A belt tear detection method, medium and system based on image data processing

Abstract

The present invention provides a belt tear detection method, medium and system based on image data processing, which belongs to the field of image data processing technology, including: first, using multiple high-resolution cameras to collect belt surface images from different angles, and obtaining standardized images through automatic exposure adjustment and automatic focus processing. Next, extracting the grayscale feature matrix of the image, and decomposing the texture features and state features using a matrix decomposition algorithm, clustering the state feature matrix through a parallel computing method, and extracting the feature vector that characterizes the belt surface state. Then, according to the preset tear discriminant equation group, the tear risk index of each area is calculated, and threshold analysis is performed to divide different belt damage levels, and obtain the identification result of the belt damage area. The method makes full use of the processing technology of the belt image data in operation, and solves the technical problem that it is difficult to obtain the risk of belt tear damage in operation efficiently and quickly in the prior art.

Description

Technical Field

[0001] The present invention belongs to the technical field of image data processing, and in particular, relates to a belt tear detection method, medium and system based on image data processing. Background Art

[0002] Industrial belts are widely used in various transportation and transmission scenarios and are an indispensable and important component of mechanical equipment. Due to long-term operation, the belts will produce various degrees of tear damage, which seriously affects the safety and operational reliability of the equipment. Therefore, timely and accurate detection of the belt status and targeted maintenance are crucial to ensure the safe and stable operation of the production process. At present, common belt tear detection methods mainly include manual visual inspection and sensor monitoring. Manual visual inspection relies on regular inspections by staff, which requires a lot of manpower costs, and at the same time cannot guarantee the comprehensiveness and objectivity of the detection. Sensor monitoring collects the changes in physical parameters of the belt during operation in real time by installing pressure sensors, vibration sensors, etc., but due to the limitations of sensor layout and data processing algorithms, it is often difficult to comprehensively analyze the complex damage patterns on the belt surface. In other words, the existing technology has the technical problem of being difficult to efficiently and quickly obtain the tear risk of the running belt. Summary of the invention

[0003] In view of this, the present invention provides a belt tear detection method, medium and system based on image data processing, which can solve the technical problem that it is difficult to obtain the tear risk of a running belt efficiently and quickly in the prior art.

[0004] The present invention is achieved in that:

[0005] A first aspect of the present invention provides a belt tear detection method based on image data processing, comprising the following steps:

[0006] S10, using multiple high-resolution cameras to collect images of the running belt surface from different angles to obtain multiple groups of original images; specifically, multiple high-resolution camera arrays are arranged, the angle between adjacent cameras is 60°, the distance between the camera and the belt surface is d, and the following conditions are satisfied:

[0007]

[0008] Where w is the belt width; θ is the camera field of view, which ranges from 30°≤θ≤45°;

[0009] S20, performing automatic exposure adjustment and automatic focus processing on the multiple groups of original images to eliminate image quality differences caused by ambient lighting and shooting angles, and obtain standardized images; wherein the processing of the original images using the automatic exposure adjustment algorithm is specifically:

[0010] Where E(x, y) is the adjusted image pixel value; I(x, y) is the original image pixel value; I max and I min are the maximum and minimum pixel values ​​of the image respectively; k is the gain coefficient, and its value range is 1.2≤k≤1.8; b is the bias coefficient, and its value range is 0.1≤b≤0.3; the x-axis direction of the image is the direction perpendicular to the direction of belt travel, and the y-axis direction is the direction of belt travel.

[0011] S30, performing grayscale transformation processing on the standardized image to extract the belt surface grayscale feature matrix; wherein the grayscale transformation processing on the standardized image is specifically:

[0012]

[0013] Where G(x, y) is the gray value of the image at the coordinate (x, y); I(x, y) is the original image pixel value; α, β are weight coefficients, and satisfy α+β=1, 0≤α, β≤1; α=0.7, β=0.3 are the preferred values;

[0014] S40, using a matrix decomposition algorithm to decompose the gray feature matrix to obtain a belt surface texture feature matrix and a state feature matrix; specifically, using a singular value decomposition method to decompose the gray feature matrix, specifically expressed as: M = USV T +E;

[0015] Where M is the input grayscale feature matrix; U is the left singular matrix; S is the singular value matrix; V is the right singular matrix; E is the noise matrix;

[0016] Furthermore, the state feature matrix is ​​extracted:

[0017] F = V·diag(w1, w2, ..., w n );

[0018] Where F is the state feature matrix; w i (i=1, 2, ..., n) is the feature weight; through experiments, it is determined that n=5 has the best effect;

[0019] S50, using a parallel computing method to perform cluster analysis on the state feature matrix to extract the belt surface state feature vector; specifically, using a K-means clustering algorithm to analyze the state feature matrix to extract the state feature vector:

[0020]

[0021] In the formula, is the state feature vector; is the i-th cluster center; i is the weight coefficient of the i-th cluster; k is the number of clusters, and its value is 3≤k≤5; γ is the gradient weight coefficient, and its value range is 0.1≤γ≤0.3;

[0022] S60, according to the belt surface state characteristic vector, using a preset belt tearing discriminant equation group to calculate the tearing risk index of each area on the belt surface; specifically, the tearing discriminant equation group is specifically expressed as:

[0023]

[0024] Where s1, s2, and s3 represent the crack degree in the x direction, crack degree in the y direction, and deformation degree of the belt surface respectively; the coefficient is calibrated by experiment: a i ∈[0.15, 0.25], d i ∈[0.2,0.3],f i ∈[0.1, 0.2]; b1=b2=b3=0.15; c1=c2=c3=0.1;

[0025] The design basis of the x-direction crack degree judgment equation s1 is:

[0026] 1. Item 1 represents the linear combination of eigenvectors, reflecting the overall state of cracks on the belt surface, where v i is the i-th component of the state eigenvector, representing the crack characteristics at different scales; a i is the weight coefficient, the larger a i Corresponding to more important crack features;

[0027] 2. Item 2 represents the rate of change of the characteristic in the lateral direction, where The lateral gradient of the anti-crack damage state; b1 is the gradient weight coefficient, which is used to balance the importance of overall characteristics and local changes;

[0028] 3. Item 3 is the nonlinear compensation term, where It decays quickly when the eigenvalue is large to prevent excessive amplification of the impact of severe cracks; c1 is the nonlinear term coefficient, which is used to adjust the compensation strength.

[0029] The design basis of the y-direction crack degree judgment equation s2 is:

[0030] 1. Item 1 represents the linear combination of eigenvectors, where d i is the weight coefficient, the larger the d i The component corresponding to the more significant crack characteristics;

[0031] 2. Item 2 represents the rate of change of the feature in the longitudinal direction, where The longitudinal gradient reflects the crack state; cracks usually show longitudinal characteristics, so the longitudinal gradient is selected;

[0032] 3. Item 3 is the logarithmic nonlinear compensation, where When the eigenvalue is small, it changes quickly, which is conducive to the discovery of early cracks; the logarithmic function can compress the range of large values ​​to prevent the eigenvalue from being too large.

[0033] The design basis of deformation degree judgment equation s3 is:

[0034] 1. Item 1 represents the linear combination of eigenvectors, where f i is the weight coefficient, the larger f i The component corresponding to the more obvious deformation characteristics;

[0035] 2. Item 2 represents the total gradient of the feature, where Calculate the omnidirectional gradient modulus of the feature; deformation usually manifests itself as regional characteristics, and changes in all directions need to be considered;

[0036] 3. Item 3 is the periodic nonlinear compensation, where The introduction of periodic changes is suitable for describing wave-like deformations; the boundedness of the sine function can prevent the eigenvalues ​​from diverging.

[0037] The overall design considerations of the system of equations are:

[0038] 1. The three equations are respectively for different types of damage characteristics, with similar mathematical structures but different focuses: Since tearing damage can be decomposed into x-direction cracks, y-direction cracks and deformation degree, three equations can be established for separate calculations. The x-direction crack occurs in the direction perpendicular to the belt travel, so when the belt is running at high speed, the x-direction crack feature is mainly manifested as a change in surface texture; the y-direction crack occurs in the direction of the belt travel, so when the belt is running at high speed, the y-direction crack feature is mainly manifested as a local mutation; the deformation feature is mainly manifested as a regional shape change;

[0039] 2. Each equation contains linear terms, gradient terms and nonlinear compensation terms: linear terms ensure basic feature extraction capabilities; gradient terms enhance sensitivity to local changes; nonlinear terms improve performance in extreme cases;

[0040] 3. Different nonlinear function selection: exponential function is used to detect the continuous change characteristics of cracks in the x direction; logarithmic function is used for crack detection in the y direction, which is suitable for early detection of sudden damage; sine function is used for deformation detection, which is suitable for the description of periodic deformation;

[0041] 4. Principles of coefficient selection: Weight coefficient (a i , d i , f i ) is calibrated through experiments; the gradient coefficients (b1, b2, b3) are kept consistent to balance various features; the nonlinear coefficients (c1, c2, c3) take the same value to simplify parameter adjustment.

[0042] Among them, the calculation formula for the tear risk index of each area on the belt surface is:

[0043]

[0044] After experimental testing, the optimal weights are: ω1 = 0.4, ω2 = 0.35, ω3 = 0.25;

[0045] Optionally, to better consider the continuous running time of the belt, the following formula is used to better calculate the tear risk index of each area:

[0046]

[0047] T is the characteristic time constant, which is 24h; δ=0.1; t represents the continuous running time of the belt.

[0048] S70, performing threshold analysis on the tear risk index of each area on the belt surface, dividing the belt surface state level, and obtaining the belt damage area identification result, which is specifically described as follows:

[0049] When SI≤0.3, the belt is in normal condition and can continue to be used;

[0050] When 0.3<SI≤0.6, the belt has slight tear damage and needs special attention;

[0051] When 0.6<SI≤0.8, the belt is seriously torn and it is recommended to repair it in time;

[0052] When SI>0.8, the belt has severe tear damage and must be shut down immediately for inspection and maintenance.

[0053] Furthermore, the damaged area is located:

[0054]

[0055] Where P(x, y) is the damage area identification matrix; L(x, y) is the local tear risk index; L this the local threshold, and its value is 0.5.

[0056] On the basis of the above technical solution, the belt tear detection method based on image data processing of the present invention can also be improved as follows:

[0057] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned belt tear detection method based on image data processing.

[0058] A third aspect of the present invention provides a belt tear detection system based on image data processing, which includes the above-mentioned computer-readable storage medium.

[0059] Compared with the prior art, the belt tear detection method, medium and system based on image data processing provided by the present invention have the following beneficial effects:

[0060] 1. Image acquisition is more intelligent. By adopting a multi-angle camera array and combining technologies such as automatic exposure adjustment and autofocus, high-quality, standardized belt surface images can be obtained, providing a reliable data basis for subsequent analysis.

[0061] 2. Feature extraction is more efficient. The present invention uses simple and efficient matrix operations such as grayscale transformation and singular value decomposition, without the need for complex deep learning algorithms, which greatly reduces the amount of calculation and processing time, and meets the real-time detection requirements in high-speed operation scenarios.

[0062] 3. The result interpretation is more detailed. The present invention not only provides the overall tear risk index of the belt, but also can make a precise quantitative analysis of the damage degree of each area on the belt surface, providing maintenance personnel with clearer maintenance suggestions.

[0063] In summary, the present invention solves the technical problem in the prior art that it is difficult to efficiently and quickly obtain the risk of tearing of the running belt. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0066] like Figure 1 FIG. 1 is a flow chart of a belt tear detection method based on image data processing provided by the first aspect of the present invention. The method comprises the following steps:

[0067] S10, collecting images of the surface of the running belt from different angles using multiple high-resolution cameras to obtain multiple groups of original images;

[0068] S20, performing automatic exposure adjustment and automatic focus processing on the multiple groups of original images to eliminate image quality differences caused by ambient lighting and shooting angles, and obtain standardized images;

[0069] S30, performing grayscale transformation processing on the standardized image to extract the grayscale feature matrix of the belt surface;

[0070] S40, decomposing the gray feature matrix using a matrix decomposition algorithm to obtain a belt surface texture feature matrix and a state feature matrix;

[0071] S50, using a parallel computing method to perform cluster analysis on the state feature matrix to extract the belt surface state feature vector;

[0072] S60, calculating the tear risk index of each area on the belt surface according to the belt surface state characteristic vector using a preset belt tear discrimination equation group;

[0073] S70, performing threshold analysis on the tear risk index of each area on the belt surface, dividing the belt surface condition level, and obtaining the belt damage area identification result.

[0074] The specific implementation methods of the above steps are described in detail below:

[0075] The specific implementation of step S10 is: by arranging multiple high-resolution camera arrays, images of the running belt surface are collected from different angles. The angle between adjacent cameras is set to 60°, and the distance d from the camera to the belt surface satisfies the formula: Where w is the belt width and θ is the camera field of view, ranging from 30°≤θ≤45°. In this way, multiple sets of high-resolution original images covering the entire belt can be obtained. The arrangement of the camera array takes into account factors such as the field of view and the distance between cameras, which can ensure the collection of comprehensive and high-quality belt surface image data, laying the foundation for subsequent image processing and analysis.

[0076] The specific implementation of step S20 is: perform automatic exposure adjustment and automatic focus processing on the original image obtained in step S10 to eliminate the image quality differences caused by factors such as ambient light and shooting angle, and obtain a standardized image. The automatic exposure adjustment adopts the formula: Where E(x, y) is the adjusted image pixel value, I(x, y) is the original image pixel value, and I max and I minare the maximum and minimum pixel values ​​of the image, respectively; k is the gain coefficient with a range of 1.2≤k≤1.8; b is the bias coefficient with a range of 0.1≤b≤0.3. By automatically adjusting the exposure parameters k and b, the influence of ambient light can be eliminated and the image can be standardized. At the same time, the autofocus algorithm ensures that clear belt surface texture details are obtained, providing high-quality data input for subsequent image analysis.

[0077] The specific implementation of step S30 is: grayscale transformation processing is performed on the standardized image to extract the grayscale feature matrix of the belt surface. The grayscale transformation formula is: Where G(x, y) is the grayscale value of the image at the coordinate (x, y), I(x, y) is the original image pixel value, α, β are weight coefficients and satisfy α+β=1, 0≤α, β≤1. The preferred values ​​are α=0.7, β=0.3. The first term α·I(x, y) reflects the average grayscale information of the pixel, and the second term It reflects the local grayscale gradient information. By setting the values ​​of α and β reasonably, the grayscale feature matrix rich in belt surface texture features can be extracted.

[0078] The specific implementation of step S40 is: using a matrix decomposition algorithm to decompose the gray feature matrix obtained in step S30 to obtain the belt surface texture feature matrix and state feature matrix. Specifically, the singular value decomposition (SVD) algorithm is used, and the formula is: M = USV T +E; where M is the input grayscale feature matrix, U is the left singular matrix, S is the singular value matrix, V is the right singular matrix, and E is the noise matrix. Through SVD decomposition, the original grayscale feature matrix M can be decomposed into three basic matrices U, S and V. Among them, the right singular matrix V contains the state feature information of the belt surface, so the state feature matrix can be further extracted: F = V·diag(w1, W2, ..., w n ), where w i (i=1, 2, ..., n) is the feature weight, and the best effect is determined by experiments when n=5. In this way, the feature matrix F containing the belt surface state information is obtained.

[0079] The specific implementation of step S50 is: using a parallel computing method to perform cluster analysis on the state feature matrix F obtained in step S40 to extract the belt surface state feature vector. Specifically, the K-means clustering algorithm is used, and the formula is: in, is the state feature vector, is the i-th cluster center, λ iis the weight coefficient of the ith cluster, k is the number of clusters and its value range is 3≤k≤5, and γ is the gradient weight coefficient and its value range is 0.1≤γ≤0.3. Through the K-means clustering algorithm, the various belt damage features contained in the state feature matrix F can be separated to extract the feature vector that comprehensively reflects the belt state. Parallel computing methods can greatly improve the computational efficiency of clustering.

[0080] The specific implementation of step S60 is: according to the belt surface state feature vector obtained in step S50 The calculation is performed using three preset belt state discrimination equations:

[0081] The x-direction crack degree judgment equation:

[0082] The y-direction crack degree judgment equation:

[0083] Deformation degree discriminant equation:

[0084] The coefficient a i ∈[0.15, 0.25], d i ∈[0.2,0.3],f i ∈[0.1, 0.2], b1=b2=b3=0.15, c1=c2=c3=0.1. The three equations are respectively aimed at different types of damage characteristics. By reasonably setting the linear term, gradient term and nonlinear term, the tearing state of the belt surface can be comprehensively evaluated.

[0085] The specific implementation of step S70 is: threshold analysis is performed on the tear risk index of each belt surface area calculated in step S60, the belt surface state level is divided, and the damaged area is located and marked. The calculation formula of the comprehensive tear risk index SI is: Among them, the calculation formula of the tear risk index of each belt surface area is:

[0086]

[0087] After experimental testing, the optimal weights are: ω1 = 0.4, ω2 = 0.35, ω3 = 0.25;

[0088] Optionally, to better consider the continuous running time of the belt, the following formula is used to better calculate the tear risk index of each area:

[0089]

[0090] T is the characteristic time constant, which is 24h; δ=0.1; t represents the continuous running time of the belt.

[0091] The belt status level is divided according to the size of SI: when SI ≤ 0.3, the belt is in normal condition; when 0.3 < SI ≤ 0.6, there is slight damage; when 0.6 < SI ≤ 0.8, the damage is serious; when SI > 0.8, there is serious damage. Furthermore, the damaged area is located according to the local tear risk index L(x, y). When L(x, y) > L th Time (L th =0.5) is marked as the damaged area. In summary, this method can comprehensively analyze the state of the belt surface and accurately locate the damaged area, providing a basis for repair and maintenance.

[0092] In summary, the present invention proposes a belt tear detection method based on image data processing. High-quality belt surface images are collected through a multi-camera array, and the feature matrix containing the belt status information is extracted after automatic exposure adjustment, grayscale transformation, singular value decomposition and other processing. The feature matrix is ​​decomposed using the K-means clustering algorithm to obtain a feature vector that comprehensively reflects the belt status. The tear risk index of each area of ​​the belt is further calculated according to the preset discriminant equation, and threshold analysis is performed to comprehensively evaluate the tear status of the belt and accurately locate the damaged area. This method utilizes a variety of advanced image processing and pattern recognition algorithms, which can realize intelligent detection and diagnosis of belt status, providing effective support for equipment maintenance.

[0093] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned belt tear detection method based on image data processing.

[0094] A third aspect of the present invention provides a belt tear detection system based on image data processing, which includes the above-mentioned computer-readable storage medium.

[0095] Specifically, the principle of the present invention is described as follows:

[0096] 1. Analysis of the microscopic mechanism of belt tearing

[0097] From the microscopic perspective of materials science, conveyor belts are usually composite materials consisting of a rubber matrix and a reinforced fabric layer. During high-speed operation, due to the combined effects of mechanical stress, thermal stress and environmental factors, stress distribution in different directions will be generated inside the material:

[0098] (1) Transverse stress (σ x ):

[0099] Generation mechanism: The belt is subjected to lateral stretching and compression during operation;

[0100] Stress distribution:

[0101] Where E is the elastic modulus, ε x is the lateral strain, v is the Poisson's ratio, u y is the longitudinal displacement;

[0102] (2) Longitudinal stress (σ y ):

[0103] Generation mechanism: The belt is subjected to longitudinal tension and dynamic impact;

[0104] Stress distribution:

[0105] Among them, ε y is the longitudinal strain, u x is the lateral displacement;

[0106] (3) Shear stress (τ xy ):

[0107] Production mechanism: The belt is subjected to torsion and uneven deformation;

[0108] Stress distribution:

[0109] Where, G is the shear modulus;

[0110] 2. Direction analysis of crack propagation;

[0111] Based on fracture mechanics theory, the crack propagation of materials follows the principle of maximum energy release rate, which can be expressed as:

[0112]

[0113] Where: G is the energy release rate; K I is the cracking stress intensity factor; K II is the sliding stress intensity factor; K III is the tearing stress intensity factor; E′ is the equivalent elastic modulus;

[0114] This shows that crack propagation can be decomposed into three basic modes:

[0115] 1. Cracking mode (Mode I): corresponds to cracks in the x direction;

[0116] 2. Sliding type (Mode II): corresponding to the crack in the y direction;

[0117] 3.Tear-off type (Mode III): corresponds to deformation damage;

[0118] 3. Physical explanation of characteristic manifestations

[0119] (1) The surface texture characteristics of the crack in the x direction are expressed as a differential equation:

[0120] Where T represents the surface texture function, α is the anisotropy coefficient;

[0121] When transverse cracks appear, the texture of the local area will change continuously. This change is mathematically expressed as the gradient change of the texture function. Therefore, the exponential function e is selected. -||v|| To describe this gradual change.

[0122] (2) The mutation characteristics of the crack in the y direction are expressed as a differential equation:

[0123] Among them, S represents the mutation function, β and γ are characteristic coefficients;

[0124] Longitudinal cracks often appear as local sudden fractures. This mutation corresponds to a function jump in mathematics. Therefore, the logarithmic function ln(1+||v||) is selected to capture this mutation feature.

[0125] (3) The regional characteristics of deformation are expressed as differential equations:

[0126] Where D represents the deformation function, f(x, y) is the external force distribution function;

[0127] Deformation usually manifests itself as a change in the shape of a continuous area, which has periodicity and spatial continuity. Therefore, the sine function sin(||v||) is selected to describe this periodic change characteristic.

[0128] 4. Theoretical basis of feature fusion

[0129] Based on the above analysis, the overall tear risk index can be expressed as:

[0130]

[0131] This combination is based on the following theoretical foundations:

[0132] 1. Orthogonality: The three damage modes are physically independent of each other;

[0133] 2. Energy superposition: follow the principle of strain energy density superposition;

[0134] 3. Weight allocation: based on the impact of various damage modes on overall safety.

[0135] In order to better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is provided below: A 500-meter-long belt conveyor is used in the production workshop of a chemical plant to transport raw materials from the storage area to the production workshop. Since the belt has been running in a high temperature and high vibration environment for a long time, it has suffered varying degrees of tearing damage. In order to detect problems in a timely manner and guide maintenance personnel to perform targeted maintenance, the factory decided to adopt the belt tear detection method based on image data processing proposed in the present invention.

[0136] In the first step, five high-resolution digital cameras were evenly arranged about 2 meters above the belt transmission line. The angle between adjacent cameras was set to 60°, and the distance d from the camera to the belt surface satisfied The belt width w = 1.2m, and the camera field of view angle θ = 40°. In this way, the five cameras can fully cover the entire surface of the 500-meter-long belt and clearly capture the belt texture and other detailed features.

[0137] The second step is to perform automatic exposure adjustment and autofocus processing on the original captured image. The automatic exposure adjustment algorithm uses the formula Where E(x, y) is the adjusted image pixel value, I(x, y) is the original image pixel value, and I max =255 and I min =0 are the maximum and minimum pixel values ​​of the 8-bit grayscale image, k=1.5 is the gain coefficient, and b=0.2 is the bias coefficient. The brightness of the image after adjustment is uniform and consistent, providing a good data basis for subsequent analysis.

[0138] The third step is to perform grayscale transformation on the standardized image. The grayscale transformation formula is: Where G(x, y) is the grayscale value of the image at the coordinate (x, y), and I(x, y) is the pixel value of the original image. Through this weighted average and gradient calculation method, the grayscale feature matrix M rich in belt surface texture features can be extracted.

[0139] The fourth step is to decompose the gray feature matrix M using the singular value decomposition (SVD) algorithm, the formula is M = USV T +E. Among them, U is the left singular matrix, S is the singular value matrix, V is the right singular matrix, and E is the noise matrix. Further, the state characteristic matrix F = V·diag(0.2, 0.25, 0.15, 0.2, 0.1) is extracted from the right singular matrix V, with a total of 5 characteristic components. In this way, the characteristic matrix F containing the belt surface state information is obtained.

[0140] The fifth step is to use the K-means clustering algorithm to analyze the state feature matrix F. The formula is: in is the state feature vector, There are three cluster centers respectively. is the gradient of the feature matrix. Through K-means clustering, the various damage features contained in the state feature matrix are successfully separated to obtain a feature vector that comprehensively reflects the belt state. in, (Calculated based on adjacent point spreads) 0.12 (calculated based on the adjacent point difference) (calculated based on the gradient vector), the running time is 12h.

[0141] Step 6: Based on the feature vector obtained in step 5 The tear risk index of each area on the belt surface is calculated using three preset discriminant equations:

[0142] Crack severity index in x direction:

[0143] s1=0.11+0.072+0.063+0.082+0.108+0.0225+0.1·e -1.29 ;

[0144] s1=0.551;

[0145] Crack severity index in y direction:

[0146] s2=0.1375+0.144+0.126+0.041+0.108+0.018+0.1·ln(2.29);

[0147] s2=0.632;

[0148] Deformation index:

[0149] s3=0.055+0.072+0.0945+0.1025+0.144+0.0285+0.1·

[0150] sin(1.29);

[0151] s3=0.576;

[0152] Comprehensive tear risk index:

[0153] SI = 0.594;

[0154] According to the above calculation results, the following analysis can be obtained:

[0155] 1. The tearing degree of the belt surface is relatively light, and the tearing risk index s1=0.551 is within the normal range;

[0156] 2. There are cracks on the belt surface to a certain extent, and the tear risk index s2 = 0.632 belongs to the level of slight tear damage;

[0157] 3. The belt surface has a certain degree of deformation, and the tear risk index s3 = 0.576 also belongs to the category of slight tear damage;

[0158] 4. The comprehensive tear risk index SI = 0.594 shows that the belt is in good overall condition, but special attention should be paid to tearing and deformation problems.

[0159] In order to further locate the damaged area on the belt surface, the local tear risk index L(x, y) of each point was calculated. The results are shown in Table 1 below:

[0160] Table 1 Point local tear risk index table

[0161] area Partial tear risk index Damage level 1 0.48 normal 2 0.62 slight 3 0.53 normal 4 0.71 More serious 5 0.58 slight 6 0.49 normal 7 0.66 More serious 8 0.52 normal

[0162] As can be seen from the table, the belt surface area 4 and 7 have serious damage, which should be the priority for attention and repair. The damage in other areas is relatively minor, so inspection and maintenance can be strengthened appropriately.

[0163] In general, by adopting the image recognition method proposed in the present invention, the chemical plant can fully and accurately grasp the running status of the belt, which can not only give an overall comprehensive evaluation, but also accurately locate the damaged area, providing an important basis for subsequent maintenance work. Compared with traditional manual inspection and sensor monitoring methods, this method greatly improves the accuracy and automation level of detection. At the same time, due to the use of simple and efficient image processing algorithms, it can also meet the requirements of real-time detection, and is very suitable for application in the production environment of the chemical plant with long belts and high-speed operation.

[0164] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A belt tear detection method based on image data processing, characterized in that: Including the following steps: S10. Collect images of the running belt surface from different angles by multiple high-resolution cameras to obtain multiple groups of original images; S20. Perform automatic exposure adjustment and autofocus processing on the multiple groups of original images to eliminate the image quality differences caused by environmental illumination and shooting angles, and obtain standardized images; S30. Perform gray-scale transformation processing on the standardized images to extract the gray-scale feature matrix of the belt surface; S40. Decompose the gray-scale feature matrix by using a matrix decomposition algorithm to obtain the texture feature matrix and the state feature matrix of the belt surface; S50. Adopt a parallel computing method to perform clustering analysis on the state feature matrix to extract the state feature vector of the belt surface; S60. According to the state feature vector of the belt surface, calculate the tearing risk index of each area on the belt surface by using a preset belt tearing discrimination equation set; S70. Perform threshold analysis on the tearing risk index of each area on the belt surface, divide the state level of the belt surface, and obtain the belt damage area identification result.

2. The belt tear detection method based on image data processing according to claim 1 is characterized in that: The arrangement of the multiple high-resolution cameras meets the following requirements: The included angle between adjacent cameras is 60°, and the distance from the camera to the belt surface is d, satisfying: In the formula, w is the belt width; θ is the camera field of view angle, and the value range is 30° ≤ θ ≤ 45°.

3. The belt tear detection method based on image data processing according to claim 2 is characterized in that: The automatic exposure adjustment is specifically to process the original images by using an automatic exposure adjustment algorithm: Where E(x,y) is the adjusted image pixel value; I(x,y) is the original image pixel value; I max and I min are the maximum and minimum pixel values ​​of the image respectively; k is the gain coefficient, and its value range is 1.2≤k≤1.8; b is the bias coefficient, and its value range is 0.1≤b≤0.3; the x-axis direction of the image is the direction perpendicular to the direction of belt travel, and the y-axis direction is the direction of belt travel.

4. The belt tear detection method based on image data processing according to claim 3 is characterized in that: Performing gray-scale transformation processing on the standardized images specifically is: In the formula, G(x, y) is the gray value of the image at the coordinate (x, y); I(x, y) is the original image pixel value; α, β are weight coefficients, and satisfy α + β = 1, 0 ≤ α, β ≤ 1.

5. The belt tear detection method based on image data processing according to claim 4 is characterized in that: Decomposing the gray-scale feature matrix specifically is: M=USV T +E; In the formula, M is the input gray-scale feature matrix; U is the left singular matrix; S is the singular value matrix; V is the right singular matrix; E is the noise matrix.

6. The belt tear detection method based on image data processing according to claim 5 is characterized in that: The state feature vector of the belt surface specifically is: In the formula, is the state feature vector; is the i-th cluster center; λ i is the weight coefficient of the ith cluster; k is the number of clusters, ranging from 3≤k≤5; γ is the gradient weight coefficient, ranging from 0.1≤γ≤0.

3.

7. The belt tear detection method based on image data processing according to claim 6 is characterized in that: The tearing discrimination equation set specifically includes: Where s1, s2, s3 represent the crack degree in the x direction, crack degree in the y direction and deformation degree of the belt surface respectively; v i is the i-th component of the state eigenvector, n represents the number of components of the state eigenvector, and the weight coefficient (a i ,d i ,f i ) is calibrated through experiments; the gradient coefficients (b1, b2, b3) are kept consistent to balance various features; the nonlinear coefficients (c1, c2, c3) take the same value to simplify parameter adjustment.

8. The belt tear detection method based on image data processing according to claim 7 is characterized in that: The tearing risk index specifically is: When SI ≤ 0.3, the belt is in normal state; When 0.3 < SI ≤ 0.6, the belt has slight tearing damage; When 0.6 < SI ≤ 0.8, the belt has relatively serious tearing damage; When SI > 0.8, the belt has serious tearing damage.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions run in a computer, they are used to execute a belt tearing detection method according to any one of claims 1-8.

10. A belt tear detection system based on image data processing, characterized in that: Including the computer-readable storage medium according to claim 9.

Images