A metal material wear positioning system and method based on machine vision

Abstract

The application discloses a metal material wear positioning system and method based on machine vision, and relates to the technical field of visual analysis.The application obtains real-time information parameters and calculates real-time metal health scores, collects real-time environment parameters to calculate real-time imaging parameters, and then re-photographs the metal surface, and sequentially evaluates the real-time severity scores of various wear types.If the severity threshold is exceeded, the system will prompt that a high-risk defect is detected and suggest shutdown for maintenance.If the severity threshold is not exceeded, the comprehensive wear score is calculated to determine the wear grade of the metal and is updated to the system, which provides a scientific basis for equipment maintenance, optimizes the maintenance plan, more efficiently allocates maintenance resources to the most needed places, and improves the operation efficiency of the equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of visual analysis, in particular to a metal material wear positioning system and method based on machine vision. BACKGROUND

[0002] The existing metal material wear detection methods mainly include manual visual inspection and image processing automatic detection methods. The manual inspection relies on the experience of the operator and has the problems of strong subjectivity and low efficiency. The image processing automatic detection method directly uses the general image segmentation and edge detection algorithm and is not optimized for the unique optical reflection characteristics and wear texture changes of different types of metal materials, resulting in a significant decrease in wear positioning accuracy in strong light reflection, low contrast or complex texture background.

[0003] Therefore, how to effectively reduce light interference and suppress noise, improve detection accuracy and stability in strong light reflection, low contrast and other situations, and quickly locate the defect position so that maintenance personnel can repair it in time is particularly important in practical applications. SUMMARY

[0004] The purpose of the present application is to provide a metal material wear positioning system and method based on machine vision to solve the problems in the prior art.

[0005] In order to solve the above technical problems, the present application provides the following technical scheme: a metal material wear positioning method based on machine vision, the method comprising:

[0006] Step S100: image acquisition is performed by an image acquisition device, abnormal image areas are identified, and a diagnosis record is generated. Based on the historical diagnosis record, the normalized information parameters and image parameters of the metal to be detected are extracted, the metal default score and the metal image score are calculated respectively, the metal health score is calculated by comprehensively considering the metal default score and the metal image score, and the metal health score range corresponding to each health level is determined according to the metal health score of all historical diagnosis records;

[0007] Step S200: for the historical diagnosis records of the same health level, the environment parameters and imaging parameters of the historical diagnosis records are combined, an imaging parameter model is trained and generated by a random forest model, at the same time, the historical diagnosis records without abnormality are screened, the average value of the image feature parameters is calculated and defined as the image standard feature parameter, and the historical diagnosis records with abnormality are screened to calculate the wear index of each wear level and generate a wear type list;

[0008] Step S300: Set the training cycle, calculate the metal health score of the diagnostic record, determine the health level corresponding to the diagnostic record, collect the environmental parameters of the detection area, obtain the imaging parameters through the imaging parameter model corresponding to the health level, upload them to the image acquisition device, re-capture the secondary surface image, divide the secondary surface image into several grid units, calculate the deviation score of each grid unit, and determine the high-frequency unit corresponding to each wear type.

[0009] Step S400: Obtain the deviation score of the high-frequency unit corresponding to each wear type in a certain diagnostic record, and calculate the severity threshold of the wear type based on the deviation score;

[0010] Step S500: Obtain real-time information parameters and calculate real-time metal health score. Collect real-time environmental parameters to calculate real-time imaging parameters, then re-photograph the metal surface. According to the wear type list, evaluate the real-time severity score of each wear type in turn. If the severity threshold is exceeded, the system will prompt that a high-risk defect has been detected and suggest stopping the machine for maintenance. If the severity threshold is not exceeded, calculate the comprehensive wear score to determine the wear level of the metal and update it to the system.

[0011] Furthermore, step S100 includes:

[0012] Step S101: By arranging several environmental monitoring devices in the detection area, environmental parameters in the detection area are collected. The metal to be detected is placed in the detection platform. The metal to be detected is image-acquired by the image acquisition device, abnormal image areas are identified, and a diagnostic record is generated.

[0013] Step S102: Obtain historical diagnostic records, extract the information parameters of the metal to be detected from a specific historical diagnostic record, normalize the information parameters, and calculate the default score for the metal according to the following formula:

[0014] ;

[0015] Where A represents the default rating for metals, B a C is represented as the normalized value of the a-th information parameter. a Let b represent the weight of the a-th information parameter, and b represent the total number of information parameters.

[0016] Step S103: Obtain the surface image of the metal to be detected from the historical diagnostic records, collect the image parameters in the surface image, normalize the image parameters, and calculate the metal image score according to the following formula:

[0017] ;

[0018] Where D represents the metal image score, E d F is represented as the normalized value of the d-th image parameter. d Let represent the weight of the d-th image parameter, and e represent the total number of image parameters;

[0019] Step S104: Summarize the default metal score and metal image score of a historical diagnostic record, and calculate the metal health score according to the following formula:

[0020] ;

[0021] Where G represents the metal health score, α1 represents the weight of the default metal score, and α2 represents the weight of the metal image score.

[0022] Step S105: Summarize the metal health scores of all historical diagnostic records, calculate the average f and variance g of the metal health scores, preset k health levels, where k is a non-zero positive integer, and calculate the threshold H for the kth interval. k =f+k×g, determine the range of metal health scores corresponding to each health level as {[G min ,H1),[H2,H3),...,[H k G max ]}, where G min and G max H1, H2, and H3 represent the minimum and maximum values ​​of the metal health score, respectively, and represent the threshold values ​​of the 1st, 2nd, and 3rd intervals, respectively.

[0023] By deploying various environmental monitoring and image acquisition devices, environmental parameters and surface images of the metal to be tested can be collected comprehensively and in real time. Abnormal image areas can be identified in a timely manner, and detailed diagnostic records can be generated, thereby providing reliable data support for subsequent decision-making.

[0024] By normalizing the information and image parameters in historical diagnostic records and calculating scores based on weights, a scientific metal scoring system was constructed. The structured scoring method can effectively reflect the health status of metal materials and provide quantitative support for maintenance decisions.

[0025] By aggregating the health scores of all historical diagnostic records and calculating their mean and variance, the health level of metal materials can be clearly classified. This not only improves management efficiency but also helps decision-makers quickly identify metal materials that need maintenance or replacement, effectively reducing potential operational risks.

[0026] The dynamic scoring and evaluation mechanism of this method can respond promptly to environmental changes and material wear, providing real-time health assessments for construction and operation. This flexibility not only improves production efficiency but also enhances the safety and reliability of equipment.

[0027] Furthermore, step S200 includes:

[0028] Step S201: Summarize historical diagnostic records at the same health level, obtain environmental parameters and imaging parameters of the image acquisition device within the detection area of ​​the historical diagnostic records at a certain health level, normalize the environmental parameters and imaging parameters, use the environmental parameters and imaging parameters as training data pairs, summarize the training data pairs of all historical diagnostic records to generate a training dataset, adopt a random forest model, use the environmental parameters as the input features of the random forest model, use the imaging parameters as the output features of the random forest model, randomly divide the training dataset into several subsets, construct a decision tree based on each subset, the splitting of the decision tree is based on minimizing the mean square error, each decision tree independently learns the mapping relationship between environmental parameters and imaging parameters, after training, the final prediction output is obtained by averaging the prediction results of multiple decision trees, save the trained random forest model and define it as the imaging parameter model, and obtain the imaging parameter model corresponding to each health level;

[0029] Step S202: Filter historical diagnostic records that have not experienced any abnormalities, collect the image feature parameters of the historical diagnostic records, summarize each image feature parameter corresponding to all historical diagnostic records, calculate the average value of each image feature parameter, and define the average value as the standard image feature parameter;

[0030] Step S203: Filter historical diagnostic records that have anomalies, summarize historical diagnostic records with the same wear type, count the number of historical diagnostic records for a certain wear type that have been repaired and the equipment has returned to normal operation as z1, calculate the repair success rate Z as Z=z1 / z2, where z2 represents the historical diagnostic records of the wear type, and define the repair success rate as the degree of reversibility; collect the repair time of a certain historical diagnostic record in a certain wear type, summarize the repair time of all historical diagnostic records of the wear type, calculate the average repair time, define the average repair time as the degree of harm, calculate the wear index of each wear level by weighted summation of the degree of reversibility and the degree of harm, sort the wear types according to the wear index from large to small, and generate a wear type list;

[0031] By summarizing historical diagnostic records of the same health level, the system can extract and integrate environmental and imaging parameters of the detection area. This systematic data processing method not only improves the efficiency of data utilization but also lays a solid foundation for subsequent analysis.

[0032] Normalizing environmental and imaging parameters eliminates computational interference caused by different units and ranges, thereby ensuring the comparability and accuracy of training data, enhancing the robustness of the model, and making subsequent predictions more stable and reliable.

[0033] The model is trained using a random forest model, which independently learns the complex mapping relationship between environmental parameters and imaging parameters using multiple decision trees. This learning method effectively improves prediction accuracy, reduces the risk of overfitting that may be caused by a single model, and ensures the generalizability and adaptability of the model in practical applications.

[0034] By minimizing the mean squared error as the basis for splitting the decision tree, the model is able to more accurately capture the intrinsic relationship between the environment and imaging parameters during the learning process, thereby improving the model's predictive ability and providing reliable guidance for subsequent health assessments.

[0035] By calculating wear types, the highest-risk wear types can be effectively identified, which helps optimize the inspection process and allows resources to be prioritized for the wear types that require the most attention, thus improving inspection efficiency and targeting.

[0036] Furthermore, step S300 includes:

[0037] Step S301: Select several consecutive days as the training period, obtain the diagnostic records within the training period, calculate the metal health score of each diagnostic record, determine the health level corresponding to the diagnostic record, collect the environmental parameters of the detection area and input them into the imaging parameter model, calculate the imaging parameters, upload the imaging parameters to the image acquisition device, and re-capture the metal to be detected to obtain a secondary surface image.

[0038] Step S302: Divide the secondary surface image into several grid cells, label and record each grid cell, collect the image feature parameters of each grid cell, and calculate the absolute difference between each image feature parameter and the standard image feature parameters to obtain the difference of each image feature parameter. Calculate the deviation score of each grid cell according to the following formula:

[0039] ;

[0040] Where P represents the deviation score, L g J is represented as the difference of the g-th image feature parameter. g This is represented as the weight of the g-th image feature parameter;

[0041] Step S303: Obtain the diagnostic records of the abnormality, extract the wear type and the corresponding grid cell in the diagnostic records, summarize the diagnostic records of the same wear type, count the number of times each grid cell appears, calculate the occurrence frequency of each grid cell, preset the occurrence frequency threshold, set the grid cell that exceeds the occurrence frequency threshold as a high frequency cell, determine the high frequency cell corresponding to each wear type, and summarize the deviation score of each high frequency cell in each diagnostic record.

[0042] By selecting several consecutive days as the training period, the system can obtain stable diagnostic records. Long-term data accumulation helps improve the reliability and accuracy of the model, ensuring that the calculation of health scores is more stable and reflects the true health status of metal materials.

[0043] The collected environmental parameters are input into the imaging parameter model, enabling dynamic calculation and optimization of imaging parameters. By re-enhancing the image of the metal under test, the generated secondary surface image can more comprehensively reflect the actual state of the metal, enhancing the timeliness and accuracy of the image data.

[0044] Dividing the secondary surface image into several grid units allows for a more detailed analysis of the wear condition in each area. The high-resolution processing helps in the identification of local wear, making the overall assessment more comprehensive and detailed, and improving the accuracy of monitoring.

[0045] By calculating the absolute difference with the standard feature parameters of the image, a deviation scoring mechanism is formed, which can quantify the changes in image feature parameters, help researchers quickly identify anomalies, and take timely measures to prevent further wear or damage and reduce economic losses.

[0046] By acquiring diagnostic records of abnormalities and extracting the wear type and its corresponding mesh cell, the system can effectively statistically analyze the wear situation and summarize records of the same wear type. This not only improves the efficiency of data analysis but also helps identify potential risk areas.

[0047] By statistically analyzing the frequency of grid cells and setting frequency thresholds, high-frequency cells can be effectively identified. High-frequency cells often indicate severe wear or potential problems, providing an important basis for subsequent maintenance decisions, helping to carry out targeted repairs and prevention, and reducing the risk of equipment failure.

[0048] Furthermore, step S400 includes:

[0049] Step S401: In the set of diagnostic records where anomalies occur, collect the deviation score of the high-frequency unit corresponding to each wear type in a certain diagnostic record, summarize the deviation scores of all high-frequency units, calculate the average deviation score of each wear type in the diagnostic record, and define the average deviation score as the severity score.

[0050] Step S402: Summarize the severity scores of a certain wear type, calculate the average value u and standard deviation t of the severity scores, and calculate the severity threshold P' of the wear type as P'=u+v×t, where v represents a preset constant;

[0051] By summarizing the deviation scores of the high-frequency units corresponding to each wear type in the abnormal diagnostic records, the average deviation score is calculated and defined as the severity score. This ensures the accurate assessment of each wear type, enabling the detection system to identify severe wear conditions in a timely manner and avoid greater losses due to delays.

[0052] By calculating the severity score for each type of wear, the system can quantify the wear status, facilitating effective control by managers, providing support for decision-making, and enabling more scientific and rational maintenance strategies to optimize the allocation of maintenance resources.

[0053] Furthermore, step S500 includes:

[0054] Step S501: Obtain real-time information parameters of the metal to be detected, calculate the real-time default score of the metal, collect image parameters of the metal to be detected, calculate the real-time image score of the metal, calculate the real-time health score of the metal based on the real-time default score and the real-time image score of the metal, and determine the real-time health level corresponding to the metal to be detected.

[0055] Step S502: Collect real-time environmental parameters within the detection area and input them into the imaging parameter model corresponding to the real-time health level to calculate the real-time imaging parameters. Then, input the real-time imaging parameters into the image acquisition device to re-capture the metal to be detected and obtain a secondary real-time surface image.

[0056] Step S503: Based on the order of the wear type list, collect the high-frequency unit corresponding to a wear type in sequence, calculate the real-time severity score of the wear type, compare the real-time severity score with the severity threshold, if it is greater than or equal to the severity threshold, interrupt the detection and jump to step S504, if it is less than the severity threshold, continue to calculate the real-time severity score of the next wear type and compare it with the severity threshold.

[0057] Step S504: If an interruption occurs during detection, the system will indicate that a high-risk defect has been detected and recommend immediate shutdown for repair. The system will also specify the location of the high-frequency unit corresponding to a real-time severity score greater than or equal to the severity threshold. If no interruption occurs, the comprehensive wear score will be calculated using the following formula:

[0058] ;

[0059] Where Q represents the overall wear score, and X n Y represents the real-time severity score for the nth wear type. n The value represents the weight of the nth wear type, and m represents the total number of wear types. Several wear levels are preset. The wear level is determined based on the comprehensive wear score of the metal to be tested and uploaded to the system.

[0060] By acquiring real-time information parameters of the metal to be detected and calculating the metal health score, the system can reflect the current status of the metal in a timely manner. The real-time monitoring mechanism enables users to grasp the health status of the equipment in an instant, reduce potential risks, and improve management efficiency.

[0061] The real-time environmental parameter acquisition function enables the imaging parameter model to be dynamically adjusted according to environmental changes, ensuring the accuracy and effectiveness of image acquisition, making the subsequent evaluation data more reliable, and adapting to different operating environments.

[0062] If the system detects that the severity score exceeds the threshold, it will interrupt the detection and immediately prompt the user to stop the machine for maintenance. The early warning mechanism can not only effectively prevent equipment accidents, but also reduce production losses caused by equipment failure and improve equipment reliability.

[0063] The comprehensive wear rating calculation mechanism can take into account the severity and weight of each wear type to form a comprehensive assessment of the equipment condition. The data-driven approach not only simplifies the judgment process but also improves the scientific nature and accuracy of decision-making.

[0064] To better implement the above method, a metal material wear localization system based on machine vision is also proposed. The system includes a health level module, an imaging analysis module, a training cycle module, a severity threshold module, and a real-time detection module.

[0065] Health level module: Image acquisition is performed through image acquisition device, abnormal image areas are identified and diagnostic records are generated. Based on historical diagnostic records, normalized information parameters and image parameters of the metal to be detected are extracted, and the default score and metal image score of the metal are calculated respectively. The metal health score is calculated by combining the default score and metal image score, and each health level and the corresponding metal health score range are determined according to the metal health scores of all historical diagnostic records.

[0066] Imaging analysis module: For historical diagnostic records of the same health level, the environmental parameters and imaging parameters of the historical diagnostic records are combined, and an imaging parameter model is generated by training through a random forest model. At the same time, historical diagnostic records without abnormalities are screened, the average value of the image feature parameters is calculated and defined as the standard image feature parameters, and historical diagnostic records with abnormalities are screened, the wear index of each wear level is calculated, and a list of wear types is generated.

[0067] Training cycle module: Set the training cycle, calculate the metal health score of the diagnostic record, determine the health level corresponding to the diagnostic record, collect environmental parameters of the detection area, obtain imaging parameters through the imaging parameter model corresponding to the health level, upload them to the image acquisition device, re-capture the secondary surface image, divide the secondary surface image into several grid units, calculate the deviation score of each grid unit, and determine the high-frequency unit corresponding to each wear type.

[0068] Severity threshold module: Obtain the deviation score of the high-frequency unit corresponding to each wear type in a certain diagnostic record, and calculate the severity threshold of the wear type based on the deviation score;

[0069] Real-time detection module: Acquires real-time information parameters and calculates real-time metal health score, collects real-time environmental parameters to calculate real-time imaging parameters, then re-images the metal surface, and evaluates the real-time severity score of each wear type according to the wear type list. If the severity threshold is exceeded, the system will prompt that a high-risk defect has been detected and suggest shutdown for maintenance; if the severity threshold is not exceeded, the comprehensive wear score is calculated to determine the wear level of the metal and updated to the system.

[0070] Furthermore, the health rating module includes a metal health scoring unit and a health rating unit:

[0071] Metal Health Scoring Unit: By deploying several environmental monitoring devices in the detection area to collect environmental parameters within the detection area, the metal to be detected is placed in the detection platform. Image acquisition devices are used to acquire images of the metal to be detected, abnormal image areas are identified, diagnostic records are generated, historical diagnostic records are obtained, information parameters of the metal to be detected in a certain historical diagnostic record are extracted, and the information parameters are normalized to calculate the default metal score. Surface images of the metal to be detected collected in the historical diagnostic records are obtained, image parameters in the surface images are collected, and the image parameters are normalized to calculate the metal image score. The default metal score and metal image score of a certain historical diagnostic record are summarized to calculate the metal health score.

[0072] Health Level Unit: This unit summarizes the metal health scores from all historical diagnostic records, calculates the average f and variance g of the metal health scores, and predefines k health levels, where k is a non-zero positive integer. The threshold for the kth interval is calculated as H. k =f+k×g, determine the range of metal health scores corresponding to each health level as {[G min ,H1),[H2,H3),...,[H k G max ]}, where G min and G max H1, H2, and H3 represent the minimum and maximum values ​​of the metal health score, respectively, and represent the threshold values ​​of the 1st, 2nd, and 3rd intervals, respectively.

[0073] Furthermore, the training cycle module includes a secondary imaging unit and a high-frequency determination unit:

[0074] Secondary imaging unit: Select several consecutive days as the training period, acquire diagnostic records within the training period, calculate the metal health score of each diagnostic record, determine the health level corresponding to the diagnostic record, collect environmental parameters of the detection area and input them into the imaging parameter model, calculate the imaging parameters, upload the imaging parameters to the image acquisition device, and re-image the metal to be detected to obtain a secondary surface image.

[0075] Determine high-frequency units: Divide the secondary surface image into several grid units, label and record each grid unit, collect the image feature parameters of each grid unit, and calculate the absolute difference between each image feature parameter and the standard image feature parameters to obtain the difference of each image feature parameter. Calculate the deviation score, obtain the diagnostic records of abnormalities, extract the wear types and the corresponding grid units in the diagnostic records, summarize the diagnostic records of the same wear type, count the number of times each grid unit appears, calculate the occurrence frequency of each grid unit, preset the occurrence frequency threshold, and set the grid units that exceed the occurrence frequency threshold as high-frequency units to determine the high-frequency units corresponding to each wear type.

[0076] Furthermore, the real-time detection module includes a real-time acquisition unit and a wear analysis unit:

[0077] Real-time acquisition unit: acquires real-time information parameters of the metal to be detected, calculates the real-time default score of the metal, acquires image parameters of the metal to be detected, calculates the real-time image score of the metal, calculates the real-time health score of the metal based on the real-time default score and the real-time image score of the metal, determines the real-time health level corresponding to the metal to be detected, acquires real-time environmental parameters in the detection area, and inputs them into the imaging parameter model corresponding to the real-time health level, calculates the real-time imaging parameters, and inputs the real-time imaging parameters into the image acquisition device to re-capture the metal to be detected and obtain a secondary real-time surface image;

[0078] Wear unit analysis: Based on the order of the wear type list, high-frequency units corresponding to one wear type are collected sequentially, and the real-time severity score of the wear type is calculated. The real-time severity score is compared with the severity threshold. If it is greater than or equal to the severity threshold, the detection is interrupted and the process jumps to step S504. If it is less than the severity threshold, the real-time severity score of the next wear type is calculated and compared with the severity threshold. If the detection is interrupted, the system prompts that a high-risk defect has been detected, recommends immediate shutdown for maintenance, and indicates the location of the high-frequency unit corresponding to the real-time severity score that is greater than or equal to the severity threshold. If the detection is not interrupted, a comprehensive wear score is calculated. Several wear levels are preset. Based on the comprehensive wear score of the metal to be detected, the wear level is determined and uploaded to the system.

[0079] Compared with the prior art, the beneficial effects of the present invention are: by collecting the health score, environmental parameters and image score of metal in real time, the detection system can reflect the current health status of metal materials in an instant, and the real-time monitoring mechanism can identify potential problems in a timely manner, greatly reducing downtime and maintenance costs caused by equipment failure.

[0080] By combining default metal scores and image scores, this invention provides a comprehensive method for assessing metal health. The multi-parameter integrated scoring mechanism can effectively improve the accuracy of diagnosis and enhance the comprehensive understanding of metal wear.

[0081] When the wear severity score exceeds the set threshold, the system can automatically issue a high-risk defect warning and recommend shutdown for maintenance, which can effectively avoid safety accidents and improve the safety of the production process.

[0082] By comprehensively determining wear and tear scores and levels, a scientific basis is provided for equipment maintenance, which can optimize maintenance plans, allocate maintenance resources more efficiently to where they are most needed, and improve equipment operating efficiency. Attached Figure Description

[0083] Figure 1 This is a schematic flowchart of a machine vision-based method for locating wear on metal materials according to the present invention.

[0084] Figure 2 This is a schematic diagram of the structure of a metal material wear positioning system based on machine vision according to the present invention. Detailed Implementation

[0085] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0086] Please see Figure 1 and Figure 2 This invention provides a technical solution: a method for locating wear patterns in metal materials based on machine vision, the method comprising:

[0087] Step S100: Image acquisition is performed through image acquisition device, abnormal image areas are identified, and diagnostic records are generated. Based on historical diagnostic records, normalized information parameters and image parameters of the metal to be detected are extracted, and the default score and metal image score of the metal are calculated respectively. The default score and metal image score of the metal are combined to calculate the metal health score, and each health level and the corresponding range of metal health score are determined according to the metal health scores of all historical diagnostic records.

[0088] Step S100 includes:

[0089] Step S101: By arranging several environmental monitoring devices in the detection area, environmental parameters in the detection area are collected. The metal to be detected is placed in the detection platform. The metal to be detected is image-acquired by the image acquisition device, abnormal image areas are identified, and a diagnostic record is generated.

[0090] Step S102: Obtain historical diagnostic records, extract the information parameters of the metal to be detected from a specific historical diagnostic record, normalize the information parameters, and calculate the default score for the metal according to the following formula:

[0091] ;

[0092] Where A represents the default rating for metals, B a C is represented as the normalized value of the a-th information parameter. a Let b represent the weight of the a-th information parameter, and b represent the total number of information parameters.

[0093] Step S103: Obtain the surface image of the metal to be detected from the historical diagnostic records, collect the image parameters in the surface image, normalize the image parameters, and calculate the metal image score according to the following formula:

[0094] ;

[0095] Where D represents the metal image score, E d F is represented as the normalized value of the d-th image parameter. d Let represent the weight of the d-th image parameter, and e represent the total number of image parameters;

[0096] Step S104: Summarize the default metal score and metal image score of a historical diagnostic record, and calculate the metal health score according to the following formula:

[0097] ;

[0098] Where G represents the metal health score, α1 represents the weight of the default metal score, and α2 represents the weight of the metal image score.

[0099] Step S105: Summarize the metal health scores of all historical diagnostic records, calculate the average f and variance g of the metal health scores, preset k health levels, where k is a non-zero positive integer, and calculate the threshold H for the kth interval. k =f+k×g, determine the range of metal health scores corresponding to each health level as {[G min ,H1),[H2,H3),...,[H k G max ]}, where G min and G max H1, H2, and H3 represent the minimum and maximum values ​​of the metal health score, respectively, and represent the threshold values ​​of the 1st, 2nd, and 3rd intervals, respectively.

[0100] For example, the information parameters extracted from the a-th historical diagnostic record are: metal thickness 6mm, wear depth 1.5mm, surface roughness 1.0μm, and metal hardness HRB85.

[0101] Assume the maximum and minimum values ​​are respectively:

[0102] Thickness: Min=5mm, Max=10mm;

[0103] Wear depth: Min=0mm, Max=5mm;

[0104] Surface roughness: Min=0.5μm, Max=2.0μm;

[0105] Hardness: Min=60, Max=100;

[0106] The normalized values ​​obtained after calculation are metal thickness 0.2, wear depth 0.3, surface roughness 0.333, and metal hardness 0.625, with corresponding weights of 0.1, 0.4, 0.3, and 0.2, respectively.

[0107] The calculated default score for the metal is 0.365;

[0108] The image parameters extracted from the a-th historical diagnostic record are: image clarity 80%, background interference 10%, and number of scratches 5.

[0109] Assume the maximum and minimum values ​​of the image parameters are as follows:

[0110] Resolution range: Min=50%, Max=100%;

[0111] Background interference range: Min=0%, Max=50%;

[0112] Scratch count range: Min=0, Max=20;

[0113] The normalized values ​​obtained are: image clarity 0.6, background interference 0.2, and scratch count 0.25, with corresponding weights of 0.5, 0.3, and 0.2, respectively.

[0114] The calculated score for the metal image is 0.41;

[0115] The weight of the default metal score is set to 0.6, and the weight of the metal image score is set to 0.4. The calculated metal health score is 0.383.

[0116] For example, the metal health score of historical diagnosis record a is 0.383, the metal health score of historical diagnosis record b is 0.25, the metal health score of historical diagnosis record c is 0.30, the metal health score of historical diagnosis record d is 0.50, and the metal health score of historical diagnosis record f is 0.43. The calculated average is 0.3726, and the calculated standard deviation is 0.0894. Assuming the health level is 3 and the maximum value is 1, then the low health level is [0.25, 0.4620), the medium health level is [0.4620, 0.5514), and the high health level is [0.5514, 1.0].

[0117] Step S200: For historical diagnostic records of the same health level, combine the environmental parameters and imaging parameters of the historical diagnostic records, train and generate an imaging parameter model through a random forest model, and at the same time, screen historical diagnostic records that have not occurred, calculate the average value of the image feature parameters and define it as the standard image feature parameters, and screen historical diagnostic records that have occurred, calculate the wear index for each wear level, and generate a wear type list.

[0118] Step S200 includes:

[0119] Step S201: Summarize historical diagnostic records at the same health level, obtain environmental parameters and imaging parameters of the image acquisition device within the detection area of ​​the historical diagnostic records at a certain health level, normalize the environmental parameters and imaging parameters, use the environmental parameters and imaging parameters as training data pairs, summarize the training data pairs of all historical diagnostic records to generate a training dataset, adopt a random forest model, use the environmental parameters as the input features of the random forest model, use the imaging parameters as the output features of the random forest model, randomly divide the training dataset into several subsets, construct a decision tree based on each subset, the splitting of the decision tree is based on minimizing the mean square error, each decision tree independently learns the mapping relationship between environmental parameters and imaging parameters, after training, the final prediction output is obtained by averaging the prediction results of multiple decision trees, save the trained random forest model and define it as the imaging parameter model, and obtain the imaging parameter model corresponding to each health level;

[0120] Step S202: Filter historical diagnostic records that have not experienced any abnormalities, collect the image feature parameters of the historical diagnostic records, summarize each image feature parameter corresponding to all historical diagnostic records, calculate the average value of each image feature parameter, and define the average value as the standard image feature parameter;

[0121] Step S203: Filter historical diagnostic records that have anomalies, summarize historical diagnostic records with the same wear type, count the number of historical diagnostic records for a certain wear type that have been repaired and the equipment has returned to normal operation as z1, calculate the repair success rate Z as Z=z1 / z2, where z2 represents the historical diagnostic record of the wear type, and define the repair success rate as the degree of reversibility; collect the repair time of a certain historical diagnostic record in a certain wear type, summarize the repair time of all historical diagnostic records of the wear type, calculate the average repair time, and perform normalization processing, define the normalized average repair time as the degree of harm, perform weighted summation of reversibility and degree of harm to calculate the wear index of each wear level, sort the wear types according to the wear index from large to small, and generate a wear type list;

[0122] For example, in the low health level, the temperature of record 1 is 22, the humidity is 55, and the light intensity is 200; the temperature of record 2 is 21, the humidity is 60, and the light intensity is 190; the temperature of record 3 is 23, the humidity is 50, and the light intensity is 210.

[0123] Record 1: Exposure time: 1 / 60s, ISO: 400, Aperture: f / 2.8; Record 2: Exposure time: 1 / 50s, ISO: 500, Aperture: f / 2.8; Record 3: Exposure time: 1 / 70s, ISO: 350, Aperture: f / 2.8;

[0124] After normalization, the temperature of record 1 is 0.5, the humidity is 0.0, and the light intensity is 0.4; the temperature of record 2 is 0.0, the humidity is 0.2, and the light intensity is 0.0; the temperature of record 3 is 0.2, the humidity is 0.1, and the light intensity is 0.5.

[0125] Record 1: Exposure time: 0.4, ISO: 0.5, Aperture: 0.3; Record 2: Exposure time: 0.6, ISO: 0.8, Aperture: 0.3; Record 3: Exposure time: 0.6, ISO: 0.8, Aperture: 0.3. Used for training a random forest model.

[0126] Record 4, with a medium health level, was selected. Its image clarity was 90, background interference was 5, and the number of scratches was 0. Record 5 had an image clarity of 92, background interference was 3, and the number of scratches was 1. Record 6 had an image clarity of 89, background interference was 4, and the number of scratches was 2. The average value was calculated to obtain the image clarity of 90.33, background interference of 4, and number of scratches of 1 in the standard image feature parameters.

[0127] Assume that wear type A is scratch wear, wear type B is pitting wear, and wear type C is crack wear. The total number of historical diagnostic records for wear type A is 20, with 18 successful repairs, resulting in a repair success rate of 0.9%; the total number of historical diagnostic records for wear type B is 15, with 10 successful repairs, resulting in a repair success rate of 0.67%; and the total number of historical diagnostic records for wear type C is 12, with 6 successful repairs, resulting in a repair success rate of 0.5%.

[0128] The average repair time for wear type A is 4.5 hours, for wear type B it is 7 hours, and for wear type C it is 9.5 hours. The maximum average repair time is 10 hours and the minimum is 4 hours. The normalized average repair time for wear type A is 0.083, for wear type B it is 0.5 hours, and for wear type C it is 0.917 hours.

[0129] The wear index for wear type A was calculated to be 0.0915, the wear index for wear type B was 0.415, and the wear index for wear type C was 0.7085.

[0130] Cracks > Pitting > Scratches form a list of wear types.

[0131] Step S300: Set the training cycle, calculate the metal health score of the diagnostic record, determine the health level corresponding to the diagnostic record, collect the environmental parameters of the detection area, obtain the imaging parameters through the imaging parameter model corresponding to the health level, upload them to the image acquisition device, re-capture the secondary surface image, divide the secondary surface image into several grid units, calculate the deviation score of each grid unit, and determine the high-frequency unit corresponding to each wear type.

[0132] Step S300 includes:

[0133] Step S301: Select several consecutive days as the training period, obtain the diagnostic records within the training period, calculate the metal health score of each diagnostic record, determine the health level corresponding to the diagnostic record, collect the environmental parameters of the detection area and input them into the imaging parameter model, calculate the imaging parameters, upload the imaging parameters to the image acquisition device, and re-capture the metal to be detected to obtain a secondary surface image.

[0134] Step S302: Divide the secondary surface image into several grid cells, label and record each grid cell, collect the image feature parameters of each grid cell, and calculate the absolute difference between each image feature parameter and the standard image feature parameters to obtain the difference of each image feature parameter. Calculate the deviation score of each grid cell according to the following formula:

[0135] ;

[0136] Where P represents the deviation score, L g J is represented as the difference of the g-th image feature parameter. g This is represented as the weight of the g-th image feature parameter;

[0137] Step S303: Obtain the diagnostic records of the abnormality, extract the wear type and the corresponding grid cell in the diagnostic records, summarize the diagnostic records of the same wear type, count the number of times each grid cell appears, calculate the occurrence frequency of each grid cell, preset the occurrence frequency threshold, set the grid cell that exceeds the occurrence frequency threshold as a high frequency cell, determine the high frequency cell corresponding to each wear type, and summarize the deviation score of each high frequency cell in each diagnostic record.

[0138] For example, the health level of the first diagnostic record is medium, the health level of the second diagnostic record is medium, the health level of the third diagnostic record is high, the health level of the fourth diagnostic record is medium, and the health level of the fifth diagnostic record is high.

[0139] The temperature of the first diagnostic record is 22°C, the humidity is 55%, and the light intensity is 200%. The temperature of the second diagnostic record is 21°C, the humidity is 60%, and the light intensity is 190%. The temperature of the third diagnostic record is 23°C, the humidity is 50%, and the light intensity is 210%. The temperature of the fourth diagnostic record is 22°C, the humidity is 55%, and the light intensity is 210%. The temperature of the fifth diagnostic record is 24°C, the humidity is 52%, and the light intensity is 220%.

[0140] The exposure times for the following diagnostic records were obtained: 1 / 60s, ISO: 400, aperture: f / 2.8; 1 / 50s, ISO: 500, aperture: f / 2.8; 1 / 70s, ISO: 350, aperture: f / 2.8; 1 / 60s, ISO: 400, aperture: f / 2.8; and 1 / 80s, ISO: 300, aperture: f / 2.8.

[0141] Assume that the image clarity of grid cell 1 is 85, the background interference is 10, and the number of scratches is 2; the image clarity of grid cell 2 is 88, the background interference is 8, and the number of scratches is 1; and the image clarity of grid cell 3 is 83, the background interference is 12, and the number of scratches is 3.

[0142] The calculated deviation scores were 4.665 for grid cell 1, 2.365 for grid cell 2, and 6.465 for grid cell 3.

[0143] Step S400: Obtain the deviation score of the high-frequency unit corresponding to each wear type in a certain diagnostic record, and calculate the severity threshold of the wear type based on the deviation score;

[0144] Step S400 includes:

[0145] Step S401: In the set of diagnostic records where anomalies occur, collect the deviation score of the high-frequency unit corresponding to each wear type in a certain diagnostic record, summarize the deviation scores of all high-frequency units, calculate the average deviation score of each wear type in the diagnostic record, and define the average deviation score as the severity score.

[0146] Step S402: Summarize the severity scores of a certain wear type, calculate the average value u and standard deviation t of the severity scores, and calculate the severity threshold P' of the wear type as P'=u+v×t, where v represents a preset constant;

[0147] For example, in a certain diagnostic record, the high-frequency unit deviation scores for wear type A are 4.5, 3.2, 5.0, 6.0, 4.7, 3.8, 5.5, 2.9, 5.2, and 4.3, and the calculated average deviation score is 4.51.

[0148] Assuming the severity scores for wear type A are 4.51, 4.65, 4.40, 4.72, 4.55, and 4.80, the calculated average severity score is 4.6, the standard deviation is 0.134, and the preset constant is 2, then the severity threshold for wear type A is 4.868.

[0149] Step S500: Obtain real-time information parameters and calculate real-time metal health score. Collect real-time environmental parameters to calculate real-time imaging parameters, then re-image the metal surface. According to the wear type list, evaluate the real-time severity score of each wear type in turn. If the severity threshold is exceeded, the system will prompt that a high-risk defect has been detected and suggest stopping the machine for maintenance. If the severity threshold is not exceeded, calculate the comprehensive wear score to determine the wear level of the metal and update it to the system.

[0150] Step S500 includes:

[0151] Step S501: Obtain real-time information parameters of the metal to be detected, calculate the real-time default score of the metal, collect image parameters of the metal to be detected, calculate the real-time image score of the metal, calculate the real-time health score of the metal based on the real-time default score and the real-time image score of the metal, and determine the real-time health level corresponding to the metal to be detected.

[0152] Step S502: Collect real-time environmental parameters within the detection area and input them into the imaging parameter model corresponding to the real-time health level to calculate the real-time imaging parameters. Then, input the real-time imaging parameters into the image acquisition device to re-capture the metal to be detected and obtain a secondary real-time surface image.

[0153] Step S503: Based on the order of the wear type list, collect the high-frequency unit corresponding to a wear type in sequence, calculate the real-time severity score of the wear type, compare the real-time severity score with the severity threshold, if it is greater than or equal to the severity threshold, interrupt the detection and jump to step S504, if it is less than the severity threshold, continue to calculate the real-time severity score of the next wear type and compare it with the severity threshold.

[0154] Step S504: If an interruption occurs during detection, the system will indicate that a high-risk defect has been detected and recommend immediate shutdown for repair. The system will also specify the location of the high-frequency unit corresponding to a real-time severity score greater than or equal to the severity threshold. If no interruption occurs, the comprehensive wear score will be calculated using the following formula:

[0155] ;

[0156] Where Q represents the overall wear score, and X n Y represents the real-time severity score for the nth wear type. n The value represents the weight of the nth wear type, and m represents the total number of wear types. Several wear levels are preset. The wear level is determined based on the comprehensive wear score of the metal to be tested and uploaded to the system.

[0157] For example, with real-time information parameters of metal thickness 7mm, wear depth 2mm, surface roughness 1.1μm, and metal hardness HRB80, the calculated real-time metal default score is 0.44. With real-time image parameters of image clarity 80%, background interference 10%, and number of scratches 5, the calculated real-time metal image score is 0.41. The calculated real-time metal health score is 0.428, which is determined to be a low health level.

[0158] The real-time severity score of the crack is calculated to be 4.5, and the detection is interrupted to proceed to step S505;

[0159] It is recommended to shut down the machine immediately for inspection and repair. Defect location: Area A of the high-frequency unit.

[0160] To better implement the above method, a metal material wear localization system based on machine vision is also proposed. The system includes a health level module, an imaging analysis module, a training cycle module, a severity threshold module, and a real-time detection module.

[0161] Health level module: Image acquisition is performed through image acquisition device, abnormal image areas are identified and diagnostic records are generated. Based on historical diagnostic records, normalized information parameters and image parameters of the metal to be detected are extracted, and the default score and metal image score of the metal are calculated respectively. The metal health score is calculated by combining the default score and metal image score, and each health level and the corresponding metal health score range are determined according to the metal health scores of all historical diagnostic records.

[0162] The health rating module includes a metal health scoring unit and a health rating unit:

[0163] Metal Health Scoring Unit: By deploying several environmental monitoring devices in the detection area to collect environmental parameters within the detection area, the metal to be detected is placed in the detection platform. Image acquisition devices are used to acquire images of the metal to be detected, abnormal image areas are identified, diagnostic records are generated, historical diagnostic records are obtained, information parameters of the metal to be detected in a certain historical diagnostic record are extracted, and the information parameters are normalized to calculate the default metal score. Surface images of the metal to be detected collected in the historical diagnostic records are obtained, image parameters in the surface images are collected, and the image parameters are normalized to calculate the metal image score. The default metal score and metal image score of a certain historical diagnostic record are summarized to calculate the metal health score.

[0164] Health Level Unit: This unit summarizes the metal health scores from all historical diagnostic records, calculates the average f and variance g of the metal health scores, and predefines k health levels, where k is a non-zero positive integer. The threshold for the kth interval is calculated as H. k =f+k×g, determine the range of metal health scores corresponding to each health level as {[G min ,H1),[H2,H3),...,[H k G max ]}, where G min and G max H1, H2, and H3 represent the minimum and maximum values ​​of the metal health score, respectively, and represent the threshold values ​​of the 1st, 2nd, and 3rd intervals, respectively.

[0165] Imaging analysis module: For historical diagnostic records of the same health level, the environmental parameters and imaging parameters of the historical diagnostic records are combined, and an imaging parameter model is generated by training through a random forest model. At the same time, historical diagnostic records without abnormalities are screened, the average value of the image feature parameters is calculated and defined as the standard image feature parameters, and historical diagnostic records with abnormalities are screened, the wear index of each wear level is calculated, and a list of wear types is generated.

[0166] Training cycle module: Set the training cycle, calculate the metal health score of the diagnostic record, determine the health level corresponding to the diagnostic record, collect environmental parameters of the detection area, obtain imaging parameters through the imaging parameter model corresponding to the health level, upload them to the image acquisition device, re-capture the secondary surface image, divide the secondary surface image into several grid units, calculate the deviation score of each grid unit, and determine the high-frequency unit corresponding to each wear type.

[0167] The training cycle module includes a secondary shooting unit and a high-frequency determination unit.

[0168] Secondary imaging unit: Select several consecutive days as the training period, acquire diagnostic records within the training period, calculate the metal health score of each diagnostic record, determine the health level corresponding to the diagnostic record, collect environmental parameters of the detection area and input them into the imaging parameter model, calculate the imaging parameters, upload the imaging parameters to the image acquisition device, and re-image the metal to be detected to obtain a secondary surface image.

[0169] Determine high-frequency units: Divide the secondary surface image into several grid units, label and record each grid unit, collect the image feature parameters of each grid unit, and calculate the absolute difference between each image feature parameter and the standard image feature parameters to obtain the difference of each image feature parameter. Calculate the deviation score, obtain the diagnostic records of abnormalities, extract the wear types and the corresponding grid units in the diagnostic records, summarize the diagnostic records of the same wear type, count the number of times each grid unit appears, calculate the occurrence frequency of each grid unit, preset the occurrence frequency threshold, and set the grid units that exceed the occurrence frequency threshold as high-frequency units to determine the high-frequency units corresponding to each wear type.

[0170] Severity threshold module: Obtain the deviation score of the high-frequency unit corresponding to each wear type in a certain diagnostic record, and calculate the severity threshold of the wear type based on the deviation score;

[0171] Real-time detection module: Acquires real-time information parameters and calculates real-time metal health score, collects real-time environmental parameters to calculate real-time imaging parameters, then re-images the metal surface, and evaluates the real-time severity score of each wear type according to the wear type list. If the severity threshold is exceeded, the system will prompt that a high-risk defect has been detected and suggest shutdown for maintenance; if the severity threshold is not exceeded, the comprehensive wear score is calculated to determine the wear level of the metal and updated to the system.

[0172] The real-time detection module includes a real-time acquisition unit and a wear analysis unit.

[0173] Real-time acquisition unit: acquires real-time information parameters of the metal to be detected, calculates the real-time default score of the metal, acquires image parameters of the metal to be detected, calculates the real-time image score of the metal, calculates the real-time health score of the metal based on the real-time default score and the real-time image score of the metal, determines the real-time health level corresponding to the metal to be detected, acquires real-time environmental parameters in the detection area, and inputs them into the imaging parameter model corresponding to the real-time health level, calculates the real-time imaging parameters, and inputs the real-time imaging parameters into the image acquisition device to re-capture the metal to be detected and obtain a secondary real-time surface image;

[0174] Wear unit analysis: Based on the order of the wear type list, high-frequency units corresponding to one wear type are collected sequentially, and the real-time severity score of the wear type is calculated. The real-time severity score is compared with the severity threshold. If it is greater than or equal to the severity threshold, the detection is interrupted and the process jumps to step S504. If it is less than the severity threshold, the real-time severity score of the next wear type is calculated and compared with the severity threshold. If the detection is interrupted, the system prompts that a high-risk defect has been detected, recommends immediate shutdown for maintenance, and indicates the location of the high-frequency unit corresponding to the real-time severity score that is greater than or equal to the severity threshold. If the detection is not interrupted, a comprehensive wear score is calculated. Several wear levels are preset. Based on the comprehensive wear score of the metal to be detected, the wear level is determined and uploaded to the system.

[0175] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for locating wear patterns in metal materials based on machine vision, characterized in that, The methods include: Step S100: Image acquisition is performed through image acquisition device, abnormal image areas are identified, and diagnostic records are generated. Based on historical diagnostic records, normalized information parameters and image parameters of the metal to be detected are extracted, and the default score and metal image score of the metal are calculated respectively. The default score and metal image score of the metal are combined to calculate the metal health score, and each health level and the corresponding range of metal health score are determined according to the metal health scores of all historical diagnostic records. Step S200: For historical diagnostic records of the same health level, combine the environmental parameters and imaging parameters of the historical diagnostic records, train and generate an imaging parameter model through a random forest model, and at the same time, screen historical diagnostic records that have not occurred, calculate the average value of the image feature parameters and define it as the standard image feature parameters, and screen historical diagnostic records that have occurred, calculate the wear index for each wear level, and generate a wear type list. Step S300: Set the training cycle, calculate the metal health score of the diagnostic record, determine the health level corresponding to the diagnostic record, collect the environmental parameters of the detection area, obtain the imaging parameters through the imaging parameter model corresponding to the health level, upload them to the image acquisition device, re-capture the secondary surface image, divide the secondary surface image into several grid units, calculate the deviation score of each grid unit, and determine the high-frequency unit corresponding to each wear type. Step S300 includes the following steps: Step S301: Select several consecutive days as the training period, obtain the diagnostic records within the training period, calculate the metal health score of each diagnostic record, determine the health level corresponding to the diagnostic record, collect the environmental parameters of the detection area and input them into the imaging parameter model, calculate the imaging parameters, upload the imaging parameters to the image acquisition device, and re-capture the metal to be detected to obtain a secondary surface image. Step S302: Divide the secondary surface image into several grid cells, label and record each grid cell, collect the image feature parameters of each grid cell, and calculate the absolute difference between each image feature parameter and the standard image feature parameters to obtain the difference of each image feature parameter. Calculate the deviation score of each grid cell according to the following formula: ； Where P represents the deviation score, L g J is represented as the difference of the g-th image feature parameter. g This is represented as the weight of the g-th image feature parameter; Step S303: Obtain the diagnostic records of the abnormality, extract the wear type and the corresponding grid cell in the diagnostic records, summarize the diagnostic records of the same wear type, count the number of times each grid cell appears, calculate the occurrence frequency of each grid cell, preset the occurrence frequency threshold, set the grid cell that exceeds the occurrence frequency threshold as a high frequency cell, determine the high frequency cell corresponding to each wear type, and summarize the deviation score of each high frequency cell in each diagnostic record. Step S400: Obtain the deviation score of the high-frequency unit corresponding to each wear type in a certain diagnostic record, and calculate the severity threshold of the wear type based on the deviation score; Step S400 includes the following steps: Step S401: In the set of diagnostic records where anomalies occur, collect the deviation score of the high-frequency unit corresponding to each wear type in a certain diagnostic record, summarize the deviation scores of all high-frequency units, calculate the average deviation score of each wear type in the diagnostic record, and define the average deviation score as the severity score. Step S402: Summarize the severity scores of a certain wear type, calculate the average value u and standard deviation t of the severity scores, and calculate the severity threshold P' of the wear type as P'=u+v×t, where v represents a preset constant; Step S500: Obtain real-time information parameters and calculate real-time metal health score. Collect real-time environmental parameters to calculate real-time imaging parameters, then re-photograph the metal surface. According to the wear type list, evaluate the real-time severity score of each wear type in turn. If the severity threshold is exceeded, the system will prompt that a high-risk defect has been detected and suggest stopping the machine for maintenance. If the severity threshold is not exceeded, calculate the comprehensive wear score to determine the wear level of the metal and update it to the system.

2. The method for locating wear on metal materials based on machine vision according to claim 1, characterized in that, Step S100 includes the following steps: Step S101: By arranging several environmental monitoring devices in the detection area, environmental parameters in the detection area are collected. The metal to be detected is placed in the detection platform. The metal to be detected is image-acquired by the image acquisition device, abnormal image areas are identified, and a diagnostic record is generated. Step S102: Obtain historical diagnostic records, extract the information parameters of the metal to be detected from a specific historical diagnostic record, normalize the information parameters, and calculate the default score for the metal according to the following formula: ； Where A represents the default rating for metals, B a C is represented as the normalized value of the a-th information parameter. a Let b represent the weight of the a-th information parameter, and b represent the total number of information parameters. Step S103: Obtain the surface image of the metal to be detected from the historical diagnostic records, collect the image parameters in the surface image, normalize the image parameters, and calculate the metal image score according to the following formula: ； Where D represents the metal image score, E d F is represented as the normalized value of the d-th image parameter. d Let represent the weight of the d-th image parameter, and e represent the total number of image parameters; Step S104: Summarize the default metal score and metal image score of a historical diagnostic record, and calculate the metal health score according to the following formula: ； Where G represents the metal health score, α1 represents the weight of the default metal score, and α2 represents the weight of the metal image score. Step S105: Summarize the metal health scores of all historical diagnostic records, calculate the average f and variance g of the metal health scores, preset k health levels, where k is a non-zero positive integer, and calculate the threshold H for the kth interval. k =f+k×g, determine the range of metal health scores corresponding to each health level as {[G min ,H1),[H2,H3),...,[H k G max ]}, where G min and G max H1, H2, and H3 represent the minimum and maximum values ​​of the metal health score, respectively, and represent the threshold values ​​of the 1st, 2nd, and 3rd intervals, respectively.

3. The method for locating wear on metal materials based on machine vision according to claim 2, characterized in that, Step S200 includes the following steps: Step S201: Summarize historical diagnostic records at the same health level, obtain environmental parameters and imaging parameters of the image acquisition device within the detection area of ​​the historical diagnostic records at a certain health level, normalize the environmental parameters and imaging parameters, use the environmental parameters and imaging parameters as training data pairs, summarize the training data pairs of all historical diagnostic records to generate a training dataset, adopt a random forest model, use the environmental parameters as the input features of the random forest model, use the imaging parameters as the output features of the random forest model, randomly divide the training dataset into several subsets, construct a decision tree based on each subset, the splitting of the decision tree is based on minimizing the mean square error, each decision tree independently learns the mapping relationship between environmental parameters and imaging parameters, after training, the final prediction output is obtained by averaging the prediction results of multiple decision trees, save the trained random forest model and define it as the imaging parameter model, and obtain the imaging parameter model corresponding to each health level; Step S202: Filter historical diagnostic records that have not experienced any abnormalities, collect the image feature parameters of the historical diagnostic records, summarize each image feature parameter corresponding to all historical diagnostic records, calculate the average value of each image feature parameter, and define the average value as the standard image feature parameter; Step S203: Filter historical diagnostic records that have anomalies, summarize historical diagnostic records with the same wear type, count the number of historical diagnostic records for a certain wear type that have been repaired and the equipment has returned to normal operation as z1, calculate the repair success rate Z as Z=z1 / z2, where z2 represents the historical diagnostic record of the wear type, and define the repair success rate as the degree of reversibility; collect the repair time of a certain historical diagnostic record in a certain wear type, summarize the repair time of all historical diagnostic records of the wear type, calculate the average repair time, define the average repair time as the degree of harm, calculate the wear index of each wear level by weighted summation of the degree of reversibility and the degree of harm, sort the wear types according to the wear index from large to small, and generate a wear type list.

4. The method for locating wear on metal materials based on machine vision according to claim 1, characterized in that, Step S500 includes the following steps: Step S501: Obtain real-time information parameters of the metal to be detected, calculate the real-time default score of the metal, collect image parameters of the metal to be detected, calculate the real-time image score of the metal, calculate the real-time health score of the metal based on the real-time default score and the real-time image score of the metal, and determine the real-time health level corresponding to the metal to be detected. Step S502: Collect real-time environmental parameters within the detection area and input them into the imaging parameter model corresponding to the real-time health level to calculate the real-time imaging parameters. Then, input the real-time imaging parameters into the image acquisition device to re-capture the metal to be detected and obtain a secondary real-time surface image. Step S503: Based on the order of the wear type list, collect the high-frequency unit corresponding to a wear type in sequence, calculate the real-time severity score of the wear type, compare the real-time severity score with the severity threshold, if it is greater than or equal to the severity threshold, interrupt the detection and jump to step S504, if it is less than the severity threshold, continue to calculate the real-time severity score of the next wear type and compare it with the severity threshold. Step S504: If an interruption occurs during detection, the system will indicate that a high-risk defect has been detected and recommend immediate shutdown for repair. The system will also specify the location of the high-frequency unit corresponding to a real-time severity score greater than or equal to the severity threshold. If no interruption occurs, the comprehensive wear score will be calculated using the following formula: ； Where Q represents the overall wear score, and X n Y represents the real-time severity score for the nth wear type. n The value represents the weight of the nth wear type, and m represents the total number of wear types. Several wear levels are preset. The wear level is determined based on the comprehensive wear score of the metal to be tested and uploaded to the system.

5. A machine vision-based metal material wear positioning system, used to implement the machine vision-based metal material wear positioning method according to any one of claims 1-4, characterized in that, The system includes a health level module, an imaging analysis module, a training cycle module, a severity threshold module, and a real-time detection module. The health level module: acquires images through an image acquisition device, identifies abnormal image areas, and generates diagnostic records. Based on historical diagnostic records, it extracts normalized information parameters and image parameters of the metal to be detected, calculates the default metal score and the metal image score respectively, combines the default metal score and the metal image score to calculate the metal health score, and determines each health level and the corresponding metal health score range based on the metal health scores of all historical diagnostic records. The imaging analysis module: For historical diagnostic records of the same health level, it combines the environmental parameters and imaging parameters of the historical diagnostic records, trains and generates an imaging parameter model through a random forest model, and at the same time, filters historical diagnostic records without abnormalities, calculates the average value of image feature parameters and defines them as standard image feature parameters, filters historical diagnostic records with abnormalities, calculates the wear index for each wear level, and generates a list of wear types. The training cycle module: sets the training cycle, calculates the metal health score of the diagnostic record, determines the health level corresponding to the diagnostic record, collects environmental parameters of the detection area, obtains imaging parameters through the imaging parameter model corresponding to the health level, uploads them to the image acquisition device, re-captures a secondary surface image, divides the secondary surface image into several grid units, calculates the deviation score of each grid unit, and determines the high-frequency unit corresponding to each wear type. The severity threshold module: obtains the deviation score of the high-frequency unit corresponding to each wear type in a certain diagnostic record, and calculates the severity threshold of the wear type based on the deviation score; The real-time detection module acquires real-time information parameters and calculates real-time metal health scores, collects real-time environmental parameters to calculate real-time imaging parameters, and then re-images the metal surface. Based on the wear type list, it sequentially evaluates the real-time severity scores of each wear type. If the severity threshold is exceeded, the system will indicate that a high-risk defect has been detected and suggest stopping the machine for maintenance. If the severity threshold is not exceeded, it calculates a comprehensive wear score to determine the wear level of the metal and updates it to the system.

6. A machine vision-based metal material wear positioning system according to claim 5, characterized in that, The health rating module includes a metal health scoring unit and a health rating unit: The metal health scoring unit works as follows: by deploying several environmental monitoring devices in the detection area to collect environmental parameters within the detection area, placing the metal to be detected in the detection platform, acquiring images of the metal to be detected through image acquisition devices, identifying abnormal image areas, generating diagnostic records, obtaining historical diagnostic records, extracting information parameters of the metal to be detected from a certain historical diagnostic record, normalizing the information parameters, calculating the default metal score, acquiring surface images of the metal to be detected collected in the historical diagnostic records, collecting image parameters from the surface images, normalizing the image parameters, calculating the metal image score, summarizing the default metal score and the metal image score of a certain historical diagnostic record, and calculating the metal health score. The health level unit: It summarizes the metal health scores from all historical diagnostic records, calculates the average f and variance g of the metal health scores, presets k health levels, where k is a non-zero positive integer, and calculates the threshold H for the kth interval. k =f+k×g, determine the range of metal health scores corresponding to each health level as {[G min ,H1),[H2,H3),...,[H k G max ]}, where G min and G max H1, H2, and H3 represent the minimum and maximum values ​​of the metal health score, respectively, and represent the threshold values ​​of the 1st, 2nd, and 3rd intervals, respectively.

7. A machine vision-based metal material wear positioning system according to claim 5, characterized in that, The training cycle module includes a secondary imaging unit and a high-frequency determination unit: The secondary imaging unit: selects several consecutive days as the training period, acquires diagnostic records within the training period, calculates the metal health score of each diagnostic record, determines the health level corresponding to the diagnostic record, collects environmental parameters of the detection area and inputs them into the imaging parameter model, calculates the imaging parameters, uploads the imaging parameters to the image acquisition device, and re-images the metal to be detected to obtain a secondary surface image. The process of determining high-frequency units involves: dividing the secondary surface image into several grid units, labeling and recording each grid unit, collecting image feature parameters for each grid unit, calculating the absolute difference between each image feature parameter and the standard image feature parameters, obtaining the difference for each image feature parameter, calculating the deviation score, obtaining diagnostic records of abnormalities, extracting the wear types and corresponding grid units from the diagnostic records, summarizing diagnostic records of the same wear type, counting the number of times each grid unit appears, calculating the occurrence frequency of each grid unit, setting a preset occurrence frequency threshold, and designating grid units exceeding the occurrence frequency threshold as high-frequency units, thus determining the high-frequency units corresponding to each wear type.

8. A machine vision-based metal material wear positioning system according to claim 5, characterized in that, The real-time detection module includes a real-time acquisition unit and a wear analysis unit: The real-time acquisition unit acquires real-time information parameters of the metal to be detected, calculates a real-time default score for the metal, acquires image parameters of the metal to be detected, calculates a real-time image score for the metal, calculates a real-time health score for the metal based on the real-time default score and the real-time image score, determines the real-time health level corresponding to the metal to be detected, acquires real-time environmental parameters in the detection area, inputs them into the imaging parameter model corresponding to the real-time health level, calculates real-time imaging parameters, and inputs the real-time imaging parameters into the image acquisition device to re-capture the metal to be detected, thereby obtaining a secondary real-time surface image. The wear analysis unit: Based on the order of the wear type list, it sequentially collects high-frequency units corresponding to one wear type and calculates the real-time severity score of the wear type. The real-time severity score is compared with a severity threshold. If it is greater than or equal to the severity threshold, the detection is interrupted and the process jumps to step S504. If it is less than the severity threshold, the real-time severity score of the next wear type is calculated and compared with the severity threshold. If the detection is interrupted, the system prompts that a high-risk defect has been detected, recommends immediate shutdown for maintenance, and indicates the location of the high-frequency unit corresponding to the real-time severity score that is greater than or equal to the severity threshold. If the detection is not interrupted, a comprehensive wear score is calculated. Several wear levels are preset. Based on the comprehensive wear score of the metal to be detected, the wear level is determined and uploaded to the system.

Images