-20 浙江大学台州研究院 A defect judgment method and device for pipe flaring based on positive light image
CN115494659A (en) * 2022-08-10 2022-12-20 北京兆维电子(集团)有限责任公司 Liquid crystal panel detection method and system
CN115494659B (en) * 2022-08-10 2025-07-08 北京兆维电子(集团)有限责任公司 Detection method and system for liquid crystal display panel
CN115351598B (en) * 2022-10-17 2024-01-09 安徽金锘轴承制造有限公司 Method for detecting bearing of numerical control machine tool
CN115351598A (en) * 2022-10-17 2022-11-18 南通钜德智能科技有限公司 Numerical control machine tool bearing detection method
CN119205721A (en) * 2024-10-31 2024-12-27 温州科技职业学院 Laser stripe detection method and system for steel components based on artificial intelligence algorithm
CN119205721B (en) * 2024-10-31 2025-02-18 温州科技职业学院 Steel member laser stripe detection method and system based on artificial intelligence algorithm

Similar Documents

Publication Publication Date Title
CN110927171A (en) A method for detecting defects on the chamfered surface of bearing rollers based on machine vision
CN111862064B (en) Silver wire surface flaw identification method based on deep learning
CN106127204B (en) A multi-directional water meter reading area detection algorithm based on fully convolutional neural network
CN108346144B (en) Automatic bridge crack monitoring and identifying method based on computer vision
CN110853015A (en) Aluminum profile defect detection method based on improved Faster-RCNN
CN114092389A (en) A surface defect detection method for glass panels based on small sample learning
CN113393426B (en) Steel rolling plate surface defect detection method
CN110136101B (en) Tire X-ray defect detection method based on twinning distance comparison
CN113487533B (en) A digital inspection system and method for parts assembly quality based on machine learning
CN108918527A (en) A kind of printed matter defect inspection method based on deep learning
CN110992349A (en) Underground pipeline abnormity automatic positioning and identification method based on deep learning
CN114119554A (en) A method and device for surface micro-defect detection based on convolutional neural network
CN111667455A (en) AI detection method for various defects of brush
CN113643268A (en) Industrial product defect quality inspection method and device based on deep learning and storage medium
CN113469951A (en) Hub defect detection method based on cascade region convolutional neural network
CN110648310A (en) Weakly Supervised Casting Defect Recognition Method Based on Attention Mechanism
CN116205876B (en) Unsupervised notebook appearance defect detection method based on multi-scale standardized flow
CN113420619A (en) Remote sensing image building extraction method
CN117952904A (en) Large equipment surface defect positioning and measuring method based on combination of image and point cloud
CN113962951B (en) Training method and device for detecting segmentation model, and target detection method and device
CN115082444B (en) Copper pipe weld defect detection method and system based on image processing
CN113610035A (en) A method for segmentation and identification of weeds in rice tillering stage based on improved encoder-decoder network
CN113674216A (en) Subway tunnel disease detection method based on deep learning
CN113313678A (en) Automatic sperm morphology analysis method based on multi-scale feature fusion
CN118967672A (en) Industrial defect detection method, system, device and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20200327