A continuous casting billet defect identification method and device, electronic equipment and storage medium

Abstract

The application relates to the technical field of steel and the technical field of defect identification, and discloses a continuous casting billet defect identification method and device, electronic equipment and a storage medium, the method comprising the following steps: a preset feature and a real defect category of a preset continuous casting billet are combined to form a sample; different samples are combined to form a training set, and a multi-modal identification model is trained based on the training set; an anchor point loss of the multi-modal identification model under the training set is generated through an anchor point loss model, a comprehensive loss value of the multi-modal identification model under the training set is generated according to the anchor point loss of the multi-modal identification model under the training set and a predefined comprehensive loss model; when the comprehensive loss value is smaller than a preset loss value, the training of the multi-modal identification model is stopped, and the trained multi-modal identification model is saved; and the current feature is identified through the trained multi-modal identification model to generate the defect category to which the current continuous casting billet belongs. The application is beneficial to the identification efficiency of the defect category to which the current continuous casting billet belongs.

Description

Technical Field

[0001] This application relates to the fields of steel technology and defect identification technology, and in particular to a method, apparatus, electronic device and storage medium for identifying defects in continuously cast billets. Background Technology

[0002] Continuously cast billets are steel billets produced by the continuous casting process. After being processed through rolling, forging, heat treatment, and other processes, continuously cast billets can be made into various types and specifications of steel products.

[0003] However, in the current continuous casting billet production process, defect identification is mainly done manually. Manual identification is time-consuming and labor-intensive, and easily affected by subjective factors, as different personnel have different standards for judging defects, leading to poor accuracy and consistency in the identification results. Therefore, how to identify the defect category of the current continuous casting billet is a technical problem that urgently needs to be solved. Summary of the Invention

[0004] This application provides a method, apparatus, electronic device, and storage medium for identifying defects in continuously cast billets, in order to solve the aforementioned technical problem of how to identify the defect category of the current continuously cast billet.

[0005] In a first aspect, embodiments of this application provide a method for identifying defects in continuously cast billets, applied to electronic devices, the method comprising: Combine the preset features and the preset actual defect categories of the continuous casting billet into a sample; Different samples are combined into a training set, and a multimodal recognition model is trained based on the training set; Anchor point loss of the multimodal recognition model under the training set is generated using a predefined anchor point loss model. Based on the anchor point loss of the multimodal recognition model under the training set and the predefined comprehensive loss model, comprehensive loss value of the multimodal recognition model under the training set is generated. When the overall loss value is less than the preset loss value, stop training the multimodal recognition model and save the trained multimodal recognition model. Based on the current multimodal data of the continuously cast billet and the predefined generation method, the current features are determined, and the current features are identified by the trained multimodal recognition model to generate the defect category of the current continuously cast billet.

[0006] In one possible implementation of the first aspect, In one possible implementation of the first aspect, the step of combining a preset feature and a preset true defect category of the continuously cast billet into a sample includes: Acquire multimodal data of a preset continuous casting billet, which includes sensor data of the preset continuous casting billet, defect images of the preset continuous casting billet, and log data of the preset continuous casting billet; Feature extraction is performed on the sensor data of the preset continuous casting billet to obtain the features of the sensor data of the continuous casting billet; feature extraction is performed on the defect image of the preset continuous casting billet to obtain the features of the defect image of the preset continuous casting billet; feature extraction is performed on the log data of the preset continuous casting billet to obtain the features of the log data of the preset continuous casting billet. The features of the sensor data of the preset continuous casting billet, the features of the defect image of the preset continuous casting billet, and the features of the log data of the preset continuous casting billet are fused to obtain the preset features. The preset features and the actual defect categories of the preset continuous casting billet are combined to form a sample.

[0007] In one possible implementation of the first aspect, generating the anchor loss of the multimodal recognition model on the training set using a predefined anchor loss model, and generating the comprehensive loss value of the multimodal recognition model on the training set based on the anchor loss of the multimodal recognition model on the training set and a predefined comprehensive loss model, includes: The consistency loss of the multimodal recognition model under the training set is obtained through the first model; the constraint loss of the multimodal recognition model under the training set is obtained through the second model; and the separation loss of the multimodal recognition model under the training set is obtained through the third model. Based on the consistency loss, constraint loss, separation loss, and predefined anchor loss model of the multimodal recognition model under the training set, the anchor loss of the multimodal recognition model under the training set is generated. The classification loss function is used to obtain the classification loss of the multimodal recognition model under the training set. The supervised contrastive loss function is used to obtain the supervised contrastive loss of the multimodal recognition model under the training set. Based on the anchor loss, classification loss, supervised contrastive loss and predefined comprehensive loss model of the multimodal recognition model under the training set, the comprehensive loss value of the multimodal recognition model under the training set is generated.

[0008] In one possible implementation of the first aspect, determining the current feature based on the multimodal data of the current continuously cast billet and a predefined determination method, identifying the current feature through a trained multimodal recognition model, and generating the defect category to which the current continuously cast billet belongs includes: Acquire the multimodal data of the current continuous casting billet, which includes the sensor data of the current continuous casting billet, the defect image of the current continuous casting billet, and the log data of the current continuous casting billet; Feature extraction is performed on the sensor data of the current continuous casting billet to obtain the features of the sensor data of the continuous casting billet; feature extraction is performed on the defect image of the current continuous casting billet to obtain the features of the defect image of the current continuous casting billet; feature extraction is performed on the log data of the current continuous casting billet to obtain the features of the log data of the current continuous casting billet. The features of the sensor data of the current continuous casting billet, the features of the defect image of the current continuous casting billet, and the features of the log data of the current continuous casting billet are fused to obtain the current features. The current features are then identified by the trained multimodal recognition model to generate the defect category of the current continuous casting billet.

[0009] In one possible implementation of the first aspect, the comprehensive loss model is defined as follows: ; This represents the overall loss value of the multimodal recognition model on the training set. The higher the overall loss value of the multimodal recognition model on the training set, the stronger the recognition ability of the multimodal recognition model on the training set. The lower the overall loss value of the multimodal recognition model on the training set, the weaker the recognition ability of the multimodal recognition model on the training set. This is the first adjustment parameter; This is the second adjustment parameter. , ; This represents the classification loss of the multimodal recognition model on the training set; This represents the anchor loss of the multimodal recognition model on the training set; This represents the supervised contrastive loss of the multimodal recognition model on the training set.

[0010] In one possible implementation of the first aspect, the anchor loss model is defined as follows: ; This represents the anchor loss of the multimodal recognition model on the training set. The larger the anchor loss of the multimodal recognition model on the training set, the worse its ability to distinguish between categories. The smaller the anchor point loss of a multimodal recognition model in the training set, the stronger the model's ability to distinguish between category anchor points. This represents the consistency loss of the multimodal recognition model on the training set. This represents the constrained loss of the multimodal recognition model on the training set; This represents the separation loss of the multimodal recognition model on the training set; This represents the first weighting coefficient. This represents the second weighting coefficient. This represents the third weighting coefficient.

[0011] In one possible implementation of the first aspect, the first model is defined as follows: ; This represents the consistency loss of the multimodal recognition model on the training set. The smaller the consistency loss of the multimodal recognition model on the training set, the more aggregated the sample features of the same real defect category on the training set. The larger the consistency loss of the multimodal recognition model on the training set, the less aggregated the sample features of the same real defect category on the training set. c represents the sequence number of the actual defect category; This represents the total number of actual defect categories; This represents the total number of samples for the c-th true defect category; This represents the set of sample indices for the c-th true defect category; Indicates that the nth sample belongs to This represents the feature vector of the nth sample belonging to the cth true defect category; This represents the central feature vector of the c-th true defect category; express and The square of the Euclidean distance between them; The second model is defined as follows: ; This represents the constraint loss of the multimodal recognition model under the training set. The smaller the constraint loss of the multimodal recognition model under the training set, the closer the class center of the multimodal recognition model is to the initial benchmark under the training set. The larger the constraint loss of the multimodal recognition model under the training set, the less the class center of the multimodal recognition model is to the initial benchmark under the training set. c represents the sequence number of the actual defect category; This represents the total number of actual defect categories; This represents the control coefficient for the c-th true defect category; This represents the central feature vector of the c-th true defect category; This represents the initial feature vector of the c-th true defect category; The third model is defined as follows: ; This represents the separation loss of the multimodal recognition model on the training set. The greater the separation loss of the multimodal recognition model on the training set, the weaker its ability to distinguish the sample features of the real defect categories on the training set; the smaller the separation loss of the multimodal recognition model on the training set, the stronger its ability to distinguish the sample features of the real defect categories on the training set. c represents the sequence number of the actual defect category; Indicates the index of the reference cluster; This represents the central feature vector of the c-th true defect category; Indicates the first The central eigenvectors of each reference cluster; This represents the temperature coefficient.

[0012] Secondly, embodiments of this application provide a continuous casting billet defect identification device, applied to electronic equipment, comprising: The acquisition module is used to combine preset features and preset real defect categories of continuous casting billets into a sample; The component module is used to combine different samples into a training set, and to train a multimodal recognition model based on the training set; The generation module is used to generate the anchor loss of the multimodal recognition model under the training set through a predefined anchor loss model, and to generate the comprehensive loss value of the multimodal recognition model under the training set based on the anchor loss of the multimodal recognition model under the training set and a predefined comprehensive loss model. The save module is used to stop training the multimodal recognition model and save the trained multimodal recognition model when the comprehensive loss value is less than the preset loss value. The identification module is used to determine the current features based on the multimodal data of the current continuous casting billet and the predefined generation method. It then identifies the current features through the trained multimodal identification model and generates the defect category to which the current continuous casting billet belongs.

[0013] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the continuous casting billet defect identification method described in the first aspect above.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the continuous casting billet defect identification method described in the first aspect above.

[0015] Fifthly, embodiments of this application provide a computer program product that, when run on an electronic device, causes the electronic device to execute the continuous casting billet defect identification method described in the first aspect.

[0016] The beneficial effects of the embodiments of this application are as follows: Firstly, based on the current multimodal data of the continuous casting billet and the predefined generation method, the current features are determined, and the current features are identified by the trained multimodal recognition model to generate the defect category of the current continuous casting billet. Since there is no need for manual identification of the defect category of the current continuous casting billet, the identification time of identifying the defect category of the current continuous casting billet can be reduced, which is conducive to improving the identification efficiency of the defect category of the current continuous casting billet. Secondly, since the trained multimodal recognition model is not affected by human subjective factors, it is beneficial to improve the reliability of the defect category of the current continuous casting billet. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is an application scenario diagram of the continuous casting billet defect identification method provided in the embodiments of this application; Figure 2 This is a schematic flowchart of the continuous casting billet defect identification method provided in the embodiments of this application; Figure 3 A flowchart showing the results of list processing is provided for embodiments of this application; Figure 4 A schematic block diagram of the continuous casting billet defect identification device provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0020] The continuous casting billet defect identification method provided in this application embodiment can be applied to electronic devices such as servers, mobile phones, tablets, wearable devices, vehicle-mounted devices, augmented reality (AR) / virtual reality (VR) devices, laptops, and netbooks. This application embodiment does not impose any restrictions on the specific type of electronic device.

[0021] Please see Figure 1 , Figure 1 The application scenario diagram of the continuous casting billet defect identification method provided in the embodiments of this application is described in detail below: The electronic device sends a data query request for a preset continuous casting billet to the database, receives the query results returned by the database based on the data query request, and obtains the multimodal data of the preset continuous casting billet from the query results.

[0022] In this embodiment, the database integrates various data from the entire production process of the pre-cast billet. The electronic device obtains the multi-modal data of the pre-cast billet from the database, eliminating the need for manual searching through massive amounts of paper documents or distributed systems. This greatly shortens the acquisition time of the multi-modal data of the pre-cast billet and helps improve the acquisition efficiency of the multi-modal data of the pre-cast billet.

[0023] Please see Figure 2 , Figure 2 This is a flowchart illustrating the continuous casting billet defect identification method provided in this application embodiment, which can be applied to electronic devices.

[0024] like Figure 2 As shown in the embodiments of this application, the method for identifying defects in continuously cast billets includes the following steps, which are detailed below: S201, combine the preset features and the preset real defect category of the continuous casting billet into a sample; The step of combining preset features and preset actual defect categories of continuously cast billets into a sample includes: Acquire multimodal data of a preset continuous casting billet, which includes sensor data of the preset continuous casting billet, defect images of the preset continuous casting billet, and log data of the preset continuous casting billet; Feature extraction is performed on the sensor data of the preset continuous casting billet to obtain the features of the sensor data of the continuous casting billet; feature extraction is performed on the defect image of the preset continuous casting billet to obtain the features of the defect image of the preset continuous casting billet; feature extraction is performed on the log data of the preset continuous casting billet to obtain the features of the log data of the preset continuous casting billet. The features of the sensor data of the preset continuous casting billet, the features of the defect image of the preset continuous casting billet, and the features of the log data of the preset continuous casting billet are fused to obtain the preset features. The preset features and the actual defect categories of the preset continuous casting billet are combined to form a sample.

[0025] For example, multimodal data of a preset continuous casting billet is acquired. This multimodal data includes sensor data of the preset continuous casting billet, defect images of the preset continuous casting billet, and log data of the preset continuous casting billet, including: The electronic device sends a data query request for a preset continuous casting billet to the database, receives the query results returned by the database based on the data query request, and obtains the multimodal data of the preset continuous casting billet from the query results. The multimodal data of the preset continuous casting billet includes sensor data of the preset continuous casting billet, defect images of the preset continuous casting billet, and log data of the preset continuous casting billet.

[0026] The sensor data for the pre-cast billet refers to the data obtained in real time by various industrial sensors deployed on the continuous casting production line, which collect the physical state and environmental parameters of the pre-cast billet throughout the entire production process.

[0027] The sensor data for the pre-cast continuous casting billet includes: surface temperature data of the pre-cast continuous casting billet collected by an infrared temperature sensor, and acoustic wave reflection signals of the internal structure of the pre-cast continuous casting billet collected by an ultrasonic sensor.

[0028] Preset defect images of continuous casting billets refer to image data acquired by industrial vision acquisition equipment that can intuitively present the surface and near-surface defect features of continuous casting billets, and are the core visual basis for defect category determination.

[0029] The defect images of the pre-cast continuous casting billet include two-dimensional images and infrared thermal images of the pre-cast continuous casting billet. The two-dimensional images of the pre-cast continuous casting billet can clearly show the appearance features such as the direction of cracks, the morphology of scales, and the distribution of roll marks. The infrared thermal images of the pre-cast continuous casting billet can show the temperature difference between the defect area and the normal area of ​​the pre-cast continuous casting billet, which is convenient for identifying hidden defects such as internal cracks in the pre-cast continuous casting billet.

[0030] The log data for pre-cast billets refers to various data and information related to pre-cast billets automatically recorded by the control system on the continuous casting production line during operation. The log data includes the operating status log of the pre-cast billet production equipment and the quality inspection log of the pre-cast billets.

[0031] S202, different samples are combined into a training set, and a multimodal recognition model is trained based on the training set; A multimodal recognition model is an artificial intelligence model that integrates two or more different types of data sources and achieves target recognition tasks through collaborative analysis and feature extraction. The core of the multimodal recognition model lies in breaking the information limitations of single-modal data and improving the accuracy and robustness of recognition by utilizing the complementarity of multi-source data.

[0032] S203, using a predefined anchor loss model, generates the anchor loss of the multimodal recognition model under the training set, and based on the anchor loss of the multimodal recognition model under the training set and the predefined comprehensive loss model, generates the comprehensive loss value of the multimodal recognition model under the training set. The process of generating the anchor loss of the multimodal recognition model on the training set using a predefined anchor loss model, and generating the comprehensive loss value of the multimodal recognition model on the training set based on the anchor loss of the multimodal recognition model on the training set and a predefined comprehensive loss model, includes: The consistency loss of the multimodal recognition model under the training set is obtained through the first model; the constraint loss of the multimodal recognition model under the training set is obtained through the second model; and the separation loss of the multimodal recognition model under the training set is obtained through the third model. Based on the consistency loss, constraint loss, separation loss, and predefined anchor loss model of the multimodal recognition model under the training set, the anchor loss of the multimodal recognition model under the training set is generated. The classification loss function is used to obtain the classification loss of the multimodal recognition model under the training set. The supervised contrastive loss function is used to obtain the supervised contrastive loss of the multimodal recognition model under the training set. Based on the anchor loss, classification loss, supervised contrastive loss and predefined comprehensive loss model of the multimodal recognition model under the training set, the comprehensive loss value of the multimodal recognition model under the training set is generated.

[0033] The first model is defined as follows: ; This represents the consistency loss of the multimodal recognition model on the training set. The smaller the consistency loss of the multimodal recognition model on the training set, the more aggregated the sample features of the same real defect category on the training set. The larger the consistency loss of the multimodal recognition model on the training set, the less aggregated the sample features of the same real defect category on the training set. c represents the sequence number of the actual defect category; This represents the total number of actual defect categories; This represents the total number of samples for the c-th true defect category; This represents the set of sample indices for the c-th true defect category; Indicates that the nth sample belongs to This represents the feature vector of the nth sample belonging to the cth true defect category; This represents the central feature vector of the c-th true defect category; express and The square of the Euclidean distance between them; The second model is defined as follows: ; This represents the constraint loss of the multimodal recognition model under the training set. The smaller the constraint loss of the multimodal recognition model under the training set, the closer the class center of the multimodal recognition model is to the initial benchmark under the training set. The larger the constraint loss of the multimodal recognition model under the training set, the less the class center of the multimodal recognition model is to the initial benchmark under the training set. c represents the sequence number of the actual defect category; This represents the total number of actual defect categories; This represents the control coefficient for the c-th true defect category; This represents the central feature vector of the c-th true defect category; This represents the initial feature vector of the c-th true defect category; The third model is defined as follows: ; This represents the separation loss of the multimodal recognition model on the training set. The greater the separation loss of the multimodal recognition model on the training set, the weaker its ability to distinguish the sample features of the real defect categories on the training set; the smaller the separation loss of the multimodal recognition model on the training set, the stronger its ability to distinguish the sample features of the real defect categories on the training set. c represents the sequence number of the actual defect category; Indicates the index of the reference cluster; This represents the central feature vector of the c-th true defect category; Indicates the first The central eigenvectors of each reference cluster; This represents the temperature coefficient.

[0034] The anchor point loss model is defined as follows: ; This represents the anchor loss of the multimodal recognition model on the training set. The larger the anchor loss of the multimodal recognition model on the training set, the worse its ability to distinguish between categories. The smaller the anchor point loss of a multimodal recognition model in the training set, the stronger the model's ability to distinguish between category anchor points. This represents the consistency loss of the multimodal recognition model on the training set. This represents the constrained loss of the multimodal recognition model on the training set; This represents the separation loss of the multimodal recognition model on the training set; This represents the first weighting coefficient. This represents the second weighting coefficient. This represents the third weighting coefficient.

[0035] The comprehensive loss model is defined as follows: ; This represents the overall loss value of the multimodal recognition model on the training set. The higher the overall loss value of the multimodal recognition model on the training set, the stronger the recognition ability of the multimodal recognition model on the training set. The lower the overall loss value of the multimodal recognition model on the training set, the weaker the recognition ability of the multimodal recognition model on the training set. This is the first adjustment parameter; This is the second adjustment parameter. , ; This represents the classification loss of the multimodal recognition model on the training set; This represents the anchor loss of the multimodal recognition model on the training set; This represents the supervised contrastive loss of the multimodal recognition model on the training set.

[0036] S204. When the comprehensive loss value is less than the preset loss value, stop training the multimodal recognition model and save the trained multimodal recognition model. Among them, when the overall loss value is less than the preset loss value, the training of the multimodal recognition model is stopped. This can balance the fitting effect and generalization ability of the multimodal recognition model, which is conducive to improving the practicality of the multimodal recognition model.

[0037] S205: Based on the multimodal data of the current continuously cast billet and the predefined generation method, determine the current features, identify the current features through the trained multimodal recognition model, and generate the defect category to which the current continuously cast billet belongs.

[0038] The current defect categories for continuously cast billets include cracks, scabs, inclusions, roll marks, bubbles, scratches, and slag.

[0039] Cracks, scabs, inclusions, roll marks, bubbles, scratches, and slag on continuously cast billets are all surface or near-surface defects caused by factors such as process, equipment, and raw materials during the continuous casting production process.

[0040] Cracks are continuous fractures that form on the surface or inside of a continuously cast billet during solidification or cooling when the stress exceeds the material's bearing capacity. Scabs are raised, scar-like defects formed when metal particles, unmelted refractory materials, or protective slag, formed by splashing molten steel during the casting process, adhere to the surface of the continuously cast billet and solidify. Inclusions are non-metallic inclusions, slag, or foreign impurities that have not floated to the surface and were not removed from the molten steel. These inclusions are either encased inside or exposed on the surface during the solidification of the continuously cast billet, forming heterogeneous defects. Roll marks are periodic and regular indentations pressed onto the surface of the continuously cast billet during the conveying process due to problems such as pits or steel sticking on the surface of the rolls, or abnormal roll gap pressure. Bubbles are circular or elliptical cavities formed inside or on the surface of the continuously cast billet when dissolved gases in molten steel fail to escape in time during solidification. Scratches are fine, elongated scratch-like defects formed on the surface of continuously cast billets during the conveying and straightening process when they rub or scrape against sharp parts of the equipment. Slag formation is a layered defect formed when protective slag, refractory material debris, etc., adhere to and solidify on the surface of the continuously cast billet during the continuous casting process, resulting in a loose texture and poor bonding with the matrix.

[0041] For ease of explanation, the following example is provided: For example, when the defect category of the current continuous casting billet is crack or scratch, the online grinding equipment is immediately triggered to precisely grind the defect location of the current continuous casting billet to obtain a qualified repaired current continuous casting billet, and then the qualified repaired current continuous casting billet is directly flowed into the subsequent rolling process. When the current continuous casting billet is classified as a defect category of scale or roll mark, an instruction is sent to the control system. The control system then moves the current continuous casting billet to the offline repair area for secondary cleaning to prevent the defect from affecting the surface accuracy of the finished product. When the current continuous casting billet is classified as having inclusions or slag, a sorting robotic arm will move the billet to the scrap area to prevent it from entering the rolling process and causing damage to the rolls or breakage of the finished product.

[0042] The beneficial effects of the embodiments of this application are as follows: Firstly, based on the current multimodal data of the continuous casting billet and the predefined generation method, the current features are determined, and the current features are identified by the trained multimodal recognition model to generate the defect category of the current continuous casting billet. Since there is no need for manual identification of the defect category of the current continuous casting billet, the identification time of identifying the defect category of the current continuous casting billet can be reduced, which is conducive to improving the identification efficiency of the defect category of the current continuous casting billet. Secondly, since the trained multimodal recognition model is not affected by human subjective factors, it is beneficial to improve the reliability of the defect category of the current continuous casting billet.

[0043] Please see Figure 3 , Figure 3 The flowchart for displaying the list processing results provided in this application embodiment is described in detail below: S301, Obtain the multimodal data of the current continuous casting billet. The multimodal data of the current continuous casting billet includes the sensor data of the current continuous casting billet, the defect image of the current continuous casting billet, and the log data of the current continuous casting billet. The sensor data for continuous casting billets currently refers to the data obtained in real time by various industrial sensors deployed on the continuous casting production line, which collect data on the physical state and environmental parameters of the entire continuous casting billet production process.

[0044] The sensor data for the current continuous casting billet includes: surface temperature data of the current continuous casting billet collected by an infrared temperature sensor, and acoustic wave reflection signals of the internal structure of the current continuous casting billet collected by an ultrasonic sensor.

[0045] Currently, defect images of continuously cast billets refer to image data acquired by industrial vision acquisition equipment that can intuitively present the surface and near-surface defect characteristics of continuously cast billets, and are the core visual basis for defect category determination.

[0046] The current defect images of the continuous casting billet include a two-dimensional image and an infrared thermal image of the current continuous casting billet. The two-dimensional image of the current continuous casting billet can clearly show the appearance features such as the direction of cracks, the morphology of scales, and the distribution of roll marks. The infrared thermal image of the current continuous casting billet can show the temperature difference between the defect area and the normal area of ​​the current continuous casting billet, which is convenient for identifying hidden defects such as internal cracks in the current continuous casting billet.

[0047] The current continuous casting billet log data refers to various data and information related to the current continuous casting billet automatically recorded by the control system on the continuous casting production line during operation. The current continuous casting billet log data includes the operating status log of the production equipment and the quality inspection log of the current continuous casting billet.

[0048] S302, extract features from the sensor data of the current continuous casting billet to obtain the features of the sensor data of the continuous casting billet, extract features from the defect image of the current continuous casting billet to obtain the features of the defect image of the current continuous casting billet, and extract features from the log data of the current continuous casting billet to obtain the features of the log data of the current continuous casting billet. S303: The features of the sensor data of the current continuous casting billet, the features of the defect image of the current continuous casting billet, and the features of the log data of the current continuous casting billet are fused to obtain the current features. The current features are then identified by the trained multimodal recognition model to generate the defect category of the current continuous casting billet.

[0049] In this embodiment of the application, the current features are identified by the trained multimodal recognition model, and the defect category of the current continuous casting billet is generated, which can greatly reduce the probability of misjudgment and ensure the reliability of the defect category of the current continuous casting billet.

[0050] For the continuous casting billet defect identification method described in the above embodiments, please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic block diagram of the continuous casting billet defect identification device provided in the embodiments of this application. Figure 4 The continuous casting billet defect identification device 400 shown can be applied to, for example... Figure 1 The application scenario diagram shows electronic devices. The following section uses electronic devices as an example to illustrate this. Figure 4 The continuous casting billet defect identification device 400 shown will be described in detail. The continuous casting billet defect identification device 400 may include an acquisition module 401, a composition module 402, a generation module 403, a storage module 404, and an identification module 405.

[0051] The acquisition module 401 is used to combine preset features and preset real defect categories of continuous casting billets into a sample; The component module 402 is used to assemble different samples into a training set and train a multimodal recognition model based on the training set. The generation module 403 is used to generate the anchor loss of the multimodal recognition model under the training set through a predefined anchor loss model, and to generate the comprehensive loss value of the multimodal recognition model under the training set based on the anchor loss of the multimodal recognition model under the training set and a predefined comprehensive loss model. The save module 404 is used to stop training the multimodal recognition model and save the trained multimodal recognition model when the comprehensive loss value is less than the preset loss value. The identification module 405 is used to determine the current features based on the multimodal data of the current continuous casting billet and the predefined generation method, identify the current features through the trained multimodal identification model, and generate the defect category to which the current continuous casting billet belongs.

[0052] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0053] The beneficial effects of the embodiments of this application are as follows: Firstly, based on the current multimodal data of the continuous casting billet and the predefined generation method, the current features are determined, and the current features are identified by the trained multimodal recognition model to generate the defect category of the current continuous casting billet. Since there is no need for manual identification of the defect category of the current continuous casting billet, the identification time of identifying the defect category of the current continuous casting billet can be reduced, which is conducive to improving the identification efficiency of the defect category of the current continuous casting billet. Secondly, since the trained multimodal recognition model is not affected by human subjective factors, it is beneficial to improve the reliability of the defect category of the current continuous casting billet.

[0054] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0055] like Figure 5 As shown, Figure 5 The electronic device 2 includes: at least one processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the at least one processor 20, wherein the processor 20 executes the computer program 22 to implement the steps in any of the above method embodiments.

[0056] The electronic device 2 may include, but is not limited to, a processor 20 and a memory 21. Those skilled in the art will understand that... Figure 5 This is merely an example of electronic device 2 and does not constitute a limitation on electronic device 2. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, it may also include input / output devices, network access devices, etc.

[0057] The processor 20 is used to run a computer program 22 stored in the memory 21, and performs the following steps when executing the computer program 22: Combine the preset features and the preset actual defect categories of the continuous casting billet into a sample; Different samples are combined into a training set, and a multimodal recognition model is trained based on the training set; Anchor point loss of the multimodal recognition model under the training set is generated using a predefined anchor point loss model. Based on the anchor point loss of the multimodal recognition model under the training set and the predefined comprehensive loss model, comprehensive loss value of the multimodal recognition model under the training set is generated. When the overall loss value is less than the preset loss value, stop training the multimodal recognition model and save the trained multimodal recognition model. Based on the current multimodal data of the continuously cast billet and the predefined generation method, the current features are determined, and the current features are identified by the trained multimodal recognition model to generate the defect category of the current continuously cast billet.

[0058] The processor 20 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors, field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0059] In some embodiments, the memory 21 may be an internal storage unit of the electronic device 2, such as a hard disk or memory of the electronic device 2. In other embodiments, the memory 21 may be an external storage device of the electronic device 2, such as a plug-in hard disk, SmartMediaCard (SMC), SecureDigital (SD) card, or FlashCard equipped on the electronic device 2.

[0060] Furthermore, the memory 21 may include both internal storage units and external storage devices of the electronic device 2. The memory 21 is used to store the operating system, applications, boot loader, data, and other programs, such as the program code of the computer program. The memory 21 can also be used to temporarily store data that has been output or will be output.

[0061] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0062] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0063] The computer-readable storage medium may also be an external storage device of the continuous casting billet defect identification device or electronic device, such as a plug-in hard drive, SmartMedia Card (SMC), Secure Digital (SD) card, Flash Card, or non-transitory computer-readable storage medium equipped on the continuous casting billet defect identification device or electronic device.

[0064] Since the computer program stored in the computer-readable storage medium can execute any of the continuous casting billet defect identification methods provided in the embodiments of this application, the computer-readable storage medium can achieve the beneficial effects that any of the continuous casting billet defect identification methods provided in the embodiments of this application can achieve, as detailed in the preceding embodiments, and will not be repeated here.

[0065] This application provides a computer program product that, when run on an electronic device, causes the electronic device to execute the aforementioned continuous casting billet defect identification method.

[0066] When a computer program is loaded into an electronic device, it can perform the following steps: Combine the preset features and the preset actual defect categories of the continuous casting billet into a sample; Different samples are combined into a training set, and a multimodal recognition model is trained based on the training set; Anchor point loss of the multimodal recognition model under the training set is generated using a predefined anchor point loss model. Based on the anchor point loss of the multimodal recognition model under the training set and the predefined comprehensive loss model, comprehensive loss value of the multimodal recognition model under the training set is generated. When the overall loss value is less than the preset loss value, stop training the multimodal recognition model and save the trained multimodal recognition model. Based on the current multimodal data of the continuously cast billet and the predefined generation method, the current features are determined, and the current features are identified by the trained multimodal recognition model to generate the defect category of the current continuously cast billet.

[0067] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0068] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for identifying defects in continuously cast billets, characterized in that, The continuous casting billet defect identification method, applied to electronic devices, includes: Combine the preset features and the preset actual defect categories of the continuous casting billet into a sample; Different samples are combined into a training set, and a multimodal recognition model is trained based on the training set; The first model obtains the consistency loss of the multimodal recognition model under the training set. The second model obtains the constraint loss of the multimodal recognition model under the training set. The third model obtains the separation loss of the multimodal recognition model under the training set. Based on the consistency loss, constraint loss, separation loss, and predefined anchor loss model, the anchor loss of the multimodal recognition model under the training set is generated. The classification loss and supervised contrast loss are obtained using the classification loss function and the supervised contrast loss function, respectively. Based on the anchor loss, classification loss, supervised contrast loss, and predefined comprehensive loss model, the comprehensive loss value of the multimodal recognition model under the training set is generated. When the overall loss value is less than the preset loss value, stop training the multimodal recognition model and save the trained multimodal recognition model. Based on the current multimodal data of the continuously cast billet and the predefined generation method, the current features are determined, and the current features are identified by the trained multimodal recognition model to generate the defect category of the current continuously cast billet.

2. The method for identifying defects in continuously cast billets according to claim 1, characterized in that, The step of combining preset features and preset continuous casting billet actual defect categories into a sample includes: Acquire multimodal data of a preset continuous casting billet, which includes sensor data of the preset continuous casting billet, defect images of the preset continuous casting billet, and log data of the preset continuous casting billet; Feature extraction is performed on the sensor data of the preset continuous casting billet to obtain the features of the sensor data of the continuous casting billet; feature extraction is performed on the defect image of the preset continuous casting billet to obtain the features of the defect image of the preset continuous casting billet; feature extraction is performed on the log data of the preset continuous casting billet to obtain the features of the log data of the preset continuous casting billet. The features of the sensor data of the preset continuous casting billet, the features of the defect image of the preset continuous casting billet, and the features of the log data of the preset continuous casting billet are fused to obtain the preset features. The preset features and the actual defect categories of the preset continuous casting billet are combined to form a sample.

3. The method for identifying defects in continuously cast billets according to claim 1, characterized in that, The process involves determining the current features based on the multimodal data of the current continuously cast billet and a predefined determination method, identifying the current features through a trained multimodal recognition model, and generating the defect category of the current continuously cast billet, including: Acquire the multimodal data of the current continuous casting billet, which includes the sensor data of the current continuous casting billet, the defect image of the current continuous casting billet, and the log data of the current continuous casting billet; Feature extraction is performed on the sensor data of the current continuous casting billet to obtain the features of the sensor data of the continuous casting billet; feature extraction is performed on the defect image of the current continuous casting billet to obtain the features of the defect image of the current continuous casting billet; feature extraction is performed on the log data of the current continuous casting billet to obtain the features of the log data of the current continuous casting billet. The features of the sensor data of the current continuous casting billet, the features of the defect image of the current continuous casting billet, and the features of the log data of the current continuous casting billet are fused to obtain the current features. The current features are then identified by the trained multimodal recognition model to generate the defect category of the current continuous casting billet.

4. The method for identifying defects in continuously cast billets according to claim 1, characterized in that, The comprehensive loss model is defined as follows: ； This represents the overall loss value of the multimodal recognition model on the training set. The higher the overall loss value of the multimodal recognition model on the training set, the stronger the recognition ability of the multimodal recognition model on the training set. The lower the overall loss value of the multimodal recognition model on the training set, the weaker the recognition ability of the multimodal recognition model on the training set. This is the first adjustment parameter; This is the second adjustment parameter. , ; This represents the classification loss of the multimodal recognition model on the training set; This represents the anchor loss of the multimodal recognition model on the training set; This represents the supervised contrastive loss of the multimodal recognition model on the training set.

5. The method for identifying defects in continuously cast billets according to claim 1, characterized in that, The anchor point loss model is defined as follows: ； This represents the anchor loss of the multimodal recognition model on the training set. The larger the anchor loss of the multimodal recognition model on the training set, the worse its ability to distinguish between categories. The smaller the anchor point loss of a multimodal recognition model in the training set, the stronger the model's ability to distinguish between category anchor points. This represents the consistency loss of the multimodal recognition model on the training set. This represents the constrained loss of the multimodal recognition model on the training set; This represents the separation loss of the multimodal recognition model on the training set; This represents the first weighting coefficient. This represents the second weighting coefficient. This represents the third weighting coefficient.

6. The method for identifying defects in continuously cast billets according to claim 3, characterized in that, The first model is defined as follows: ； This represents the consistency loss of the multimodal recognition model on the training set. The smaller the consistency loss of the multimodal recognition model on the training set, the more aggregated the sample features of the same real defect category are in the training set. The greater the consistency loss of a multimodal recognition model on the training set, the less aggregated the sample features of the same real defect category on the training set. c represents the sequence number of the actual defect category; This represents the total number of actual defect categories; This represents the total number of samples for the c-th true defect category; This represents the set of sample indices for the c-th true defect category; Indicates that the nth sample belongs to This represents the feature vector of the nth sample belonging to the cth true defect category; This represents the central feature vector of the c-th true defect category; express and The square of the Euclidean distance between them; The second model is defined as follows: ； This represents the constraint loss of the multimodal recognition model under the training set. The smaller the constraint loss of the multimodal recognition model under the training set, the closer the class center of the multimodal recognition model is to the initial benchmark under the training set. The larger the constraint loss of the multimodal recognition model under the training set, the less the class center of the multimodal recognition model is to the initial benchmark under the training set. c represents the sequence number of the actual defect category; This represents the total number of actual defect categories; This represents the control coefficient for the c-th true defect category; This represents the central feature vector of the c-th true defect category; This represents the initial feature vector of the c-th true defect category; The third model is defined as follows: ； This represents the separation loss of the multimodal recognition model on the training set. The greater the separation loss of the multimodal recognition model on the training set, the weaker its ability to distinguish the sample features of the real defect categories on the training set; the smaller the separation loss of the multimodal recognition model on the training set, the stronger its ability to distinguish the sample features of the real defect categories on the training set. c represents the sequence number of the actual defect category; Indicates the index of the reference cluster; This represents the central feature vector of the c-th true defect category; Indicates the first The central eigenvectors of each reference cluster; This represents the temperature coefficient.

7. A continuous casting billet defect identification device, characterized in that, Applied to electronic devices, including: The acquisition module is used to combine preset features and preset real defect categories of continuous casting billets into a sample; The component module is used to combine different samples into a training set, and to train a multimodal recognition model based on the training set; The generation module is used to obtain the consistency loss of the multimodal recognition model under the training set through the first model, the constraint loss of the multimodal recognition model under the training set through the second model, and the separation loss of the multimodal recognition model under the training set through the third model. Based on the consistency loss, constraint loss, separation loss and predefined anchor loss model of the multimodal recognition model under the training set, the module generates the anchor loss of the multimodal recognition model under the training set. Based on the classification loss function, the module obtains the classification loss and the supervised contrastive loss of the multimodal recognition model under the training set. Based on the anchor loss, classification loss and supervised contrastive loss and predefined comprehensive loss model of the multimodal recognition model under the training set, the module generates the comprehensive loss value of the multimodal recognition model under the training set. The save module is used to stop training the multimodal recognition model and save the trained multimodal recognition model when the comprehensive loss value is less than the preset loss value. The identification module is used to determine the current features based on the multimodal data of the current continuous casting billet and the predefined generation method. It then identifies the current features through the trained multimodal identification model and generates the defect category to which the current continuous casting billet belongs.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the continuous casting billet defect identification method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the continuous casting billet defect identification method as described in any one of claims 1 to 6.

Images