CN113642597A - Self-making method of defect image data set of lining layer of solid rocket engine - Google Patents
Self-making method of defect image data set of lining layer of solid rocket engine Download PDFInfo
- Publication number
- CN113642597A CN113642597A CN202110699614.3A CN202110699614A CN113642597A CN 113642597 A CN113642597 A CN 113642597A CN 202110699614 A CN202110699614 A CN 202110699614A CN 113642597 A CN113642597 A CN 113642597A
- Authority
- CN
- China
- Prior art keywords
- defect
- lining
- random
- image
- self
- 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
Links
- 230000007547 defect Effects 0.000 title claims abstract description 168
- 238000000034 method Methods 0.000 title claims abstract description 40
- 239000007787 solid Substances 0.000 title claims abstract description 31
- 238000007665 sagging Methods 0.000 claims description 22
- 239000011248 coating agent Substances 0.000 claims description 20
- 238000000576 coating method Methods 0.000 claims description 20
- 238000013519 translation Methods 0.000 claims description 18
- 239000003086 colorant Substances 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 3
- 238000000465 moulding Methods 0.000 claims description 2
- 238000012549 training Methods 0.000 abstract description 8
- 238000013528 artificial neural network Methods 0.000 abstract description 5
- 238000004519 manufacturing process Methods 0.000 abstract description 5
- 238000004088 simulation Methods 0.000 description 8
- 238000001514 detection method Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- 238000005498 polishing Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 239000003380 propellant Substances 0.000 description 2
- 230000007306 turnover Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000002485 combustion reaction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000009413 insulation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000013021 overheating Methods 0.000 description 1
- -1 sagging Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000003313 weakening effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The invention relates to the field of manufacturing of label data sets required for training a neural network, and discloses a self-made method of a defect image data set of a solid rocket engine lining. A large number of marked lining defect images can be obtained by the method and can be used for training a neural network.
Description
Technical Field
The invention relates to the field of manufacturing of label data sets required by training of neural networks, in particular to a self-control method of a defect image data set of a lining layer of a solid rocket engine.
Background
The solid rocket engine is used as a power device of a missile or a space vehicle, and has specific requirements on the coating integrity of the lining of the inner wall of the engine shell. The liner primarily functions to bond the propellant to the insulation or casing, and is required to be strong and able to withstand the stresses encountered during operation of the engine. The lining is not firmly bonded, and the phenomenon of debonding can cause overheating and weakening of the inside of the shell or expansion of the combustion surface of the propellant, so that the engine fails.
Under the existing lining forming technology, the surface of the lining may have the defects of missing coating, sagging, air bubbles, accumulation and the like, the integrity of the surface of the lining is poor, and the bonding performance of the lining is reduced. The solid rocket engine generally has the characteristic of large length-diameter ratio, the light inside the shell is weak, and the effect is poor by adopting a mode of manually detecting the defects of the lining layer, so that a new method for detecting the defects of the lining layer is very necessary to be explored. The method is a novel detection method for detecting the lining defects by adopting the technology based on computer vision and deep learning, can automatically complete the detection requirements of the lining defects and locate the positions of the defects by establishing a neural network lining defect identification model. However, building a lining defect identification model requires a large amount of lining defect image data. For a solid rocket engine, the production capacity is less, the number of defects of the lining layer of the inner wall of the engine is less, and the image data volume requirement required by training a neural network model cannot be met.
In conclusion, the automatic detection of the lining layer defects can be realized by adopting a deep learning method to establish a lining layer defect detection model, but a necessary lining layer image data set is lacked. Therefore, it is necessary to self-manufacture the liner defect image data set according to the forming method of the liner of the solid rocket engine shell.
Disclosure of Invention
The invention aims to overcome the defects and provides a self-making method of a defect image data set of a solid rocket engine lining, which is characterized in that a subset image is taken out from the upper left corner to the lower right corner of an original lining image in a specific step length, the subset image and a defect graph generated by a defect generator are combined randomly with equal probability, and the problem that a neural network model cannot be trained by applying a deep learning method due to the lack of a lining defect image data set at present is solved.
In order to achieve the above purpose, the invention provides the following technical scheme:
a solid rocket engine lining defect image data set self-making method comprises the following steps:
(1) acquiring a flawless engine shell lining image as an original lining image;
(2) sequentially extracting subset images with pixels as a x a from the original lining image in a mode of step length as a, wherein the number of the subset images is more than or equal to 100;
(3) generating a defect graph by a defect generator, wherein the defect graph comprises an uncoated defect, a sagging defect, a bubble defect and a stacking defect graph;
(4) randomly combining the subset image obtained in the step (2) with the defect pattern obtained in the step (3) in an equal probability manner to obtain a combined image;
(5) setting a corresponding label for the combined image obtained in the step (4) and then storing the combined image;
(6) and (4) performing data enhancement on all the combined images saved in the step (5) at least once.
Further, in the step (1), the pixels of the original lining image are more than or equal to 4096 × 2048; in the step (2), a subset image with 64 × 64 pixels is sequentially extracted from the original underlayer image, with the step size a being 64.
Further, in the step (1), acquiring an original lining image by a linear array camera to acquire a defect-free engine shell lining in a cast molding mode;
in the step (3), the missing coating defect pattern is an ellipse, the coordinate of the outline of the ellipse in a coordinate system taking the center of the ellipse as an origin is (x, y), and | x | or | y | is obtained by adding the basic value and the variation value; the sagging defect pattern is in a keyway shape, and the keyway shape is defined by a lower arc, an upper arc, a left straight line and a right straight line; the bubble defect graph is circular, and the circular radius is obtained by adding the basic value and the variation value;
in the step (6), the data enhancement comprises normalization, random width translation, random height translation, random rotation and random horizontal inversion.
Further, in the step (3), the contour curve of the piled defect pattern is a random closed curve in the rectangular area of the a × a pixels.
Further, the method for obtaining the random closed curve comprises the steps of firstly drawing a basic closed curve in a rectangular area of a pixel a and a pixel a; secondly, randomly changing coordinate points on the basic closed curve, wherein the random change value is larger than zero and smaller than a; and finally, selecting a closed curve of the rectangular area of the a x a pixels as an outline curve of the stacking defect pattern.
Further, in the step (4), the equal probability random combination method is to divide the interval a into 4 equal length sub-intervals a1, a2, A3 and a4, to correspond to the missing coating defect, the sagging defect, the bubble defect and the piled-up defect pattern, respectively, to generate a random number in the interval a, and to combine the defect pattern corresponding to the random number with the subset image according to the sub-interval to which the random number belongs.
Further, in the step (3), filling colors inside the defect pattern; and (4) in the step (3), filling the inside of the bubble defect graph into white.
Further, in the step (3), the missing coating defect pattern is an ellipse, the coordinate of the outline of the ellipse in a coordinate system with the center of the ellipse as the origin is (x, y), | x | or | y | is obtained by adding a base value and a variation value, the base value is 32, and the variation value is any value within the range of (10,30) or (-30, -10).
Further, in the step (3), the sagging defect pattern is in a key groove shape, and the key groove shape is defined by a lower arc, an upper arc, a left straight line and a right straight line;
the distance between the centers of the upper arc and the lower arc is any value within the range of [13.33,21.33 ]; the radius of the upper arc and the lower arc is any value within the range of [11.25,18.75 ]; the central angle corresponding to the upper arc and the lower arc is any value in the range of [150 degrees and 250 degrees ].
Further, in the step (3), the bubble defect pattern is a circle, the radius of the circle is obtained by adding a base value and a variation value, the base value is 20, and the variation value is in a range of [ -2.5,2.5 ].
Further, in the step (6), the data enhancement comprises normalization, standardization, random width translation, random height translation, random rotation and random horizontal turnover; the range of the random width translation is [ -0.15,0.15 ]; the range of random height translation is [ -0.15,0.15 ]; the random rotation angle was 45 °.
Compared with the prior art, the invention has the following beneficial effects:
(1) the invention relates to a self-made method of a solid rocket engine lining defect image data set, which is characterized in that subset images are sequentially taken from the upper left corner to the lower right corner of an original lining image in a specific step length, taking a preferred scheme as an example, the size of each original lining image is 4096 x 2048, the subset images (or subset areas) with the size of 64 x 64 are sequentially selected to be randomly combined with a defect generator, the combined number of one original image is 4096/64 x 2048/64 x 2048, when the pixels of the original lining image are larger or the pixels of the subset images are smaller, the combined number which is far larger than 2048 can be obtained, the number of training data sets is greatly expanded, and the 64 area can fully display about 3mm of defects;
(2) the invention relates to a self-making method of a solid rocket engine lining defect image data set, which designs four common defect graph forms, wherein four defects in a defect generator are determined in the form of basic graphs or coordinate points, so that the shape and pixels are convenient to adjust, and the actual lining defects can be effectively simulated;
(3) the invention relates to a self-making method of a solid rocket engine lining defect image data set, which is characterized in that a subset image and a defect graph are combined randomly with equal probability, the storage format of the subset is 'original lining image name _64 x 64 subset region position sequence number', and the defect position can be conveniently positioned according to the position sequence number;
(4) according to the self-making method of the solid rocket engine lining defect image data set, after data enhancement is carried out on the combined result, the data volume can be increased by at least one time, the generalization capability of the model can be improved, and the overfitting problem is restrained.
Drawings
FIG. 1 is a schematic diagram of a 64 x 64 subset area and a marking method of the solid rocket engine lining defect image dataset self-making method of the invention;
FIG. 2 is four defect edge shapes for the solid rocket engine lining defect image dataset self-manufacturing method of the present invention; wherein, FIGS. 2(a) and (b) are edge shapes of the missing coating defect; FIGS. 2(c), (d) are edge shapes of sagging defects; FIGS. 2(e), (f) are schematic views showing the shape of the edge of the blow-by bubble; FIGS. 2(g), (h) are stacking fault edge shapes;
FIG. 3 is a flow chart of a method for self-preparing a defect image data set of a solid rocket engine lining according to the present invention.
Detailed Description
The features and advantages of the present invention will become more apparent and appreciated from the following detailed description of the invention.
The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments. While the various aspects of the embodiments are presented in drawings, the drawings are not necessarily drawn to scale unless specifically indicated.
A self-making method of a solid rocket engine lining defect image data set for training a neural network model comprises the following steps:
(1) acquiring a flawless engine shell lining image as an original lining image;
(2) taking subset images with pixels of a from the upper left corner to the lower right corner in the original lining image one by one in a mode of step length a, wherein the number of the subset images is more than or equal to 100;
(3) generating a defect graph by a defect generator, wherein the defect graph comprises an uncoated defect, a sagging defect, a bubble defect and a stacking defect graph;
(4) randomly combining the subset image obtained in the step (2) with the defect pattern obtained in the step (3) in an equal probability manner to obtain a combined image;
(5) setting a corresponding label for the combined image obtained in the step (4) and then storing the combined image; after the combination is completed, the combination result and the corresponding label need to be stored, after one original lining layer image is completed, one original lining layer image is taken out to be combined and stored, and the steps are repeated until all the lining layer images are combined and stored;
(6) and (4) performing data enhancement on all combined image results saved in the step (5) at least once.
Further, the self-making method of the defect image data set of the lining layer of the solid rocket engine comprises the following steps:
(1) acquiring a defect-free engine shell lining image as an original lining image, wherein the pixel of the original lining image is 4096 × 2048;
(2) sequentially extracting a subset image with 64 x 64 pixels from the original underlayer image in a mode of step size being 64;
(3) generating a defect graph by a defect generator, wherein the defect graph comprises an uncoated defect, a sagging defect, a bubble defect and a stacking defect graph;
(4) randomly combining the subset image obtained in the step (2) with the defect pattern obtained in the step (3) in an equal probability manner to obtain a combined image;
(5) setting a corresponding label for the combined image obtained in the step (4) and then storing the combined image;
(6) and (4) performing data enhancement on all the combined images saved in the step (5) at least once.
Further, in the step (3), the missing coating defect pattern is an ellipse, the coordinate of the outline of the ellipse in the coordinate system with the center of the ellipse as the origin is (x, y), | x | or | y | is obtained by adding a base value and a variation value, the base value is 32, and the variation value is any value in the range of (10,30) or (-30, -10).
Further, in the step (3), the sagging defect pattern is in a key groove shape, and the key groove shape is defined by a lower arc, an upper arc, a left straight line and a right straight line;
the distance between the centers of the upper arc and the lower arc is any value within the range of [13.33,21.33 ]; the radius of the upper arc and the lower arc is any value within the range of [11.25,18.75 ]; the central angle corresponding to the upper arc and the lower arc is any value in the range of [150 degrees and 250 degrees ].
Further, in the step (3), the bubble defect pattern is a circle, the radius of the circle is obtained by adding a base value and a variation value, the base value is 20, and the variation value is in a range of [ -2.5,2.5 ].
Further, in the step (3), the contour curve of the piled defect pattern is a random closed curve in the rectangular area of the a × a pixels.
Further, the method for obtaining the random closed curve comprises the steps of firstly drawing a basic closed curve in a rectangular area of a pixel a and a pixel a; secondly, randomly changing the coordinate point on the basic closed curve, wherein the random change value is larger than zero and smaller than 64; and finally, selecting a closed curve of the rectangular area of the a x a pixels as an outline curve of the stacking defect pattern.
Further, in the step (4), the equal probability random combination method is to divide the interval a into 4 equal length sub-intervals a1, a2, A3 and a4, to correspond to the missing coating defect, the sagging defect, the bubble defect and the piled-up defect pattern, respectively, to generate a random number in the interval a, and to combine the defect pattern corresponding to the random number with the subset image according to the sub-interval to which the random number belongs. The equiprobable random combination is controlled by a random number, and the subset image is combined with a defect pattern in the defect generator according to the size of the random number.
Further, in the step (3), filling colors inside the defect pattern; and (4) in the step (3), filling the inside of the bubble defect graph into white.
Further, in the step (1), the original lining layer image is obtained by acquiring a defect-free engine shell lining layer in a cast-coating forming mode by a linear array camera.
Further, in the step (6), the data enhancement comprises normalization, standardization, random width translation, random height translation, random rotation and random horizontal turnover; the range of the random width translation is [ -0.15,0.15 ]; the range of random height translation is [ -0.15,0.15 ]; the random rotation angle was 45 °.
In the present invention, parentheses are used to indicate the values of the intervals excluding the endpoints, and parentheses are used to indicate the values of the intervals including the endpoints.
Example 1
As shown in fig. 1-3, the present embodiment provides a technical solution: a solid rocket engine lining defect image data set self-control method comprises an original lining image, wherein the original lining image is obtained by acquiring a defect-free lining of an engine shell through a linear array camera, the pixel size of each image of the original lining image is 4096 x 2048, 64 x 64 subset images are sequentially taken out from the upper left corner to the lower right corner of the lining image one by one according to a mode that the step length is 64, as shown in figure 1, the subset represents a row with i and a column with k, the row with i is marked through i-k, the subset is stored in a format of 'original lining image name _64 x 64 subset area position sequence number', and the defect position can be conveniently positioned according to the position sequence number; and combining the subset images with defect patterns generated by a defect generator at equal probability randomly, wherein the defect patterns comprise missing coating defects, sagging defects, bubble defects and stacking defect patterns, the equal probability random combination is controlled by a random number, combining the subset images with one of the defect generators according to the size of the random number, storing a combination result and a corresponding label after the combination is finished, taking out one original lining image, combining and storing the original lining image, repeating the steps until all the lining images are combined and stored, and finally performing data enhancement on all the combined results.
The basic pattern of the missing coating defect is an ellipse, the ellipse is drawn by a coordinate point (x, y), the coordinate (x, y) takes the center of the ellipse as an origin, the basic value is 32, the variation value comprises a major axis variation value and a minor axis variation value, the major axis variation value is used for calculating the coordinate x, the minor axis variation value is used for calculating the coordinate y, the range of the major axis variation value and the minor axis variation value is (10,30) or (-30, -10), the coordinate group (x, y) is approximately enclosed into an elliptical contour, the elliptical contour needs to be filled inside, and the filling can specify colors. The skip-coating defects have randomness and simulation, wherein the randomness refers to that the shapes of the skip-coating defects generated every time are different, and the simulation refers to that the actual skip-coating defects in a lining layer polishing and forming mode can be accurately simulated.
The basic figure of the sagging defect is 'keyway-shaped', the keyway-shaped comprises a lower arc, an upper arc, a left straight line and a right straight line, the lower arc and the upper arc are drawn by coordinate points (x, y), the distance between the center of the lower arc and the center of the upper arc randomly varies, the random variation range is [13.33,21.33], the random variation range of the radius of the lower arc and the radius of the upper arc is [11.25,18.75], the random variation range of the center angle of the lower arc and the center angle of the upper arc is [150 degrees ], 250 degrees ], the left straight line is formed by connecting the counterclockwise starting point of the lower arc and the counterclockwise ending point of the upper arc, the right straight line is formed by connecting the counterclockwise ending point of the lower arc and the counterclockwise starting point of the upper arc, and the inside of the sagging defect can be filled with colors. The sagging defects have randomness and simulation, wherein the randomness refers to that the shapes of the sagging defects generated every time are different, and the simulation refers to that the actual sagging defects in a lining throwing and coating forming mode can be accurately simulated.
The bubble defect base pattern is a circle, the radius of the circle is obtained by adding a base value and a variation value, the base value is 20, the variation value is [ -2.5,2.5], and the inside of the bubble defect is generally white. The bubble defects have randomness and simulation, wherein the randomness refers to that the shapes of the bubble defects generated each time are different, and the simulation refers to that the actual bubble defects in a lining layer polishing and forming mode can be accurately simulated.
The shape of the edge curve of the stacking fault is obtained by combining coordinate points and random variation values, the coordinate points draw a zigzag and smooth closed curve in a 64-by-64 rectangular area by using a spline curve through SolidWorks, dozens of points are densely selected on the curve, the random variation values depend on the addition and multiplication products of the coordinate values and the random numbers, the coordinate points and the random variation values are larger than zero and smaller than 64, and the interior of the stacking fault can be filled with colors. The stacking defects have randomness and simulation, wherein the randomness refers to that the shapes of the stacking defects generated each time are different, and the simulation refers to that the actual stacking defects in a lining layer polishing and forming mode can be accurately simulated.
The original lining image and the defect generator are combined in an equal probability way, namely, the original lining image with the size of 4096 × 2048 is combined with one of the defect generators by taking out 64 × 64 subset areas and one combination of the defect generators one by one from the upper left corner to the lower right corner in step 64 and storing the combination to a designated storage path, the combination is an equal probability combination, the equal probability combination is controlled by a random number in a [0,1) interval, if the random number is greater than or equal to 0 and less than 0.2, the 64 × 64 subset area is combined with the missing coating defect, the combined image and record label are stored as 1, if the random number is greater than or equal to 0.2 and less than 0.4, the 64 × 64 subset area is combined with the sagging defect, the combined image and record label are stored as 2, if the random number is greater than or equal to 0.4 and less than 0.6, the 64 × 64 subset area is combined with the bubble defect, and saving the combined image and recording label as 3, if the random number is greater than or equal to 0.6 and less than 0.8, combining the 64 × 64 subset area with the sagging defect, and saving the combined image and recording label as 4, if the random number is greater than or equal to 0.8 and less than 1, directly saving the 64 × 64 subset area and recording label as 5, firstly, sequentially combining one original lining image, and then combining the next lining image after completing one lining image, and repeating the steps until all the original lining images are completed.
The data enhancement comprises normalization, standardization, random width translation [ -0.15,0.15], random height translation [ -0.15,0.15], random rotation of 45 degrees and random horizontal inversion, after the original underlayer image and the defect generator are combined, the data enhancement is carried out on the combined result once, the training data can be doubled, and the training samples are copied once according to the original sequence and added at the back.
In conclusion, the working process of the invention is as follows: firstly, a plurality of lining images in a defect-free cast-coating forming mode are collected by using a linear array camera, the size of each lining image is 4096 x 2048, one lining image is taken out, 64 x 64 subset images are sequentially taken out from the upper left corner to the lower right corner of the image according to a mode of step length 64, the subset images and one of defect generators are combined according to an equal probability mode, a combination result and corresponding labels are stored, after one lining image is completed, one lining image is taken out to be combined and stored, the process is repeated until all the lining images are combined and stored, finally, all the combined results are subjected to data enhancement once, all the stored label texts are copied and added behind the last line. This results in a large number of labeled images of the underlying defects that can be used to train neural networks.
The invention has been described in detail with reference to specific embodiments and illustrative examples, but the description is not intended to be construed in a limiting sense. Those skilled in the art will appreciate that various equivalent substitutions, modifications or improvements may be made to the technical solution of the present invention and its embodiments without departing from the spirit and scope of the present invention, which fall within the scope of the present invention. The scope of the invention is defined by the appended claims.
Those skilled in the art will appreciate that those matters not described in detail in the present specification are well known in the art.
Claims (11)
1. A solid rocket engine lining defect image data set self-making method is characterized by comprising the following steps:
(1) acquiring a flawless engine shell lining image as an original lining image;
(2) sequentially extracting subset images with pixels as a x a from the original lining image in a mode of step length as a, wherein the number of the subset images is more than or equal to 100;
(3) generating a defect graph by a defect generator, wherein the defect graph comprises an uncoated defect, a sagging defect, a bubble defect and a stacking defect graph;
(4) randomly combining the subset image obtained in the step (2) with the defect pattern obtained in the step (3) in an equal probability manner to obtain a combined image;
(5) setting a corresponding label for the combined image obtained in the step (4) and then storing the combined image;
(6) and (4) performing data enhancement on all the combined images saved in the step (5) at least once.
2. The method for self-making the solid rocket engine lining defect image data set according to claim 1, wherein in the step (1), the pixels of the original lining image are more than or equal to 4096 x 2048; in the step (2), a subset image with 64 × 64 pixels is sequentially extracted from the original underlayer image, with the step size a being 64.
3. The method for self-making the image data set of the defects of the lining of the solid rocket engine according to claim 1 or 2,
in the step (1), the original lining layer image is obtained by acquiring a defect-free engine shell lining layer in a cast molding mode by a linear array camera;
in the step (3), the missing coating defect pattern is an ellipse, the coordinate of the outline of the ellipse in a coordinate system taking the center of the ellipse as an origin is (x, y), and | x | or | y | is obtained by adding the basic value and the variation value; the sagging defect pattern is in a keyway shape, and the keyway shape is defined by a lower arc, an upper arc, a left straight line and a right straight line; the bubble defect graph is circular, and the circular radius is obtained by adding the basic value and the variation value;
in the step (6), the data enhancement comprises normalization, random width translation, random height translation, random rotation and random horizontal inversion.
4. The method for self-making the image data set of the defects of the lining of the solid rocket engine according to claim 1 or 2, wherein in the step (3), the outline curve of the stacking defect pattern is a random closed curve in a rectangular area of a pixel.
5. The method for self-making the solid rocket engine lining defect image data set according to claim 4, wherein the random closed curve is obtained by firstly drawing a basic closed curve in a rectangular region of a pixels; secondly, randomly changing coordinate points on the basic closed curve, wherein the random change value is larger than zero and smaller than a; and finally, selecting a closed curve of the rectangular area of the a x a pixels as an outline curve of the stacking defect pattern.
6. The method for self-making the defect image data set of the lining of the solid rocket engine according to claim 1 or 2, wherein in the step (4), the equal probability random combination method comprises the steps of dividing the interval A into 4 equal length sub-intervals A1, A2, A3 and A4, respectively corresponding to the missing coating defect, the sagging defect, the bubble defect and the stacking defect pattern, generating random numbers in the interval A, and combining the defect pattern corresponding to the random numbers with the sub-image according to the sub-interval to which the random numbers belong.
7. The method for self-making the defect image data set of the lining of the solid rocket engine according to claim 1 or 2, wherein in the step (3), the defect patterns are filled with colors; and (4) in the step (3), filling the inside of the bubble defect graph into white.
8. The method for self-making a defect image data set of a solid rocket engine lining according to claim 2, wherein in the step (3), the missing coating defect pattern is an ellipse, the coordinate of the outline of the ellipse in a coordinate system with the center of the ellipse as the origin is (x, y), | x | or | y | is obtained by adding a base value and a variable value, the base value is 32, and the variable value is any value in the range of (10,30) or (-30, -10).
9. The method for self-making the defect image data set of the lining of the solid rocket engine according to claim 2, wherein in the step (3), the sagging defect pattern is in a shape of a key groove, and the shape of the key groove is defined by a lower arc, an upper arc, a left straight line and a right straight line;
the distance between the centers of the upper arc and the lower arc is any value within the range of [13.33,21.33 ]; the radius of the upper arc and the lower arc is any value within the range of [11.25,18.75 ]; the central angle corresponding to the upper arc and the lower arc is any value in the range of [150 degrees and 250 degrees ].
10. The method for self-making a defect image data set of a solid rocket engine lining according to claim 2, wherein in the step (3), the defect pattern of the bubble is a circle, the radius of the circle is obtained by adding a base value and a variation value, the base value is 20, and the variation value is in the range of [ -2.5,2.5 ].
11. The method for self-making a solid rocket engine lining defect image data set according to claim 2, wherein in the step (6), data enhancement comprises normalization, random width translation, random height translation, random rotation and random horizontal inversion; the range of the random width translation is [ -0.15,0.15 ]; the range of random height translation is [ -0.15,0.15 ]; the random rotation angle was 45 °.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110699614.3A CN113642597B (en) | 2021-06-23 | 2021-06-23 | Self-control method for liner defect image data set of solid rocket engine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110699614.3A CN113642597B (en) | 2021-06-23 | 2021-06-23 | Self-control method for liner defect image data set of solid rocket engine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113642597A true CN113642597A (en) | 2021-11-12 |
| CN113642597B CN113642597B (en) | 2024-09-27 |
Family
ID=78416091
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110699614.3A Active CN113642597B (en) | 2021-06-23 | 2021-06-23 | Self-control method for liner defect image data set of solid rocket engine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113642597B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114937004A (en) * | 2022-04-28 | 2022-08-23 | 南通启锦智能科技有限公司 | Method for detecting surface pore defects of mechanical parts based on computer vision |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106442557A (en) * | 2015-08-06 | 2017-02-22 | 青岛软控机电工程有限公司 | Detection method and defect detection system for composite part joint defect detection system |
| CN108596876A (en) * | 2018-03-28 | 2018-09-28 | 潍坊路加精工有限公司 | A kind of method of automatic setting detection zone |
| US20190272627A1 (en) * | 2018-03-05 | 2019-09-05 | Element Ai Inc. | Automatically generating image datasets for use in image recognition and detection |
| CN111524100A (en) * | 2020-04-09 | 2020-08-11 | 武汉精立电子技术有限公司 | Defect image sample generation method and device and panel defect detection method |
| CN111582294A (en) * | 2019-03-05 | 2020-08-25 | 慧泉智能科技（苏州）有限公司 | Method for constructing convolutional neural network model for surface defect detection and application thereof |
-
2021
- 2021-06-23 CN CN202110699614.3A patent/CN113642597B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106442557A (en) * | 2015-08-06 | 2017-02-22 | 青岛软控机电工程有限公司 | Detection method and defect detection system for composite part joint defect detection system |
| US20190272627A1 (en) * | 2018-03-05 | 2019-09-05 | Element Ai Inc. | Automatically generating image datasets for use in image recognition and detection |
| CN108596876A (en) * | 2018-03-28 | 2018-09-28 | 潍坊路加精工有限公司 | A kind of method of automatic setting detection zone |
| CN111582294A (en) * | 2019-03-05 | 2020-08-25 | 慧泉智能科技（苏州）有限公司 | Method for constructing convolutional neural network model for surface defect detection and application thereof |
| CN111524100A (en) * | 2020-04-09 | 2020-08-11 | 武汉精立电子技术有限公司 | Defect image sample generation method and device and panel defect detection method |
Non-Patent Citations (3)
| Title |
|---|
| FIRECAT全宏: ""halcon视觉缺陷检测系列（3）表面缺陷检测数据集汇总及其相关项目推荐"", pages 1, Retrieved from the Internet <URL:https://juejin.cn/post/6963598984145272839> * |
| GONGJIE ZHANG: ""Defect-GAN: High-Fidelity Defect Synthesis for Automated Defect Inspection"", 《ARXIV》, pages 1 - 16 * |
| 戴铭: ""基于先验分布活动轮廓模型的纹理缺陷检测"", 《控制与决策》, pages 1226 - 1230 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114937004A (en) * | 2022-04-28 | 2022-08-23 | 南通启锦智能科技有限公司 | Method for detecting surface pore defects of mechanical parts based on computer vision |
| CN114937004B (en) * | 2022-04-28 | 2024-01-02 | 烟台大视工业智能科技有限公司 | Method for detecting surface air hole defects of mechanical part based on computer vision |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113642597B (en) | 2024-09-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104036544B (en) | A kind of building roof method for reconstructing based on on-board LiDAR data | |
| CN111582104B (en) | Remote sensing image semantic segmentation method and device based on self-attention feature aggregation network | |
| CN103971338B (en) | Variable-block image repair method based on saliency map | |
| CN110084304A (en) | A kind of object detection method based on generated data collection | |
| CN107392947A (en) | 2D 3D rendering method for registering based on coplanar four point set of profile | |
| CN104123417B (en) | A kind of method of the image segmentation based on Cluster-Fusion | |
| CN104463925B (en) | Based on the digital camouflage method for designing of spotting template Automatic Combined | |
| CN110399880A (en) | Recognition methods, device and the equipment of a kind of characters on license plate and license plate classification | |
| CN113642597A (en) | Self-making method of defect image data set of lining layer of solid rocket engine | |
| CN106936964A (en) | A kind of mobile phone screen angular-point detection method based on Hough transformation template matches | |
| CN110757804A (en) | Full-color layered slicing algorithm based on texture mapping three-dimensional model | |
| CN115223054A (en) | Remote sensing image change detection method based on partition clustering and convolution | |
| CN106296773A (en) | A kind of method and apparatus generating multi-level vectogram | |
| CN109872396A (en) | A kind of quick cross section profile generation method suitable for triangle grid model | |
| CN109409388B (en) | Dual-mode deep learning descriptor construction method based on graphic primitives | |
| CN110097581B (en) | Method for constructing K-D tree based on point cloud registration ICP algorithm | |
| CN113516123A (en) | Detection and identification method for tire embossed characters | |
| CN113989793A (en) | Graphite electrode embossed seal character recognition method | |
| CN106469460B (en) | The method and apparatus and electronic equipment that mosaic image generates | |
| US7339584B1 (en) | Method of generating a surface mesh | |
| CN113205019A (en) | Method for detecting defective insulator | |
| CN114529666B (en) | Three-dimensional scene rapid construction method based on fusion of oblique photography and next generation modeling technology | |
| CN114425856B (en) | Irregular target scanning and printing method based on three-dimensional camera | |
| Verma et al. | Dual mesh convolutional networks for human shape correspondence | |
| CN118279491B (en) | Urban live-action three-dimensional model construction system and method based on deep learning |
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 |