CN115170569A - Failure detection method of high-entropy material coating cutter based on image - Google Patents

Failure detection method of high-entropy material coating cutter based on image Download PDF

Info

Publication number
CN115170569A
CN115170569A CN202211087012.3A CN202211087012A CN115170569A CN 115170569 A CN115170569 A CN 115170569A CN 202211087012 A CN202211087012 A CN 202211087012A CN 115170569 A CN115170569 A CN 115170569A
Authority
CN
China
Prior art keywords
image
cutter
entropy material
material coating
pixel
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.)
Granted
Application number
CN202211087012.3A
Other languages
Chinese (zh)
Other versions
CN115170569B (en
Inventor
姬清华
曹鋆汇
杨婷婷
张野
邢奇
侯树森
马飞豹
梁婧怡
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.)
Xinxiang University
Original Assignee
Xinxiang University
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 Xinxiang University filed Critical Xinxiang University
Priority to CN202211087012.3A priority Critical patent/CN115170569B/en
Publication of CN115170569A publication Critical patent/CN115170569A/en
Application granted granted Critical
Publication of CN115170569B publication Critical patent/CN115170569B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • 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
    • 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
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20076—Probabilistic image processing
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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
    • G06T2207/30136—Metal

Landscapes

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

Abstract

A failure detection method of a high-entropy material coating cutter based on images is characterized in that an image analysis method is adopted to automatically detect the high-entropy material coating cutter in a non-contact and non-damage mode, images of the high-entropy material coating cutter are analyzed according to apparent characteristics of the high-entropy material coating cutter, whether the running state of the high-entropy material coating cutter is normal or failed is judged, the good state of the cutter in use is ensured, and the reliability and the safety of industrial production are improved.

Description

Failure detection method of high-entropy material coating cutter based on image
Technical Field
The invention relates to the field of cutter coating detection, in particular to a failure detection method of a high-entropy material coating cutter based on an image.
Background
Hard alloys composed of refractory metal compounds and metal binders have become one of the main high-speed cutting tool materials due to their excellent mechanical properties and cutting performance. However, in the cutting process, with the increase of the cutting speed, the heat quantity of transformation of deformation work and friction work is increased rapidly, the cutting temperature is increased remarkably, and the service life of the cutter is seriously influenced. This problem has become one of the problems that will affect the rapid development of equipment manufacturing industry and needs to be solved urgently. In order to increase the reliability of the operation of the cutter and prolong the service life of the cutter, three methods are usually adopted to improve the cutter in practice: firstly, hard alloy materials with better performance are researched and developed; secondly, optimizing a cutting process; thirdly, a surface strengthening technology is adopted. For decades, cutter materials are developed rapidly, the performance improvement space is limited, the cutting machining process is relatively mature, and the service life of the cutter can be effectively prolonged only by performing coating treatment. At present, more than 90% of cutting tools for high-precision machine tools in developed countries are subjected to coating treatment. Therefore, the development of the novel wear-resistant heat-resistant cutter coating is an effective way for prolonging the service life of the cutter, improving the production efficiency, saving resources and reducing environmental pollution.
The multi-element nano composite of the coating components is a main way for improving the toughness and the thermal stability of the hard wear-resistant coating material, and the multi-element nano composite coating is prepared and can obtain a novel hard wear-resistant coating material with more excellent performance. On the premise of excellent microstructure, the CoCrCuFeNi binding phase can fully play a high-entropy effect at high temperature, on one hand, the stable oxidation lubricating layer is promoted to be formed, on the other hand, the high-strength combination with a hard phase is realized, and the generation of abrasive particles is avoided, so that the high-temperature friction and wear performance of the cutter is remarkably improved. Therefore, the high-entropy alloy material (HEA) coating is an effective method for improving the performance of the cutter, and has application prospects. The failure detection of the high-speed cutting tool is an effective and necessary way for reducing waste generation and avoiding the potential safety hazard of tool operation.
The traditional detection of the coating requires a destructive operation, which makes the method only applicable to spot inspection, but not to real-time detection of the production line. The image-based industrial detection method is an emerging intelligent detection tool in recent years, and is widely applied to the field of industrial detection due to the characteristics of no contact with a detection target, no damage to a target, low cost and high efficiency. The current method of using neural network image processing algorithms includes, however, the input of the current neural network algorithm is either the whole tool image or is specific to a particular region (e.g. a knife edge). The former has extremely low detection efficiency, and the latter depends on the positioning accuracy of the knife edge area. Moreover, most of the existing neural network algorithms only aim at detecting defects of the cutter, and are not specially designed aiming at the degradation of the coating, so that the detection accuracy of the coating failure is not high. Although the neural network model is used for distinguishing images in other fields, the network structure difference is huge due to different detection objects, and the neural network model cannot be directly applied to accurate detection of the high-entropy material coating of the cutter.
Disclosure of Invention
Based on the industrial application requirements and the characteristics of the high-entropy material coating cutter, the invention innovatively provides a failure detection method of the high-entropy material coating cutter based on images.
A failure detection method of a high-entropy material coating cutter based on images,
step 1: extraction of key region of high-entropy material coating cutter image
Step 1.1, establishing a probability model of the color distribution of the high-entropy material coating cutter image: performing channel activation operation on the color according to the color value of the RGB channel of the image; the activated image calculation method comprises the following steps:
Figure DEST_PATH_IMAGE002
r, g, b in equation (1) represent the RGB three channels of the image;
Figure DEST_PATH_IMAGE004
、
Figure DEST_PATH_IMAGE006
is a linear coefficient, and the linear coefficient,
Figure DEST_PATH_IMAGE008
、
Figure DEST_PATH_IMAGE010
is composed of
Figure DEST_PATH_IMAGE012
As a function of (a) or (b),
Figure 620330DEST_PATH_IMAGE012
the color value range is (0, 1)](ii) a Wherein:
Figure DEST_PATH_IMAGE014
based on activation images
Figure DEST_PATH_IMAGE016
The input image is judged according to the distribution probability of the pixel values, and the probability that the input image is the high-entropy material coating cutter image is judged:
Figure DEST_PATH_IMAGE018
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE020
for activating images
Figure 851591DEST_PATH_IMAGE016
The value of the pixel of (a) is,
Figure DEST_PATH_IMAGE022
coating knives for high entropy materialsThe device is provided with an image classification label,
Figure DEST_PATH_IMAGE024
representing a probability function, argmax representing a make probability function
Figure 970857DEST_PATH_IMAGE024
Taking maximum time parameter
Figure 832634DEST_PATH_IMAGE022
A value of (d);
step 1.2, extracting a key region of the high-entropy material coating cutter image from the image: to pair
Figure 466878DEST_PATH_IMAGE016
Each pixel of (2)
Figure DEST_PATH_IMAGE026
According to a probabilistic model
Figure DEST_PATH_IMAGE028
Solving the classification label to obtain a key area
Figure DEST_PATH_IMAGE030
；
Step 2: state failure detection of high-entropy material coating cutter based on key area
Taking each pixel
Figure 265944DEST_PATH_IMAGE020
Of the neighborhood of pixels
Figure DEST_PATH_IMAGE032
：
Figure DEST_PATH_IMAGE034
Wherein each pixel in the neighborhood
Figure DEST_PATH_IMAGE036
Is that
Figure 911820DEST_PATH_IMAGE016
Neutralization of
Figure 6815DEST_PATH_IMAGE020
Is not more than
Figure DEST_PATH_IMAGE038
The set of pixels of (a) is,
Figure 925093DEST_PATH_IMAGE038
is the radius of the neighborhood;
activating an image from one
Figure 966998DEST_PATH_IMAGE016
The number of pixels in which the key region is obtained is
Figure DEST_PATH_IMAGE040
Is recorded as
Figure DEST_PATH_IMAGE042
(ii) a Correspondingly have
Figure 293812DEST_PATH_IMAGE040
A neighborhood of marked as
Figure DEST_PATH_IMAGE044
,
Figure DEST_PATH_IMAGE046
,…,
Figure DEST_PATH_IMAGE048
Will turn over
Figure 28550DEST_PATH_IMAGE040
The neighborhoods are arranged from small to large according to the pixel mean value to obtain a sequence:
Figure DEST_PATH_IMAGE050
Figure DEST_PATH_IMAGE052
representing a sequence
Figure DEST_PATH_IMAGE054
One element of
Figure DEST_PATH_IMAGE056
(is a neighborhood) pixel mean;
in a certain neighborhood as described above
Figure 840648DEST_PATH_IMAGE056
Establishing a calculation model for input, and calculating the high-entropy material coating cutter state voting coefficient according to input data; setting the corresponding model output as
Figure DEST_PATH_IMAGE058
，
Figure DEST_PATH_IMAGE060
For coefficients ordered according to equation (6), calculate:
Figure DEST_PATH_IMAGE062
the calculation result of the equation (13) is a tool state voting coefficient; and when the voting coefficient is larger than 0.3, judging that the cutter is invalid.
The production line is provided with image acquisition equipment, the image acquisition equipment is connected with a communication unit, the communication unit is connected with a remote server through a communication network, and the server is connected with a controller of the production line.
The image acquisition equipment is used for acquiring a tool image.
The communication unit is used for transmitting the cutter image to the server.
The server is used for implementing the method, judging whether the cutter is invalid or not and feeding back the result to the controller.
The controller is used for receiving the discrimination signal fed back by the server, thereby controlling the action of the production line to eliminate unqualified cutters.
The key area refers to an image pixel set used for detecting the performance state of the high-entropy material coating cutter.
Solving conditional probability functions
Figure DEST_PATH_IMAGE064
The method comprises the following steps:
Figure DEST_PATH_IMAGE066
wherein
Figure DEST_PATH_IMAGE068
Is a pixel
Figure 794566DEST_PATH_IMAGE020
Is determined by the conditional probability function of (a),
Figure DEST_PATH_IMAGE070
is to classify
Figure 336406DEST_PATH_IMAGE022
Is determined by the prior probability function of (a),
Figure DEST_PATH_IMAGE072
is a pixel
Figure 507624DEST_PATH_IMAGE020
Is calculated as a function of the prior probability.
The probability is a model parameter and is obtained through a training sample of the high-entropy material coating cutter image.
A cutter on-line use method comprises the detection method.
The invention has the advantages that:
1. the extraction method of the key area of the high-entropy material coating cutter image is different from a mode of dividing the area in advance in the prior art, a pixel set conforming to an algorithm is used as the key area, and the probability that the pixel belongs to the key area can be accurately calculated by activating the processing of the image, so that the extraction method of the key area can be quickly and accurately finished, and is quicker and more accurate than the prior art. And furthermore, a foundation is further provided for the operation of the following neural network, and the high efficiency and accuracy of the operation of the neural network are ensured.
2. Calculating the key area pixels of the high-entropy material coating cutter image according to the step 1, performing neighborhood expansion, taking the neighborhood as an input parameter, calculating the state voting coefficient of the high-entropy material coating cutter according to the key area by using an optimized special neural network model (special excitation function and the like), and setting a special algorithm for the voting coefficient to judge whether the high-entropy material coating cutter fails, so that the non-contact and non-damage automatic detection of the high-entropy material coating cutter is realized, and the detection result is accurate and quick.
Detailed Description
Step 1: extracting a key area of the high-entropy material coating cutter image:and detecting and extracting a key region of the high-entropy material coating cutter image from the shot image.
Unlike the prior art, the critical area is not a fixed geometric area that has been manually specified in advance. In the invention, the key area of the high-entropy material coating cutter image refers to an image pixel set for detecting the performance state of the high-entropy material coating cutter, and is a subset of an original shot image. And such a set of pixels is not necessarily a standard geometric area.
Due to the special properties of the high-entropy material coating, the high-entropy material coating cutter shows special color distribution in an image. A probability model of the color distribution of the high-entropy material coating cutter image is established, model parameters are obtained through training (step 1.1), and the probability model is used for extracting a key area of the high-entropy material coating cutter image from the image (step 1.2).
Step 1.1, a probability model of the color distribution of the high-entropy material coating cutter image is established, and parameters required by the model are obtained through training.
In order to reflect the characteristics of the high-entropy material coating cutter, the collected high-entropy material coating cutter image adopts a three-color RGB color image, wherein RGB is three color channels of the image.
And performing channel activation operation on the color according to the color value of the RGB channel of the image to obtain an activated image, so that the color of the high-entropy material coating cutter in the image is more obvious, and the model detection is facilitated.
The activated image calculation method comprises the following steps:
Figure DEST_PATH_IMAGE002A
in equation (1), r, g, b represent the RGB three channels of the image.
Figure 541439DEST_PATH_IMAGE004
、
Figure 987464DEST_PATH_IMAGE006
Is a linear coefficient of the linear coefficient,
Figure 23291DEST_PATH_IMAGE008
、
Figure 896569DEST_PATH_IMAGE010
is composed of
Figure 276735DEST_PATH_IMAGE012
Wherein:
Figure DEST_PATH_IMAGE014A
Figure 464134DEST_PATH_IMAGE012
the color value range is (0, 1)]I.e. the result after normalization. e and ln represent natural index and natural logarithm, respectively.
Obtaining an activation image by the calculation of equation (1)
Figure 183828DEST_PATH_IMAGE016
The method can make colors with high correlation degree with the high-entropy material coating in the original image more prominent, and inhibit other colors, thereby being beneficial to model detection.
Based on activation images
Figure 165691DEST_PATH_IMAGE016
The input image is judged according to the distribution probability of the pixel values, and the probability that the input image is the high-entropy material coating cutter image is judged:
Figure DEST_PATH_IMAGE018A
wherein, the first and the second end of the pipe are connected with each other,
Figure 174098DEST_PATH_IMAGE020
for activating images
Figure 961925DEST_PATH_IMAGE016
The value of the pixel of (a) is,
Figure 598443DEST_PATH_IMAGE022
the cutter image classification label is coated on the high-entropy material,
Figure DEST_PATH_IMAGE074
the time indicates that the pixel belongs to the high-entropy material coating cutter image,
Figure DEST_PATH_IMAGE076
the time indicates that the pixel cannot belong to the high-entropy material coating cutter image;
Figure 718583DEST_PATH_IMAGE024
representing a probability function, argmax representing a make probability function
Figure 807762DEST_PATH_IMAGE024
Taking maximum time parameter
Figure 399281DEST_PATH_IMAGE022
The value of (c). The function f will activate the image
Figure 765671DEST_PATH_IMAGE016
Is marked as 1 or 0, respectively, which indicates belonging to or not belonging to the high-entropy material coating tool image.
Classifying according to Bayes' theorem
Figure 948391DEST_PATH_IMAGE022
Conditional probability function of (2)
Figure 728128DEST_PATH_IMAGE064
Can be obtained by sample training:
Figure DEST_PATH_IMAGE066A
wherein
Figure 326599DEST_PATH_IMAGE068
Is a pixel
Figure 609813DEST_PATH_IMAGE020
Is determined by the conditional probability function of (a),
Figure 104380DEST_PATH_IMAGE070
is to classify
Figure 168150DEST_PATH_IMAGE022
Is determined by the prior probability function of (a),
Figure 101471DEST_PATH_IMAGE072
is a pixel
Figure 675410DEST_PATH_IMAGE020
The probability is a model parameter, and is obtained through a training sample of the high-entropy material coating cutter image. Therefore, the conditional probability function can be obtained by equation (4)
Figure 668774DEST_PATH_IMAGE064
。
Step 1.2, extracting key areas of high-entropy material coating cutter images from images
Establishing a probabilistic model according to step 1.1
Figure 219841DEST_PATH_IMAGE028
And solving model parameters through training
Figure 691273DEST_PATH_IMAGE068
、
Figure 886763DEST_PATH_IMAGE070
、
Figure 847765DEST_PATH_IMAGE072
。
Calculating the input image according to equations (1) and (2) to obtain an activation image
Figure 558232DEST_PATH_IMAGE016
To, for
Figure 505460DEST_PATH_IMAGE016
Each pixel of (2)
Figure 945668DEST_PATH_IMAGE020
According to a probabilistic model
Figure 15256DEST_PATH_IMAGE028
Solving the classification label to obtain a key area
Figure 416281DEST_PATH_IMAGE030
：
Figure DEST_PATH_IMAGE078
Namely the critical zone
Figure 760675DEST_PATH_IMAGE030
By activating the image
Figure 163712DEST_PATH_IMAGE016
The middle classification label is composed of pixels with 1,
Figure 466517DEST_PATH_IMAGE030
is that
Figure 417156DEST_PATH_IMAGE016
A subset of (a).
Step 2: state failure detection of high-entropy material coating cutter based on key area: calculating a key area of the high-entropy material coating cutter image according to the step 1, calculating a high-entropy material coating cutter state voting coefficient according to the key area, and judging whether the high-entropy material coating cutter is invalid or not according to the voting coefficient.
Obtaining the Key region according to step 1
Figure 971765DEST_PATH_IMAGE030
。
For each pixel
Figure DEST_PATH_IMAGE080
Take the set of its neighborhood pixels
Figure 855407DEST_PATH_IMAGE032
：
Figure DEST_PATH_IMAGE082
Each pixel in the neighborhood
Figure 938901DEST_PATH_IMAGE036
Is that
Figure 173573DEST_PATH_IMAGE016
Neutralization of
Figure 328611DEST_PATH_IMAGE020
Is not more than
Figure 473285DEST_PATH_IMAGE038
Set of pixels, vertical line
Figure DEST_PATH_IMAGE084
Representing the manhattan distance of two pixels,
Figure 852313DEST_PATH_IMAGE038
referred to as the radius of the neighborhood. Manhattan distance is robust to noise compared to Euclidean distanceThe performance is higher.
By expanding each pixel in the key region into its neighborhood for describing the local information of the key region, the neighborhood features have high correlation with the state of the tool compared with a single pixel, and thus can be used for judging the state of the tool.
Setting from one activation image
Figure 682604DEST_PATH_IMAGE016
The number of pixels in which the key region is obtained is
Figure 641333DEST_PATH_IMAGE040
Is marked as
Figure 765146DEST_PATH_IMAGE042
(ii) a Correspondingly have
Figure 783918DEST_PATH_IMAGE040
A neighborhood of marked as
Figure 868549DEST_PATH_IMAGE044
,
Figure 365389DEST_PATH_IMAGE046
,…,
Figure 343709DEST_PATH_IMAGE048
Will turn over
Figure 533382DEST_PATH_IMAGE040
Each neighborhood gets a sequence according to the small-to-large arrangement of its pixel mean:
Figure DEST_PATH_IMAGE050A
Figure 636468DEST_PATH_IMAGE052
representing a sequence
Figure 936999DEST_PATH_IMAGE054
One element of
Figure 645192DEST_PATH_IMAGE056
(is a neighborhood) of the mean of the pixels.
In a certain neighborhood as described above
Figure 802504DEST_PATH_IMAGE056
And establishing a calculation model for inputting, and calculating the tool state voting coefficient of the high-entropy material coating according to the input data.
Input device
Figure 924043DEST_PATH_IMAGE056
Expressed as:
Figure DEST_PATH_IMAGE086
Figure DEST_PATH_IMAGE088
is a neighborhood
Figure 136588DEST_PATH_IMAGE056
The coordinates of the middle pixel.
Order:
Figure DEST_PATH_IMAGE090
in the formula (7), the reaction mixture is,
Figure DEST_PATH_IMAGE092
is prepared by reacting with
Figure 230446DEST_PATH_IMAGE056
Are of equal size and are composed of
Figure 761921DEST_PATH_IMAGE056
Derived neighborhood of wherein
Figure DEST_PATH_IMAGE094
Is that
Figure 574019DEST_PATH_IMAGE056
The coordinates in the neighborhood of the point of interest,
Figure 481933DEST_PATH_IMAGE088
is a neighborhood
Figure 289352DEST_PATH_IMAGE092
The coordinates of (a) are (b),
Figure DEST_PATH_IMAGE096
representing the two aforementioned neighborhoods
Figure 162367DEST_PATH_IMAGE092
And
Figure 992920DEST_PATH_IMAGE056
the correlation coefficient between two related pixels in the image,
Figure DEST_PATH_IMAGE098
is a linear offset.
Figure DEST_PATH_IMAGE100
The function is used to non-linearize the model, and
Figure DEST_PATH_IMAGE102
Figure 579890DEST_PATH_IMAGE100
the non-linearisation of the function allows the model to model the input data more accurately.
Further, let:
Figure DEST_PATH_IMAGE104
in the formula (9), the reaction mixture is,
Figure DEST_PATH_IMAGE106
is a radius of
Figure DEST_PATH_IMAGE108
And is made of
Figure 913920DEST_PATH_IMAGE092
Derived neighborhood of wherein
Figure 223416DEST_PATH_IMAGE094
Is that
Figure 541265DEST_PATH_IMAGE092
The coordinates in the neighborhood of the point of interest,
Figure 587718DEST_PATH_IMAGE088
is a neighborhood
Figure 979517DEST_PATH_IMAGE106
The coordinate(s) of (a) is (are),
Figure DEST_PATH_IMAGE110
representing the two aforementioned neighborhoods
Figure 820434DEST_PATH_IMAGE106
And
Figure 625579DEST_PATH_IMAGE092
the correlation coefficient between two related pixels in the image,
Figure DEST_PATH_IMAGE112
is a linear offset.
Figure 616668DEST_PATH_IMAGE100
The function is the same as equation (8).
Further, let:
Figure DEST_PATH_IMAGE114
in the formula (10), the compound represented by the formula (10),
Figure DEST_PATH_IMAGE116
is a radius of
Figure DEST_PATH_IMAGE118
And is composed of
Figure 299191DEST_PATH_IMAGE106
Derived neighborhood of wherein
Figure 576589DEST_PATH_IMAGE094
Is that
Figure 603451DEST_PATH_IMAGE106
The coordinates in the neighborhood of the point of interest,
Figure 867073DEST_PATH_IMAGE088
is a neighborhood
Figure 358097DEST_PATH_IMAGE116
The coordinates of (a) are (b),
Figure DEST_PATH_IMAGE120
representing the two aforementioned neighborhoods
Figure 416183DEST_PATH_IMAGE116
And with
Figure 992658DEST_PATH_IMAGE106
The correlation coefficient between two related pixels in the image,
Figure DEST_PATH_IMAGE122
is a linear offset.
Figure 794392DEST_PATH_IMAGE100
The function is the same as equation (8).
Three-layer pyramid neighborhoods are derived from the original neighborhoods according to the formulas (7), (9) and (10) and are used for modeling different scale structures of the neighborhoods, so that the local features of the image can be more accurately described. The cracks causing the cutter state failure are represented as local repeated features in the image, and whether the cracks appear in the local image can be detected by adopting the method.
Order:
Figure DEST_PATH_IMAGE124
in the formula (11), the reaction mixture is,
Figure 802841DEST_PATH_IMAGE094
is that
Figure 625303DEST_PATH_IMAGE116
The coordinates in the neighborhood of the point of interest,
Figure DEST_PATH_IMAGE126
to represent
Figure 423495DEST_PATH_IMAGE116
One pixel in the neighborhood and the output
Figure DEST_PATH_IMAGE128
The correlation coefficient of (a) is calculated,
Figure DEST_PATH_IMAGE130
output of
Figure 294499DEST_PATH_IMAGE128
Indicating whether there is a crack in the neighborhood of the input, when the output is
Figure DEST_PATH_IMAGE132
Is indicative of the presence of a crack, when output
Figure DEST_PATH_IMAGE134
Indicating the absence of cracks.
Equations (7) - (11) define a computational model for detecting the presence of cracks in the neighborhood of the critical region. Defining a cost function:
Figure DEST_PATH_IMAGE136
the cost function is used for optimizing the calculation model to enable the model to reach an optimal value under a training set, wherein
Figure DEST_PATH_IMAGE138
Representing the labeled true values of the training set,
Figure 104323DEST_PATH_IMAGE128
the output value of the sample data of the training set under the model,
Figure DEST_PATH_IMAGE140
、
Figure DEST_PATH_IMAGE142
is a linear coefficient, preferably
Figure DEST_PATH_IMAGE144
. According to (12), the unknown parameters of the models (7) to (11) can be solved by adopting BP algorithm
Figure DEST_PATH_IMAGE146
、
Figure DEST_PATH_IMAGE148
、
Figure DEST_PATH_IMAGE150
、
Figure DEST_PATH_IMAGE152
、
Figure 471588DEST_PATH_IMAGE098
、
Figure 632442DEST_PATH_IMAGE112
、
Figure 369454DEST_PATH_IMAGE122
、
Figure DEST_PATH_IMAGE154
.
The training set is an image of a high-entropy material coating cutter with known cracks, a key area is extracted, whether cracks exist in the neighborhood is observed manually, and the key area is marked as the crack
Figure DEST_PATH_IMAGE156
Or
Figure DEST_PATH_IMAGE158
And (5) substituting (12) the iterative computation to finish the training.
After training is completed, a neighborhood is selected
Figure 627260DEST_PATH_IMAGE056
For inputting, calculating whether a crack exists in the neighborhood according to the trained calculation model, and setting the corresponding model output as
Figure 525946DEST_PATH_IMAGE058
，
Figure 938211DEST_PATH_IMAGE060
For coefficients ordered according to equation (6), calculate:
Figure DEST_PATH_IMAGE062A
the calculation result of equation (13) is a tool state voting coefficient. And when the voting coefficient is larger than 0.3, judging that the cutter is invalid. Reducing the error of single neighborhood detection.
The invention provides a failure detection method of a high-entropy material coating cutter based on an image. According to the comparison of the following results, the detection result of the method disclosed by the invention has a small standard error with the detection standard of a microscope, and the accurate automatic detection of the failure of the high-entropy material coating cutter is realized.
Figure DEST_PATH_IMAGE160
In addition, compared with the existing image processing method, the method of the invention also has obvious advantages. For example, compared with the traditional edge detection mode, the accuracy rate of the method is improved by 72%; compared with a common ResNet neural network model, the accuracy rate is improved by 35%.
The hardware structure of the present invention can use an off-the-shelf image acquisition device. For example, an image acquisition device is arranged on a detection table of a production line, the image acquisition device is connected with a communication unit, the communication unit is connected with a remote server through a communication network, and the server is connected with a controller of the production line.
The image acquisition equipment is used for acquiring a cutter image;
the communication unit is used for transmitting the cutter image to the server;
the server is used for implementing the method, judging whether the cutter is invalid or not and feeding back a result to the controller;
the controller is used for receiving the discrimination signal fed back by the server, thereby controlling the action of the production line to eliminate unqualified cutters.
It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present invention and the technical principles employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims (10)

1. A failure detection method of a high-entropy material coating cutter based on images is characterized by comprising the following steps:
step 1: extraction of key region of high-entropy material coating cutter image
Step 1.1, establishing a probability model of the color distribution of the high-entropy material coating cutter image: performing channel activation operation on the color according to the color value of the RGB channel of the image; the activated image calculation method comprises the following steps:
Figure 419674DEST_PATH_IMAGE002
r, g, b in equation (1) represent the RGB three channels of the image;
Figure DEST_PATH_IMAGE003
、
Figure 867973DEST_PATH_IMAGE004
is a linear coefficient, and the linear coefficient,
Figure DEST_PATH_IMAGE005
、
Figure 319814DEST_PATH_IMAGE006
is composed of
Figure DEST_PATH_IMAGE007
Is a function of (a) a function of (b),
Figure 620083DEST_PATH_IMAGE007
the color value range is (0, 1)](ii) a Wherein:
Figure DEST_PATH_IMAGE009
according to activation image
Figure 231193DEST_PATH_IMAGE010
The probability of the distribution of the pixel values is used for judging the input image, and the probability that the input image is the high-entropy material coating cutter image is judged:
Figure 194601DEST_PATH_IMAGE012
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE013
for activating images
Figure 258372DEST_PATH_IMAGE010
The value of the pixel of (a) is,
Figure 598217DEST_PATH_IMAGE014
the cutter image classification label is coated for the high-entropy material,
Figure DEST_PATH_IMAGE015
representing a probability function, argmax representing making the probability function
Figure 798255DEST_PATH_IMAGE015
Taking maximum time parameter
Figure 198143DEST_PATH_IMAGE014
A value of (d);
step 1.2, extracting a key area of the high-entropy material coating cutter image from the image: to pair
Figure 218052DEST_PATH_IMAGE010
Each pixel of (2)
Figure DEST_PATH_IMAGE017
According to a probabilistic model
Figure 486222DEST_PATH_IMAGE018
Solving the classification label to obtain the key area
Figure DEST_PATH_IMAGE019
；
Step 2: state failure detection of high-entropy material coating cutter based on key area
Taking each pixel
Figure 914667DEST_PATH_IMAGE013
Of (2) a set of neighborhood pixels
Figure 610091DEST_PATH_IMAGE020
：
Figure DEST_PATH_IMAGE021
Wherein each pixel in the neighborhood
Figure 258241DEST_PATH_IMAGE022
Is that
Figure 798944DEST_PATH_IMAGE010
Neutralization of
Figure 363786DEST_PATH_IMAGE013
Is not more than
Figure DEST_PATH_IMAGE023
Is selected to be the set of pixels of (a),
Figure 167794DEST_PATH_IMAGE023
is the radius of the neighborhood;
activating an image from one
Figure 162295DEST_PATH_IMAGE010
The number of pixels in which the key region is obtained is
Figure 146169DEST_PATH_IMAGE024
Is marked as
Figure DEST_PATH_IMAGE025
(ii) a Correspondingly have
Figure 316250DEST_PATH_IMAGE024
A neighborhood of
Figure 619056DEST_PATH_IMAGE026
,
Figure DEST_PATH_IMAGE027
,…,
Figure 976219DEST_PATH_IMAGE028
Will turn over
Figure 389883DEST_PATH_IMAGE024
Each neighborhood gets a sequence according to the small-to-large arrangement of its pixel mean:
Figure 148891DEST_PATH_IMAGE030
Figure DEST_PATH_IMAGE031
representing a sequence
Figure 622598DEST_PATH_IMAGE032
One element in
Figure DEST_PATH_IMAGE033
Pixel mean of (is a neighborhood);
with a certain neighborhood mentioned above
Figure 965592DEST_PATH_IMAGE033
Establishing a calculation model for input, and calculating the high-entropy material coating cutter state voting coefficient according to input data; let the corresponding model output be
Figure 651788DEST_PATH_IMAGE034
，
Figure DEST_PATH_IMAGE035
For coefficients ordered according to equation (6), calculate:
Figure DEST_PATH_IMAGE037
the calculation result of the formula (13) is a tool state voting coefficient; and when the voting coefficient is larger than 0.3, judging that the cutter is invalid.
2. The method for detecting the failure of the high-entropy material-coated cutter based on the image as claimed in claim 1, wherein: the production line is provided with image acquisition equipment, the image acquisition equipment is connected with a communication unit, the communication unit is connected with a remote server through a communication network, and the server is connected with a controller of the production line.
3. The method for detecting the failure of the high-entropy material-coated cutter based on the image as claimed in claim 2, wherein: the image acquisition equipment is used for acquiring a tool image.
4. The method for detecting the failure of the high-entropy material-coated cutter based on the image as claimed in claim 2, wherein: the communication unit is used for transmitting the cutter image to the server.
5. The method for detecting the failure of the high-entropy material-coated cutter based on the image as claimed in claim 2, wherein: the server is used for implementing the method, judging whether the cutter is invalid or not and feeding back the result to the controller.
6. The method for detecting the failure of the high-entropy material-coated cutter based on the image as claimed in claim 2, wherein: the controller is used for receiving the discrimination signal fed back by the server, thereby controlling the production line to act to remove the unqualified cutter.
7. The method for detecting the failure of the high-entropy material-coated cutter based on the image as claimed in claim 1, wherein the method comprises the following steps: the key area refers to an image pixel set for detecting the performance state of the high-entropy material coating cutter.
8. The method for detecting the failure of the high-entropy material-coated cutter based on the image as claimed in claim 1, wherein the method comprises the following steps: solving conditional probability function
Figure 796462DEST_PATH_IMAGE038
The method comprises the following steps:
Figure 441070DEST_PATH_IMAGE040
wherein
Figure DEST_PATH_IMAGE041
Is a pixel
Figure 772825DEST_PATH_IMAGE013
Is determined by the conditional probability function of (a),
Figure 262712DEST_PATH_IMAGE042
is to classify
Figure 589789DEST_PATH_IMAGE014
Is determined by the prior probability function of (a),
Figure DEST_PATH_IMAGE043
is a pixel
Figure 280664DEST_PATH_IMAGE013
Is calculated as a function of the prior probability.
9. The method for detecting the failure of the high-entropy material-coated cutter based on the image as claimed in claim 8, wherein: the probability is a model parameter and is obtained through a training sample of the high-entropy material coating cutter image.
10. A cutter online use method is characterized in that: a method of failure detection comprising an image-based high entropy material coated tool according to claims 1-9.
CN202211087012.3A 2022-09-07 2022-09-07 Failure detection method for high-entropy material coating cutter based on image Active CN115170569B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211087012.3A CN115170569B (en) 2022-09-07 2022-09-07 Failure detection method for high-entropy material coating cutter based on image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211087012.3A CN115170569B (en) 2022-09-07 2022-09-07 Failure detection method for high-entropy material coating cutter based on image

Publications (2)

Publication Number Publication Date
CN115170569A true CN115170569A (en) 2022-10-11
CN115170569B CN115170569B (en) 2022-12-02

Family

ID=83481183

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211087012.3A Active CN115170569B (en) 2022-09-07 2022-09-07 Failure detection method for high-entropy material coating cutter based on image

Country Status (1)

Country Link
CN (1) CN115170569B (en)

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107451970A (en) * 2017-07-28 2017-12-08 电子科技大学 A kind of high dynamic range images generation method based on single-frame images
CN107464233A (en) * 2017-07-19 2017-12-12 国家电网公司 The image detecting method and system of composite insulator based on Lab color modes
CN109382702A (en) * 2018-11-06 2019-02-26 重庆大学 A kind of chain digital control gear hobbing machine rolling blade losing efficacy form automatic identifying method
CN111127499A (en) * 2019-12-20 2020-05-08 北京工业大学 Security inspection image cutter detection segmentation method based on semantic contour information
US20200177805A1 (en) * 2017-09-18 2020-06-04 Shenzhen University Pipeline detection method and apparatus, and storage medium
CN111242202A (en) * 2020-01-07 2020-06-05 燕山大学 Method for monitoring wear state of turning tool based on metric learning
CN111300144A (en) * 2019-11-25 2020-06-19 上海大学 Automatic detection method for tool wear state based on image processing
CN112192318A (en) * 2020-09-28 2021-01-08 北京天泽智云科技有限公司 Machining tool state monitoring method and system
CN112200797A (en) * 2020-10-28 2021-01-08 佛山市南海区广工大数控装备协同创新研究院 Effective training method based on PCB noise labeling data
CN112241692A (en) * 2020-09-25 2021-01-19 天津大学 Channel foreign matter intelligent detection and classification method based on aerial image super-pixel texture
CN113436204A (en) * 2021-06-10 2021-09-24 中国地质大学（武汉） High-resolution remote sensing image weak supervision building extraction method

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107464233A (en) * 2017-07-19 2017-12-12 国家电网公司 The image detecting method and system of composite insulator based on Lab color modes
CN107451970A (en) * 2017-07-28 2017-12-08 电子科技大学 A kind of high dynamic range images generation method based on single-frame images
US20200177805A1 (en) * 2017-09-18 2020-06-04 Shenzhen University Pipeline detection method and apparatus, and storage medium
CN109382702A (en) * 2018-11-06 2019-02-26 重庆大学 A kind of chain digital control gear hobbing machine rolling blade losing efficacy form automatic identifying method
CN111300144A (en) * 2019-11-25 2020-06-19 上海大学 Automatic detection method for tool wear state based on image processing
CN111127499A (en) * 2019-12-20 2020-05-08 北京工业大学 Security inspection image cutter detection segmentation method based on semantic contour information
CN111242202A (en) * 2020-01-07 2020-06-05 燕山大学 Method for monitoring wear state of turning tool based on metric learning
CN112241692A (en) * 2020-09-25 2021-01-19 天津大学 Channel foreign matter intelligent detection and classification method based on aerial image super-pixel texture
CN112192318A (en) * 2020-09-28 2021-01-08 北京天泽智云科技有限公司 Machining tool state monitoring method and system
CN112200797A (en) * 2020-10-28 2021-01-08 佛山市南海区广工大数控装备协同创新研究院 Effective training method based on PCB noise labeling data
CN113436204A (en) * 2021-06-10 2021-09-24 中国地质大学（武汉） High-resolution remote sensing image weak supervision building extraction method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
ZHAOYANG HUANG 等: "VS-Net: Voting with Segmentation for Visual Localization", 《IEEE》 *
杨延栋 等: "盘形滚刀磨损预测模型", 《煤炭学报》 *

Also Published As

Publication number Publication date
CN115170569B (en) 2022-12-02

Similar Documents

Publication Publication Date Title
CN110561193A (en) Cutter wear assessment and monitoring method and system based on feature fusion
CN110216319B (en) Medium plate optimized shearing control method based on surface quality online evaluation
CN110427825A (en) The video flame recognition methods merged based on key frame with quick support vector machines
CN108907896A (en) A kind of cutter remaining life on-line prediction method and system
CN106584209A (en) Real-time online monitoring method for tool wear based on cloud manufacturing
CN111539845B (en) Enterprise environment-friendly management and control response studying and judging method based on power consumption mode membership grade
CN109191367A (en) The joining method of tool wear image and the life-span prediction method of cutter
CN112231971B (en) Blast furnace fault diagnosis method based on relative integral trend diffusion fault sample generation
CN115047162A (en) Defect detection method and system for steel pipe heat treatment
CN108446706B (en) Automatic abrasive grain material identification method based on principal component extraction of colors
CN113962997A (en) Strip steel edge crack defect detection method and system based on image processing
CN115170569B (en) Failure detection method for high-entropy material coating cutter based on image
Wu et al. Design of online surface inspection system of hot rolled strips
Zhu et al. Multi-classification recognition and quantitative characterization of surface defects in belt grinding based on YOLOv7
CN113962954A (en) Surface defect detection method based on SE-R-YOLOV4 automobile steel part
CN116740044B (en) Copper pipe milling surface processing method and system based on visual detection and control
CN114386679A (en) Ball mill energy consumption state monitoring method based on ensemble learning
CN117001423A (en) Tool state online monitoring method based on evolutionary learning
CN116871632A (en) Robot backing welding forming defect monitoring method based on arc voltage and molten pool vision
CN115625317A (en) Surface water wave optimization processing method and system for die-casting regulation
CN105354848A (en) Optimization method of Cognex surface quality detection system of hot galvanizing production line
CN115472246A (en) Prediction method and model for delamination defect in high-temperature alloy cross-piercing process
CN114972353A (en) Quality evaluation method and system for metal high-pressure special pipe fitting
CN111680448B (en) Continuous casting billet longitudinal crack prediction method based on SVM classification
CN115099266A (en) Hard vehicle surface white layer prediction method based on gradient lifting decision tree

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