CN-121982364-A - Continuous casting billet defect identification method and device, electronic equipment and storage medium

CN121982364ACN 121982364 ACN121982364 ACN 121982364ACN-121982364-A

Abstract

The application relates to the technical field of steel and defect identification, and discloses a continuous casting billet defect identification method, a device, electronic equipment and a storage medium, wherein the method comprises the steps of forming a sample by preset characteristics and the real defect category of a preset continuous casting billet; the method comprises the steps of forming a training set by different samples, training a multi-mode identification model based on the training set, generating anchor point loss of the multi-mode identification model under the training set through the anchor point loss model, generating a comprehensive loss value of the multi-mode identification model under the training set according to the anchor point loss of the multi-mode identification model under the training set and a predefined comprehensive loss model, stopping training the multi-mode identification model when the comprehensive loss value is smaller than a preset loss value, storing the trained multi-mode identification model, and identifying current characteristics through the trained multi-mode identification model to generate the defect category of the current continuous casting billet. The method is beneficial to the identification efficiency of the defect category of the current continuous casting billet.

Inventors

ZHANG YUHANG
PENG HAN
MA TAO
XU BO

Assignees

湖南红普创新科技发展有限公司

Dates

Publication Date: 20260505
Application Date: 20251223

Claims (10)

1. A method for identifying a defect of a continuous casting billet, which is applied to an electronic device, the method comprising: Forming a sample by the preset characteristics and the real defect types of the preset continuous casting blank; Different samples form a training set, and a multi-mode recognition model is trained based on the training set; generating anchor point loss of the multi-modal identification model under the training set through a predefined anchor point loss model, and generating a comprehensive loss value of the multi-modal identification model under the training set according to the anchor point loss of the multi-modal identification model under the training set and the predefined comprehensive loss model; when the comprehensive loss value is smaller than the preset loss value, stopping training the multi-modal identification model, and storing the trained multi-modal identification model; and determining the current characteristics based on the multi-mode data of the current continuous casting billet and the predefined generation mode, and identifying the current characteristics through the trained multi-mode identification model to generate the defect category of the current continuous casting billet.
2. The method for identifying defects in a continuous casting slab according to claim 1, wherein the step of forming a sample of the predetermined feature and the true defect class of the predetermined continuous casting slab comprises: Acquiring multi-mode data of a preset continuous casting billet, wherein the multi-mode data of the preset continuous casting billet comprise 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; Performing feature extraction on sensor data of a preset continuous casting billet to obtain features of the sensor data of the continuous casting billet, performing feature extraction on a defect image of the preset continuous casting billet to obtain features of the defect image of the preset continuous casting billet, and performing feature extraction on log data of the preset continuous casting billet to obtain features of the log data of the preset continuous casting billet; Fusing the characteristics of sensor data of the preset continuous casting billet, the characteristics of defect images of the preset continuous casting billet and the characteristics of log data of the preset continuous casting billet to obtain preset characteristics, and forming a sample by the preset characteristics and the real defect types of the preset continuous casting billet.
3. The method for identifying defects of a continuous casting billet according to claim 1, wherein the generating, by the predefined anchor point loss model, anchor point loss of the multi-modal identification model under the training set, generating, according to the anchor point loss of the multi-modal identification model under the training set and the predefined integrated loss model, an integrated loss value of the multi-modal identification model under the training set includes: Obtaining consistency loss of the multi-mode recognition model under the training set through the first model, obtaining constraint loss of the multi-mode recognition model under the training set through the second model, and obtaining separation loss of the multi-mode recognition model under the training set through the third model; Generating anchor point loss of the multi-modal identification model under the training set according to consistency loss of the multi-modal identification model under the training set, constraint loss of the multi-modal identification model under the training set, separation loss of the multi-modal identification model under the training set and a predefined anchor point loss model; The method comprises the steps of obtaining the classification loss of a multi-modal identification model under a training set through a classification loss function, obtaining the supervision comparison loss of the multi-modal identification model under the training set through a supervision comparison loss function, and generating the comprehensive loss value of the multi-modal identification model under the training set according to the anchor point loss of the multi-modal identification model under the training set, the classification loss of the multi-modal identification model under the training set, the supervision comparison loss of the multi-modal identification model under the training set and the predefined comprehensive loss model.
4. The method for identifying defects of continuous casting billets according to claim 1, wherein the steps of determining the current characteristics based on the multi-modal data of the current continuous casting billets and the predefined determination mode, identifying the current characteristics through the trained multi-modal identification model, and generating the defect category of the current continuous casting billets comprise: Acquiring multi-mode data of a current continuous casting billet, wherein the multi-mode data of the current continuous casting billet comprise sensor data of the current continuous casting billet, defect images of the current continuous casting billet and log data of the current continuous casting billet; extracting features of sensor data of the current continuous casting billet to obtain features of the sensor data of the continuous casting billet, extracting features of a defect image of the current continuous casting billet to obtain features of the defect image of the current continuous casting billet, extracting features of log data of the current continuous casting billet to obtain features of the log data of the current continuous casting billet; The characteristics of sensor data of the current continuous casting billet, the characteristics of defect images of the current continuous casting billet and the characteristics of log data of the current continuous casting billet are fused to obtain the current characteristics, and the current characteristics are identified through the trained multi-mode identification model to generate the defect category of the current continuous casting billet.
5. The method for identifying defects in a continuous casting slab according to claim 1, wherein the integrated loss model is defined as follows: ; the comprehensive loss value of the multi-modal identification model under the training set is represented, the higher the comprehensive loss value of the multi-modal identification model under the training set is, the stronger the identification capability of the multi-modal identification model to the training set is, the lower the comprehensive loss value of the multi-modal identification model under the training set is, and the weaker the identification capability of the multi-modal identification model to the training set is; Is a first adjustment parameter; For the second adjustment parameter, a second adjustment parameter, 、 ; Representing the classification loss of the multi-modal recognition model under the training set; Representing anchor point loss of the multi-modal identification model under the training set; and the supervision contrast loss of the multi-modal identification model under the training set is represented.
6. The method for identifying defects in a continuous casting slab according to claim 1, An anchor point loss model is defined as follows: ; The anchor point loss of the multi-modal identification model under the training set is represented, and the larger the anchor point loss of the multi-modal identification model under the training set is, the worse the distinguishing capability of the multi-modal identification model in the aspect of category anchor points is explained; The smaller the anchor point loss of the multi-mode identification model under the training set is, the stronger the distinguishing capability of the multi-mode identification model in the aspect of category anchor points is; representing a consistency loss of the multi-modal identification model under the training set; Representing constraint loss of the multi-modal identification model under the training set; Representing the separation loss of the multi-modal recognition model under the training set; a first weight coefficient is represented and a second weight coefficient is represented, A second weight coefficient is represented and is used to represent, Representing a third weight coefficient.
7. A method of identifying a defect in a continuous casting billet according to claim 3, wherein the first model is defined as follows: ; the method comprises the steps of representing consistency loss of a multi-modal identification model under a training set, wherein the smaller the consistency loss of the multi-modal identification model under the training set is, the more aggregated sample features of the same real defect type of the multi-modal identification model under the training set are, and the larger the consistency loss of the multi-modal identification model under the training set is, the less aggregated sample features of the same real defect type of the multi-modal identification model under the training set are; c represents the serial number of the true defect class; Representing the total number of real defect categories; a sample total amount representing the c-th real defect class; a sample index set representing the c-th real defect class; Indicating that the nth sample belongs to A feature vector representing an nth sample belonging to a c-th true defect class; A center feature vector representing a c-th true defect class; Representation of And (3) with Square value of euclidean distance between them; A second model, defined as follows: ; The method comprises the steps of representing constraint loss of a multi-modal identification model under a training set, wherein the smaller the constraint loss of the multi-modal identification model under the training set is, the closer the category center of the multi-modal identification model under the training set is to an initial reference; c represents the serial number of the true defect class; Representing the total number of real defect categories; a control coefficient representing the c-th real defect class; A center feature vector representing a c-th true defect class; an initial feature vector representing a c-th true defect class; a third model, defined as follows: ; The method comprises the steps of obtaining a training set, representing the separation degree loss of a multi-modal identification model under the training set, wherein the larger the separation degree loss of the multi-modal identification model under the training set is, the weaker the multi-modal identification model is capable of distinguishing sample characteristics of real defect types under the training set is, and the smaller the separation degree loss of the multi-modal identification model under the training set is, the stronger the multi-modal identification model is capable of distinguishing the sample characteristics of the real defect types under the training set is; c represents the serial number of the true defect class; a sequence number representing a reference cluster; A center feature vector representing a c-th true defect class; represent the first Central feature vectors of the reference clusters; representing the temperature coefficient.
8. A continuous casting billet defect recognition apparatus, characterized by being applied to an electronic device, comprising: The acquisition module is used for forming a sample from preset characteristics and the real defect types of the preset continuous casting blank; the composition module is used for composing different samples into a training set and training a multi-mode identification model based on the training set; the generation module is used for generating anchor point loss of the multi-modal identification model under the training set through a predefined anchor point loss model, and generating a comprehensive loss value of the multi-modal identification model under the training set according to the anchor point loss of the multi-modal identification model under the training set and the predefined comprehensive loss model; The storage module is used for stopping training the multi-modal identification model when the comprehensive loss value is smaller than the preset loss value and storing the trained multi-modal identification model; The identification module is used for determining the current characteristics based on the multi-mode data of the current continuous casting billet and the predefined generation mode, identifying the current characteristics through the trained multi-mode identification model and generating the defect category of the current continuous casting billet.
9. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the method of identifying a strand defect according to any one of claims 1 to 7 when the computer program is executed.
10. A computer-readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the strand defect identification method according to any one of claims 1 to 7.

Description

Continuous casting billet defect identification method and device, electronic equipment and storage medium Technical Field The application relates to the technical field of steel and defect identification, in particular to a continuous casting billet defect identification method, a device, electronic equipment and a storage medium. Background The continuous casting blank is a steel blank produced by a continuous casting technology. After being processed by rolling, forging, heat treatment and other processes, the continuous casting blank can be manufactured into various steel products with different types and specifications. However, in the current continuous casting billet production process, the defect identification work is mainly finished by manpower, and the manual identification mode is time-consuming and labor-consuming, and is easily affected by subjective factors, because different people have differences in the judgment standards of defects, the accuracy and consistency of the identification result are poor. Therefore, how to identify the defect type of the current continuous casting billet is a technical problem to be solved. Disclosure of Invention The embodiment of the application provides a method, a device, electronic equipment and a storage medium for identifying defects of a continuous casting blank, which are used for solving the technical problem of how to identify the defect category of the current continuous casting blank. In a first aspect, an embodiment of the present application provides a method for identifying a defect of a continuous casting billet, which is applied to an electronic device, where the method includes: Forming a sample by the preset characteristics and the real defect types of the preset continuous casting blank; Different samples form a training set, and a multi-mode recognition model is trained based on the training set; generating anchor point loss of the multi-modal identification model under the training set through a predefined anchor point loss model, and generating a comprehensive loss value of the multi-modal identification model under the training set according to the anchor point loss of the multi-modal identification model under the training set and the predefined comprehensive loss model; when the comprehensive loss value is smaller than the preset loss value, stopping training the multi-modal identification model, and storing the trained multi-modal identification model; and determining the current characteristics based on the multi-mode data of the current continuous casting billet and the predefined generation mode, and identifying the current characteristics through the trained multi-mode identification model to generate the defect category of the current continuous casting billet. In one possible implementation manner of the first aspect, In a possible implementation manner of the first aspect, the forming a sample from the preset feature and the true defect class of the preset continuous casting slab includes: Acquiring multi-mode data of a preset continuous casting billet, wherein the multi-mode data of the preset continuous casting billet comprise 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; Performing feature extraction on sensor data of a preset continuous casting billet to obtain features of the sensor data of the continuous casting billet, performing feature extraction on a defect image of the preset continuous casting billet to obtain features of the defect image of the preset continuous casting billet, and performing feature extraction on log data of the preset continuous casting billet to obtain features of the log data of the preset continuous casting billet; Fusing the characteristics of sensor data of the preset continuous casting billet, the characteristics of defect images of the preset continuous casting billet and the characteristics of log data of the preset continuous casting billet to obtain preset characteristics, and forming a sample by the preset characteristics and the real defect types of the preset continuous casting billet. In a possible implementation manner of the first aspect, the generating, by using a predefined anchor point loss model, anchor point loss of the multi-modal identification model under the training set, generating, according to the anchor point loss of the multi-modal identification model under the training set and the predefined comprehensive loss model, a comprehensive loss value of the multi-modal identification model under the training set includes: Obtaining consistency loss of the multi-mode recognition model under the training set through the first model, obtaining constraint loss of the multi-mode recognition model under the training set through the second model, and obtaining separation loss of the multi-mode recognition model under the training set through the third model; Genera