CN111257422B - Wheel axle defect identification model construction method and defect identification method based on machine vision - Google Patents
Wheel axle defect identification model construction method and defect identification method based on machine vision Download PDFInfo
- Publication number
- CN111257422B CN111257422B CN202010128206.8A CN202010128206A CN111257422B CN 111257422 B CN111257422 B CN 111257422B CN 202010128206 A CN202010128206 A CN 202010128206A CN 111257422 B CN111257422 B CN 111257422B
- Authority
- CN
- China
- Prior art keywords
- defect
- image
- machine vision
- sample library
- steps
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/04—Analysing solids
- G01N29/048—Marking the faulty objects
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
Abstract
The invention relates to a wheel axle defect recognition model construction method and a defect recognition method based on machine vision, wherein the method constructs a defect recognition model based on a machine vision technology, and can continuously improve the accuracy of defect recognition through continuous autonomous learning and training; meanwhile, the machine vision is not limited by personnel experience and quality, so that the defect recognition accuracy, the operation efficiency and the operation quality are greatly improved, and misjudgment and missed judgment of defects are effectively reduced. The identification method comprises the following steps: performing defect identification by using a defect identification model; screening the identified defects; displaying and alarming the screened defects; and continuously perfecting the defect identification model by utilizing new defect data.
Description
Technical Field
The invention is suitable for the field of ultrasonic nondestructive inspection, and particularly relates to a wheel axle defect identification model construction method and a defect identification method based on machine vision.
Background
At present, ultrasonic detection is adopted for wheel axle flaw detection, and manual analysis and detection data are taken as main means. The axle ultrasonic detection results are displayed in the modes of A display, B display and C display, and then the display images are observed by naked eyes to judge whether the axle has defects. The defect judgment result depends on the quality and experience of the inspector to a great extent, so that missed judgment, misjudgment and low efficiency are easily caused; due to the quality and experience of the detection personnel, the analysis results of different detection personnel on the same detection data are possibly inconsistent, so that accurate basis cannot be provided for decision making; in addition, as rail transit rolling stock increases year by year and the types of wheel axles of the rolling stock are various, the contradiction between the increase of the wheel axle detection tasks and the insufficient number of detection personnel is increasingly sharp; the efficiency of nondestructive test operation has been restricted to the insufficient quantity of detection personnel, is unfavorable for carrying out the short-term test to a large amount of shaft, and then influences production efficiency. Meanwhile, the existing automatic analysis method for wheel axle flaw detection cannot meet the requirements of wheel axle flaw detection operation technology, defect identification accuracy and consistency. There is a need in the art for a rapid and accurate automated analysis of ultrasound test data that addresses the above-described problems.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a vehicle defect identification model construction method and a defect identification method based on machine vision, wherein the method constructs a defect identification model based on the machine vision technology, and can continuously improve the accuracy of defect identification through continuous autonomous learning and training; meanwhile, the machine vision is not limited by personnel experience and quality, so that the defect recognition accuracy, the operation efficiency and the operation quality are greatly improved, and misjudgment and missed judgment of defects are effectively reduced.
The invention provides a wheel axle defect identification model construction method based on machine vision, which comprises the following steps:
step S100, acquiring detection data of a wheel axle by utilizing ultrasonic flaw detection equipment, and storing the detection data in a binary file;
step 200, obtaining sample data through the binary file, and constructing a defect positive sample library and a defect negative sample library;
step S300, preprocessing a defect positive sample library to generate a defect positive sample template file;
and step 400, constructing a defect identification model through the defect positive sample template file, the sample matching algorithm and the defect negative sample library.
Further, the ultrasonic flaw detection device acquires detection data according to an axle type or wheel type, corresponding channel configuration and scanning mode.
Further, the step S200 further includes the steps of:
step S210, clustering analysis is carried out on the binary files;
step S220, extracting images from the binary files after cluster analysis;
step S230, capturing a defect position image according to the image characteristics of the defect to obtain sample data;
step S240, defective sample data form a defective positive sample library, and non-defective sample data form a defective negative sample library.
Further, the defect positive sample library comprises defect images; the defect negative sample library comprises a non-defect image, a noise image and a transitional arc image.
Further, the defect image comprises a flat bottom hole image, a transverse defect image, a longitudinal defect image, an outer surface defect image and an inner surface defect image.
Further, the preprocessing includes normalization processing, i.e., unifying the size and format of the defect image.
The invention also provides a wheel axle defect identification method based on machine vision, which comprises the following steps:
step Z100, acquiring detection data of the wheel axle by utilizing ultrasonic flaw detection equipment, storing the detection data in a binary file, and extracting a detection image from the binary file;
step Z200, performing defect recognition on the detection image by using the wheel axle defect recognition model;
step Z300, performing defect screening on the defect identification result according to specific parameters;
and step Z400, displaying the screened defects and alarming.
Further, the step Z300 further includes: the specific parameters include at least one of defect type, threshold value, location.
Further, the wheel axle defect identification method further includes: and step Z500, the wheel axle defect recognition model is constructed in a supplementary mode by using the image contained in the defect recognition result.
Further, step Z500 further includes: manually selecting to add the image contained in the identification result to the defect positive sample library or the defect negative sample library; the computer can also automatically increase the image contained in the defect identification result to the defect positive sample library or the defect negative sample library according to the comparison result of the defect image matching degree and the matching degree threshold value.
The invention has the following beneficial effects: first, the invention introduces artificial intelligence technology into the field of nondestructive testing, promotes the development of the nondestructive testing technology to the intelligent direction, and greatly improves the quality and efficiency of nondestructive testing. Acquiring detection data of the wheel axle by utilizing ultrasonic flaw detection equipment based on a machine vision technology, acquiring sample data according to image characteristics of wheel axle defects, and further constructing a defect identification model; the method of machine learning is adopted to obtain the characteristic information of the defects, so that the requirements on quality and experience of detection personnel are reduced; by adopting a computer analysis method, the problems of misjudgment, missed judgment and the like caused by artificial subjective factors during manual data analysis are solved, and the detection quality is improved, so that the automatic analysis of ultrasonic detection data is realized rapidly and accurately.
Secondly, the wheel axle defect recognition model is utilized to carry out defect recognition, and a large amount of data analysis work borne by the original manual work is processed by a computer, so that the data processing speed is greatly improved, the rapid nondestructive detection of a large amount of workpieces is realized, the defect recognition efficiency is greatly improved, and the production efficiency of the wheel axle is improved; meanwhile, the defect sample data are continuously supplemented and perfected along with the increase of detection samples and the enrichment of defect types, and the defect identification model of the wheel axle is continuously supplemented and optimized through the defect identification result, so that the accuracy of data analysis is further improved.
Drawings
FIG. 1 is a flow chart of a machine vision based defect identification model construction method of the present invention.
FIG. 2 is a flow chart of a machine vision based defect identification method of the present invention.
Fig. 3 is a schematic diagram of the rim spoke defect recognition result.
Fig. 4 is a schematic diagram of the results of identifying defects of a hollow axle before screening.
Fig. 5 is a schematic diagram of the result of identifying defects of the screened hollow axle.
Detailed Description
The following description of the embodiments of the present invention will be made apparent and fully in view of the accompanying drawings, in which some, but not all embodiments of the invention are shown. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
Referring to fig. 1: a wheel axle defect identification model construction method based on machine vision comprises the following steps:
and S100, acquiring detection data of the wheel axle by using ultrasonic flaw detection equipment, and storing the detection data in a binary file.
Because the geometric characteristics of the rim, the web, the rim and the hollow shaft of the wheel are different, the flaw detection modes are different, and the detected images are different. The ultrasonic wheel axle flaw detection equipment acquires detection data of the wheel axle parts according to different axle types or wheel types, corresponding channels are configured and scanning modes are carried out, the detection data are stored in an upper computer of the flaw detection equipment in the form of binary files, and the binary files contain information such as positions, amplitudes, detection parameters, images and the like.
The channels are probes, one channel is one probe, each probe scans according to a certain direction and angle, a plurality of probes scan together in one scanning process, and each probe has an independent scanning result.
The scanning mode comprises A scanning, B scanning and C scanning, and the output results are also divided into A scanning display, B scanning display and C scanning display; the A scanning display is that the X axis represents time, the Y axis represents the ultrasonic signal display mode of amplitude, namely the X axis represents time, and the Y axis represents amplitude; b scanning is shown as a cross-sectional view of a detected piece drawn according to the relation between the sound path length of an echo signal with amplitude in a preset range and the sound beam axis position when the probe scans along only one direction, and for a hollow shaft, an X axis is expressed as an axial direction, and the sound path length of a Y axis is expressed as a radius; c scanning is displayed as a two-dimensional plane display of the detected piece, and according to the scanning position of the probe, the existence of echo signals with the amplitude or the sound range within a preset range is drawn, and for a hollow shaft, an X axis is expressed as an axial direction, and a Y axis is expressed as a circumferential angle.
The A scan display, the B scan display and the C scan display are stored in a binary file in a host computer of the flaw detection device, for example, the A scan display is stored as ". A. The B scan display is stored as". B. The C scan display is stored as ". C).
And step 200, acquiring sample data through the binary file, and constructing a defect positive sample library and a defect negative sample library.
And S210, clustering and analyzing the binary files.
The clustering analysis adopts a clustering algorithm to obtain a clustering result corresponding to a preset dimension, wherein the preset dimension in the embodiment refers to a clustering mode according to one or more of an axial mode or a round-robin mode, a channel and a scanning mode, and the clustering algorithm comprises at least one of a k-means clustering algorithm, a mean shift clustering algorithm, a pixel clustering algorithm, a hierarchical clustering algorithm, a spectral clustering algorithm and a depth embedded clustering algorithm based on a depth convolutional neural network. The binary files may be shared and transferred over a computer network. And step S220, extracting an image from the binary file after cluster analysis.
And S230, capturing a defect position image according to the image characteristics of the defect to obtain sample data.
Each defect has its unique image characteristics, such as a C-scan display, the characteristics of the defect progressively decrease from center point to edge amplitude, and the corresponding image characteristics progressively decrease from center point to edge color table. B scanning shows that for the flaw detection of the axle, as the probe advances spirally along a certain direction, the same flaw can be continuously found for a plurality of times by the same probe, and the corresponding image features are point sets which are continuously and discontinuously arranged along a certain direction; for wheel flaw detection, the probe is used for fixing the wheel to rotate, the defect can be continuously found for a plurality of times by the same probe, but the corresponding image features are continuously intermittent point sets along a certain direction, each point set presents a parabola (dovetail shape), and the color table gradually decreases from the top point to the two sides.
Sample data is obtained from the defect image using a feature extraction algorithm, which may include, but is not limited to, at least one of: RPN (Region Proposal Network, candidate region generation network) algorithm, FPN (Feature Pyramid Networks, feature pyramid network) algorithm, and DN (Deep Net) algorithm, among others.
Step S240, defective sample data form a defective positive sample library, and non-defective sample data form a defective negative sample library.
Determining a defect position according to the image characteristics of the defect, and cutting out images of the defect position to form a defect positive sample library and a defect negative sample library, wherein the defect positive sample library comprises defect images such as a flat bottom hole image, a transverse defect image, a longitudinal defect image, an outer surface defect image, an inner surface defect image and the like; the defect negative sample library comprises non-defect images, noise images, transition arc images and the like.
And step S300, preprocessing the defect positive sample library to generate a defect positive sample template file.
Pre-processing the defect positive sample, wherein the pre-processing method comprises at least one of the following steps: adjusting at least one of brightness, gray scale and contrast; at least one of a resizing and an offset; and performing self-adaptive graph cutting and the like, and performing normalization processing on the size and the format of the defect image, wherein the processing size of the defect image is 20 pixels by 20 pixels, the format is stored as BMP, and a normalized defect positive sample library generates a defect positive sample template file.
And step 400, constructing a defect identification model through the defect positive sample template file, the sample matching algorithm and the defect negative sample library.
And inputting parameters such as a defect positive sample template file, a sample matching algorithm, a defect negative sample library, a training stage and the like, performing machine learning, and building a defect identification model after learning is finished, wherein the defect identification model is stored in the form of an XML file. The sample matching algorithm may be a gray value based matching algorithm, a shape based matching algorithm, a feature point based matching algorithm, or the like.
Example 2
Referring to fig. 2: a wheel axle defect identification method based on machine vision comprises the following steps:
and step Z100, acquiring detection data of the wheel axle by using ultrasonic flaw detection equipment, storing the detection data in a binary file, and extracting a detection image from the binary file.
This step is the same as that of embodiment 1, and will not be described here again.
And step Z200, performing defect recognition on the detected image by using a defect recognition model.
After the processing program loads the defect recognition model, parameters such as a detection image, a defect lower limit (a target image pixel size lower limit), a defect upper limit (a target image pixel size upper limit), steps (an image division scale according to a certain size), a neighborhood (hit times) and the like are input, a computer automatically matches and recognizes the defect and outputs a defect recognition result, the recognition result comprises an image, a position, a size, a matching degree (the matching degree of the defect image and the defect recognition model) and a type, and a schematic diagram of a display result is shown in fig. 3.
Further processing in the program may obtain an amplitude, a sound path, e.g. the sound path, the amplitude may be obtained from the detection data based on the position information.
If the input detection image, CRH1A, is a 45+ forward C display image, the defect lower limit is 5, the defect upper limit is 40, the step is 1.1, the neighborhood is 3 and other parameters; the computer can output the result of the defect size of 5*5 to 40 x 40 pixels in the image with the hit times of more than or equal to 3 times according to the defect recognition model, the output result has a coincident block diagram or a non-defect block diagram, the output result is stored in a structured data after frame regression and cluster analysis, and the detection result is shown as follows:
[{id:"1",t＝"1",channel:"45+",x:"244",y:"267",h:"28",w:"28",l:"9",v:"45",k:"1.3",d＝"55"},{id:"2",t＝"1",channel:"45+",x:"382",y:"88",h:"27",w:"27",l:"9",v:"65",k:"1.3",d＝"60"},{id:"3",t＝"1",channel:"45+",x:"500",y:"177",h:"29",w:"29",l:"9",v:"70",k:"1.3",d＝"55"},{id:"4",t＝"1",channel:"45+",x:"630",y:"267",h:"28",w:"28",l:"9",v:"80",k:"1.3",d＝"55"},{id:"5",t＝"1",channel:"45+",x:"1057",y:"87",h:"28",w:"28",l:"9",v:"66",k:"1.3",d＝"60"},{id:"6",t＝"1",channel:"45+",x:"1330",y:"88",h:"26",w:"26",l:"9",v:"90",k:"1.3",d＝"55"},{id:"7",t＝"2",channel:"45+",x:"1589",y:"267",h:"29",w:"29",l:"9",v:"77",k:"1.3",d＝"55"},{id:"8",t＝"1",channel:"45+",x:"1876",y:"179",h:"27",w:"27",l:"9",v:"80",k:"1.3",d＝"70"},{id:"9",t＝"1",channel:"45+",x:"1986",y:"87",h:"29",w:"29",l:"9",v:"77",k:"1.3",d＝"55"},{id:"10",t＝"1",channel:"45+",x:"2086",y:"2",h:"27",w:"27",l:"9",v:"89",k:"1.3",d＝"60"},{id:"11",t＝"1",channel:"45+",x:"2125",y:"266",h:"29",w:"29",l:"9",v:"90",k:"1.3",d＝"55"}]。
the above results indicate that 11 defects were found in total, where id is the serial number, t is the defect type, x is the defect x-axis coordinate, y is the y-axis coordinate, channel is the channel name, h is the defect height, w is the defect width, l is the stage value, v is the amplitude, k is the degree of matching, and d is the sound path.
The output results are tabulated as follows:
| id | t | Channel | x | y | h | w | l | v | k | d |
| 1 | 1 | 45+ | 244 | 267 | 28 | 28 | 9 | 45 | 1.3 | 55 |
| 2 | 1 | 45+ | 382 | 88 | 27 | 27 | 9 | 65 | 1.3 | 60 |
| 3 | 1 | 45+ | 500 | 177 | 29 | 29 | 9 | 70 | 1.3 | 55 |
| 4 | 1 | 45+ | 630 | 267 | 28 | 28 | 9 | 80 | 1.3 | 55 |
| 5 | 1 | 45+ | 1057 | 87 | 28 | 28 | 9 | 66 | 1.3 | 60 |
| 6 | 1 | 45+ | 1330 | 88 | 26 | 26 | 9 | 90 | 1.3 | 55 |
| 7 | 2 | 45+ | 1589 | 267 | 29 | 29 | 9 | 77 | 1.3 | 55 |
| 8 | 1 | 45+ | 1876 | 179 | 27 | 27 | 9 | 80 | 1.3 | 70 |
| 9 | 1 | 45+ | 1986 | 87 | 29 | 29 | 9 | 77 | 1.3 | 55 |
| 10 | 1 | 45+ | 2086 | 2 | 27 | 27 | 9 | 89 | 1.3 | 60 |
| 11 | 1 | 45+ | 2125 | 266 | 29 | 29 | 9 | 90 | 1.3 | 55 |
and step Z300, performing defect screening on the defect identification result according to the specific parameters.
After the computer loads the defect identification result, defect screening is carried out according to specific parameters, wherein the screened parameters comprise defect types, threshold values (amplitudes) and positions.
And step Z310, screening the defects through defect types.
Assuming that 1 represents a transverse defect, 2 represents a longitudinal defect, 3 represents a flat bottom hole, 4 represents a surface defect and so on, if only the transverse defect is selected, matching the result with t being 1 in the result file and displaying, and the other types are not displayed;
and step Z320, screening the defects through an amplitude threshold.
If the amplitude threshold is adjusted to 50, displaying the defects with the amplitude of the defects plus a floating value (the default value of the floating value is 0) being greater than 50, and displaying the defects with the amplitude being less than or equal to 50.
And step Z330, screening defects through positions.
Screening according to the radial, axial and circumferential position information, wherein if the defect position is characterized by short wave amplitude of the sound path, comparing the sound path of a certain position with the contour line (axial contour and wheel contour) of the position according to the characteristic, if the floating value (default value of the floating value is 0) of the sound path is smaller than the contour line, displaying the sound path, otherwise, not displaying the sound path; if the hollow axle B scanning display or C scanning display is adopted, the axial display range is set to be 100-500 mm, and the defect display in the closed region is not displayed otherwise; if the hollow axle C is scanned and displayed, the circumferential display range is set to be 0-180 degrees, and the defect display in the closed section is displayed, otherwise, the defect display is not displayed;
several parameters may be used alone or in combination, if the amplitude threshold is adjusted to 50 and the type 1 is a lateral defect, then the defect will be shown to be lateral, the amplitude of the defect will float above 50, and the unconditional defect will not be shown.
As shown in fig. 4, the defect recognition result of the hollow axle before screening is shown in fig. 5, and the defect recognition result after defect screening is more accurate.
And step Z400, displaying the screened defects and alarming.
And loading the screened defect data into an image to display the defect data in the form of a defect display frame, wherein the information of the defect display frame comprises a serial number, an amplitude, a type and the like.
The defect alarm and various alarm modes are not limited to early warning and displaying the information related to the defect in the forms of images, sounds, photoelectricity, lists and the like.
And step Z500, constructing a defect recognition model by using the image supplement contained in the defect recognition result.
Preferably, the label of the defect display frame can display information such as serial number, amplitude, type and the like, and a corresponding response event can be caused by clicking, double clicking and dragging the defect display frame by a mouse.
The mouse response event includes but is not limited to events such as movement, clicking, double clicking and the like, for example, a mouse presses a left key to drag a certain defect display frame, the defect frame position can be moved and defect position information can be corrected, a right key clicks a certain defect frame, a popup menu is displayed, and the popup menu comprises 'delete', 'add to positive sample', 'add to negative sample';
according to a certain defect display frame mouse response event, the defect image of the defect display frame can be manually selected to be added to the defect positive sample library or the defect negative sample library.
More preferably, the computer automatically adds the defect image to the defect positive sample library or the defect negative sample library according to a comparison result of the defect image matching degree and a matching degree threshold, wherein the matching degree threshold comprises a positive sample matching degree threshold and a negative sample matching degree threshold, and the matching degree threshold can be obtained according to the defect identification model. The matching degree can be obtained through a matching algorithm, and the matching algorithm can be one of a matching algorithm based on gray values, a matching algorithm based on shapes and a matching algorithm based on feature points.
The defect image matching degree is greater than or equal to the positive sample matching degree threshold value and is added into the defect positive sample library, and the defect image matching degree is less than or equal to the negative sample matching degree threshold value and is added into the defect negative sample library, and the intermediate value is not processed. If the positive sample matching degree threshold is 1 and the negative sample matching degree threshold is-1, adding a defect image with the matching degree greater than or equal to 1 to a defect positive sample library, adding a defect image with the matching degree less than or equal to-1 to a defect negative sample library, and preprocessing the defect positive sample, wherein the preprocessing method comprises at least one of the following steps: adjusting at least one of brightness, gray scale and contrast; at least one of a resizing and an offset; and performing self-adaptive graph cutting and normalization processing, namely unifying and performing normalization processing on the size and the format of the defect image, wherein the processing size of the defect image is 20 x 20 pixels, the format is stored as BMP, and a defect positive sample library after normalization processing generates a defect positive sample template file.
The invention adopts the machine vision technology in the artificial intelligence field to acquire and construct the image characteristic information of the defects, and a computer utilizes the image characteristic information to analyze the detection data/images of ultrasonic detection, identify the defects in the detected workpiece, and can prompt the detection personnel of the defect information (including but not limited to positions, sizes, amplitudes, types and the like) in the forms of images, tables, sounds, lights (including but not limited to warning lights) and the like.
Meanwhile, the image characteristic information of the defects is continuously supplemented and perfected along with the increase of detection samples and the enrichment of defect types, so that the accuracy of data analysis is further improved.
Compared with a non-artificial intelligent method, the method has autonomous learning capability, and can adapt to more types of defects by newly adding the types of the positive samples; and the non-defect interference information can be accurately distinguished by putting the non-defect interference sample into the defect negative sample library, so that the defect identification accuracy is further improved.
The above examples are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above examples, and any other changes, modifications, substitutions, combinations, and simplifications that do not depart from the spirit and principle of the present invention should be made in the equivalent manner, and the embodiments are included in the protection scope of the present invention.
Claims (9)
1. A wheel axle defect identification model construction method based on machine vision is characterized by comprising the following steps of: the method comprises the following steps:
step S100, respectively acquiring at least one detection data corresponding to the wheel and/or the shaft by utilizing ultrasonic flaw detection equipment, and storing the detection data in a binary file; wherein, the detection data obtained by the same probe of the ultrasonic flaw detection equipment are assigned to the same type of channel;
step 200, performing cluster analysis on the binary files based on the wheel type, the shaft type and the channel type, wherein the binary files belonging to the same type correspond to the same defect characteristics; obtaining sample data by using defect characteristics corresponding to the binary file, and constructing a defect positive sample library and a defect negative sample library;
step S300, preprocessing a defect positive sample library to generate a defect positive sample template file;
and step 400, constructing a defect identification model through the defect positive sample template file, the sample matching algorithm and the defect negative sample library.
2. The machine vision-based axle defect recognition model construction method of claim 1, wherein the method comprises the steps of: the ultrasonic flaw detection equipment is configured with corresponding channels and scanning modes to obtain detection data according to the shaft type or the wheel type.
3. The machine vision-based axle defect recognition model construction method of claim 1, wherein the method comprises the steps of: the step S200 further includes the steps of:
step S210, clustering analysis is carried out on the binary files;
step S220, extracting images from the binary files after cluster analysis;
step S230, capturing a defect position image according to the image characteristics of the defect to obtain sample data;
step S240, defective sample data form a defective positive sample library, and non-defective sample data form a defective negative sample library.
4. The machine vision-based axle defect recognition model construction method of claim 1, wherein the method comprises the steps of: the defect positive sample library comprises defect images; the defect negative sample library comprises a non-defect image, a noise image and a transitional arc image.
5. The machine vision-based axle defect recognition model construction method of claim 4, wherein the method comprises the steps of: the defect image comprises a flat bottom hole image, a transverse defect image, a longitudinal defect image, an outer surface defect image and an inner surface defect image.
6. The machine vision-based axle defect recognition model construction method of claim 1, wherein the method comprises the steps of: the preprocessing includes normalization processing, i.e., unifying the size and format of the defect image.
7. A wheel axle defect identification method based on machine vision is characterized in that: the method comprises the following steps:
step Z100, acquiring detection data of the wheel axle by utilizing ultrasonic flaw detection equipment, storing the detection data in a binary file, and extracting a detection image from the binary file;
step Z200, performing defect recognition on the detection image by using the wheel axle defect recognition model according to any one of claims 1-6;
step Z300, performing defect screening on the defect identification result according to the position of the defect;
and step Z400, displaying the screened defects and alarming.
8. The machine vision-based axle defect identification method of claim 7, wherein: the wheel axle defect identification method further comprises the following steps: and step Z500, constructing the wheel axle defect recognition model according to any one of claims 1-6 by using the image supplement contained in the defect recognition result.
9. The machine vision-based axle defect identification method of claim 8, wherein: step Z500 further comprises: manually selecting to add the image contained in the defect identification result to the defect positive sample library or the defect negative sample library; the computer can also automatically increase the image contained in the defect identification result to a defect positive sample library or a defect negative sample library according to the comparison result of the defect image matching degree and the matching degree threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010128206.8A CN111257422B (en) | 2020-02-28 | 2020-02-28 | Wheel axle defect identification model construction method and defect identification method based on machine vision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010128206.8A CN111257422B (en) | 2020-02-28 | 2020-02-28 | Wheel axle defect identification model construction method and defect identification method based on machine vision |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111257422A CN111257422A (en) | 2020-06-09 |
| CN111257422B true CN111257422B (en) | 2023-09-08 |
Family
ID=70946130
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010128206.8A Active CN111257422B (en) | 2020-02-28 | 2020-02-28 | Wheel axle defect identification model construction method and defect identification method based on machine vision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111257422B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113470018B (en) * | 2021-09-01 | 2021-12-21 | 深圳市信润富联数字科技有限公司 | Hub defect identification method, electronic device, device and readable storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2012008142A (en) * | 2011-08-30 | 2012-01-12 | Mitsubishi Heavy Ind Ltd | Program, processing unit and processing method for processing ultrasonic flaw detection data |
| CN107451997A (en) * | 2017-07-31 | 2017-12-08 | 南昌航空大学 | A kind of automatic identifying method of the welding line ultrasonic TOFD D scanning defect types based on deep learning |
| CN107749058A (en) * | 2017-10-23 | 2018-03-02 | 中南大学 | A kind of machine vision detection method and system of boiler tubing surface defect |
| CN109658397A (en) * | 2018-12-12 | 2019-04-19 | 广州地铁集团有限公司 | A kind of rail polling method and system |
| CN109829907A (en) * | 2019-01-31 | 2019-05-31 | 浙江工业大学 | A kind of metal shaft surface defect recognition method based on deep learning |
| CN110297041A (en) * | 2019-07-01 | 2019-10-01 | 天津工业大学 | A kind of 3D woven composite defect inspection method based on FCN and GRU |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11078631B2 (en) * | 2017-11-30 | 2021-08-03 | Sperry Rail Holdings, Inc. | System and method for inspecting a rail using machine learning |
-
2020
- 2020-02-28 CN CN202010128206.8A patent/CN111257422B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2012008142A (en) * | 2011-08-30 | 2012-01-12 | Mitsubishi Heavy Ind Ltd | Program, processing unit and processing method for processing ultrasonic flaw detection data |
| CN107451997A (en) * | 2017-07-31 | 2017-12-08 | 南昌航空大学 | A kind of automatic identifying method of the welding line ultrasonic TOFD D scanning defect types based on deep learning |
| CN107749058A (en) * | 2017-10-23 | 2018-03-02 | 中南大学 | A kind of machine vision detection method and system of boiler tubing surface defect |
| CN109658397A (en) * | 2018-12-12 | 2019-04-19 | 广州地铁集团有限公司 | A kind of rail polling method and system |
| CN109829907A (en) * | 2019-01-31 | 2019-05-31 | 浙江工业大学 | A kind of metal shaft surface defect recognition method based on deep learning |
| CN110297041A (en) * | 2019-07-01 | 2019-10-01 | 天津工业大学 | A kind of 3D woven composite defect inspection method based on FCN and GRU |
Non-Patent Citations (1)
| Title |
|---|
| 基于深度学习的不规则特征识别检测技术;赵欣洋等;《轻工机械》;20190613(第03期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111257422A (en) | 2020-06-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113362326B (en) | Method and device for detecting defects of welding spots of battery | |
| CN114757949B (en) | Wire and cable defect detection method and system based on computer vision | |
| CN111862064B (en) | Silver wire surface flaw identification method based on deep learning | |
| CN110779937B (en) | Casting product internal defect intelligent detection device | |
| CN112083017A (en) | Intelligent detection device, method and medium for internal defects of welding seam | |
| CN110253019B (en) | Quality monitoring and control method for selective laser melting | |
| CN111667462B (en) | Automatic detection method and system for small and weak targets in welding defects | |
| CN111815555A (en) | Metal additive manufacturing image detection method and device combining anti-neural network with local binary | |
| CN111798419A (en) | Metal paint spraying surface defect detection method | |
| CN112529884B (en) | Welding spot quality evaluation method based on indentation characteristic image recognition | |
| CN112288760B (en) | Adherent cell image screening method and system and cell image analysis method | |
| CN114910480A (en) | Wafer surface defect detection method based on machine vision | |
| CN111161237A (en) | Fruit and vegetable surface quality detection method, storage medium and sorting device thereof | |
| CN113012123A (en) | Classification recognition and quantitative analysis method and system for defect and damage of carbon fiber composite material | |
| CN115797361B (en) | Aluminum template surface defect detection method | |
| US20220076404A1 (en) | Defect management apparatus, method and non-transitory computer readable medium | |
| CN111257422B (en) | Wheel axle defect identification model construction method and defect identification method based on machine vision | |
| CN116432988B (en) | Intelligent management method, medium and equipment for valve production process data | |
| Tang et al. | Surface inspection system of steel strip based on machine vision | |
| CN116511998B (en) | Defect detection method and system for circular cutterhead | |
| CN117934467A (en) | Titanium metal welding quality detection method and system based on image processing | |
| CN117274251A (en) | Tablet quality detection method in medicine production process based on image data | |
| CN111738991A (en) | Method for creating digital ray detection model of weld defects | |
| CN107024416A (en) | With reference to similitude and the quasi-circular particle mean size detection method of discontinuity | |
| CN116381053A (en) | Ultrasonic detection method and system for welding metal materials |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |