Continuous casting round billet core defect classification method, system, terminal and medium

Abstract

The invention belongs to the field of billet defect detection, and particularly discloses a continuous casting round billet core defect classification method and system, a terminal and a medium, multiple groups of original time domain signals of a continuous casting round billet core are collected, multiple waveform time domain features are extracted through the original time domain signals, and multiple multi-dimensional time domain feature vectors are constructed; constructing a BP neural network, and optimizing the BP neural network by using a genetic algorithm; configuring a sample label for each multi-dimensional time-domain feature vector according to the type of the core defect of the continuous casting round billet, forming a sample set by the plurality of multi-dimensional time-domain feature vectors and the sample labels, and training the optimized BP neural network by using the sample set; and collecting a target time domain signal of the core part of the target continuous casting round billet, and inputting the target time domain signal into the trained BP neural network to obtain defect classification of the core part of the target continuous casting round billet. According to the method, the BP neural network is optimized to process the time domain signal of the continuous casting round billet core so as to realize defect classification, and the defect classification efficiency and accuracy are improved.

Description

Technical Field

[0001] The invention belongs to the field of steel billet defect detection, and in particular relates to a method, system, terminal and medium for classifying core defects of a continuous casting round billet. Background Art

[0002] As an intermediate product of steel, the core defects of high-quality special steel billets, such as cracks, shrinkage cavities, and porosity, will directly affect the yield rate of finished products, mainly in terms of affecting the mechanical properties and microstructure of the products. In terms of affecting mechanical properties, core cracks, porosity, and shrinkage cavities will directly affect the mechanical properties of rolled products. These defects will destroy the continuity of the metal, resulting in a decrease in the mechanical properties of the material, such as strength, toughness, and plasticity. For example, the presence of cracks will reduce the tensile strength and yield strength of the material, making it more likely to break when subjected to stress. Porosity may cause stress concentration, causing cracks or fractures in the material when subjected to stress. In terms of affecting the microstructure, core defects may also affect the microstructure of the rolled product. For example, shrinkage cavities may be stretched into long and thin streamlines during the rolling process, forming a banded structure, which will cause the metal to have obvious anisotropy. This anisotropy will affect the mechanical properties and processing properties of the material, resulting in differences in the performance of the material in different directions. In addition, defects such as porosity can also make the material's microstructure uneven, thus affecting its overall performance.

[0003] According to the different core defects of continuous casting round billets, different rolling processes can be selected to significantly improve the qualified rate of rolled materials. For example, through the liquid core large reduction rolling technology and the optimization of the heating process, the core looseness of the large-size round billets of special steel can be effectively eliminated or improved during the rolling process; through temperature-controlled rolling, reasonable setting of material shape, modification of hole shape, optimization of rolling process, heat treatment and other rolling process means, the core cracks of large-size round billets of special steel can be effectively eliminated or improved during the rolling process. Therefore, before the rolling operation, the core defects of the continuous casting round billets need to be detected to determine the defect type.

[0004] At present, the main method for detecting core defects of continuous casting round billets is to use ultrasonic flaw detection equipment to obtain the original time domain signal of the core of the continuous casting round billet to generate the corresponding time domain sequence waveform, and then manually check the waveform to determine which waveform it is based on experience. This method relies on manual detection, but the accuracy cannot meet the requirements and the efficiency is low. Another current method is to use a neural network to process the relevant parameters in the process of continuous casting round billet generation, and then predict the defects of the continuous casting round billet. However, this method is based on production parameters, etc., and does not analyze the actual products produced, which affects the accuracy of defect analysis. Summary of the invention

[0005] To solve the above problems, the present invention provides a method, system, terminal and medium for classifying core defects of continuous casting round billets. By optimizing the BP neural network to process the time-domain signals of the core of continuous casting round billets, the defect classification efficiency and accuracy are improved.

[0006] In a first aspect, the technical solution of the present invention provides a method for classifying core defects of continuous casting round billets, including the following steps: Collect multiple groups of original time-domain signals of the core of continuous casting round billets, extract multiple waveform time-domain features from the original time-domain signals, and construct multiple multi-dimensional time-domain feature vectors; Construct a BP neural network and optimize the BP neural network using a genetic algorithm; Configure sample labels for each multi-dimensional time-domain feature vector according to the core defect types of continuous casting round billets, form a sample set with multiple multi-dimensional time-domain feature vectors and sample labels, and use the sample set to train the optimized BP neural network; Collect the target time-domain signal of the core of the target continuous casting round billet, and input the target time-domain signal into the trained BP neural network to obtain the defect classification of the core of the target continuous casting round billet.

[0007] In an optional embodiment, extracting multiple waveform time-domain features from the original time-domain signal specifically includes: Calculate the average value, standard deviation, skewness, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, amplitude index, waveform index, impact index, margin index, and energy of the original time-domain signal through the waveform amplitude of the original time-domain signal.

[0008] In an optional embodiment, the constructed BP neural network includes an input layer, a hidden layer, and an output layer. The input layer includes 13 input neurons, the hidden layer includes 8 hidden layer neurons, and the output layer includes 4 output neurons; the activation function of the BP neural network is the ReLU function, the loss function uses cross-entropy loss, the optimizer is Adam, and the output layer uses the Softmax function.

[0009] In an optional embodiment, using a genetic algorithm to optimize the BP neural network specifically includes: Set the training parameters of the BP neural network, including the maximum number of iterations, error threshold, and learning rate for the training of the BP neural network; Set the optimization parameters, including the number of genetic generations, population size, number of optimization parameters, and optimization variable boundaries; Initialize the population, including initializing the real number coding, parameters of the selection function, parameters of the crossover function, and parameters of the mutation function; Run the genetic algorithm to optimize the BP neural network to obtain the optimal population; Decode the optimal population to obtain the optimal parameters of the BP neural network.

[0010] In an alternative embodiment, the optimal parameters of the BP neural network include the weight from input to hidden layer, the bias from input to hidden layer, the weight from hidden to output layer, and the bias from hidden to output layer.

[0011] In an alternative embodiment, the core defect types of the continuous casting round billet include crack defects, shrinkage cavity defects, porosity defects, and normal morphology.

[0012] In a second aspect, the technical solution of the present invention provides a core defect classification system for continuous casting round billets, including: An original feature collection module, configured to collect multiple groups of original time-domain signals of the core of the continuous casting round billet, extract multiple waveform time-domain features from the original time-domain signals, and construct multiple multi-dimensional time-domain feature vectors; A classification network model optimization module, configured to construct a BP neural network and optimize the BP neural network using a genetic algorithm; A classification network model training module, configured to configure sample labels for each multi-dimensional time-domain feature vector according to the core defect types of the continuous casting round billet, form a sample set with the multiple multi-dimensional time-domain feature vectors and the sample labels, and use the sample set to train the optimized BP neural network; A defect classification module, configured to collect a target time-domain signal of the core of the target continuous casting round billet, and input the target time-domain signal into the trained BP neural network to obtain the defect classification of the core of the target continuous casting round billet.

[0013] In an alternative embodiment, extracting multiple waveform time-domain features from the original time-domain signal specifically includes: calculating the average value, standard deviation, skewness, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, amplitude index, waveform index, impact index, margin index, and energy of the original time-domain signal through the waveform amplitude of the original time-domain signal.

[0014] In a third aspect, the technical solution of the present invention provides a terminal, including: A memory, configured to store a core defect classification program for continuous casting round billets; A processor, configured to implement the steps of the core defect classification method for continuous casting round billets as described in any one of the above when executing the core defect classification program for continuous casting round billets.

[0015] In a fourth aspect, the technical solution of the present invention provides a computer-readable storage medium, on which a core defect classification program for continuous casting round billets is stored, and when the core defect classification program for continuous casting round billets is executed by a processor, the steps of the core defect classification method for continuous casting round billets as described in any one of the above are implemented.

[0016] A method, system, terminal and medium for classifying core defects of continuous casting round billets provided by the present invention has the following beneficial effects compared with the prior art: constructing a BP neural network and optimizing it using a genetic algorithm, collecting the original time-domain signals of the core of the continuous casting round billet and extracting multiple waveform time-domain features, then using the multiple waveform time-domain features as the input for optimizing the BP neural network for network training, and finally using the trained optimized BP neural network to classify the core defects of the continuous casting round billet. The present invention combines the global search ability of the genetic algorithm and the non-linear mapping ability of the BP neural network, showing good performance in multi-feature classification prediction, thereby realizing data classification of typical core defects of continuous casting round billets, solving the problem of low efficiency and accuracy relying on manual detection, and classifying based on the time-domain signals of the actual core of the continuous casting round billet, which can have higher classification accuracy compared with predicting using production parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a schematic flow chart of a method for classifying core defects of continuous casting round billets provided by an embodiment of the present invention.

[0019] Figure 2 It is a schematic diagram of the circumferential point-taking position of the round billet.

[0020] Figure 3 It is an ultrasonic waveform diagram of the time-domain ultrasonic data of the round billet.

[0021] Figure 4 It is a structural diagram of the BP neural network model.

[0022] Figure 5 It is a fitness change curve.

[0023] Figure 6 It is a comparison curve diagram of the prediction results of the training set.

[0024] Figure 7 It is a comparison curve diagram of the prediction results of the test set.

[0025] Figure 8 It is a scatter diagram comparing the predicted values and the true values of the training set.

[0026] Figure 9 It is a scatter diagram comparing the predicted values and the true values of the test set.

[0027] Figure 10Schematic block diagram of a core defect classification system for continuous casting round billets provided by an embodiment of the present invention.

[0028] Figure 11 Schematic diagram of the structure of a terminal provided by an embodiment of the present invention. Detailed implementation manners

[0029] To make the objectives, features, and advantages of the present invention more obvious and understandable, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the specific embodiments of the present invention. Obviously, the embodiments described below are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments, and are not intended to limit the present invention.

[0031] The current main method for detecting the core defects of continuous casting round billets is to use ultrasonic flaw detection equipment to obtain the original time-domain signals of the core of the continuous casting round billet to generate corresponding time-domain sequence waveforms, and then manually view the waveforms and judge which waveform it is according to experience. This method relying on manual detection cannot meet the accuracy requirements and has low efficiency. Another current method is to use a neural network to process the relevant parameters in the process of continuous casting round billet generation, and then predict the defects of the continuous casting round billet. However, this method is to predict according to production parameters, etc., and does not analyze the actually produced products, which affects the accuracy of defect analysis. Aiming at the problems of low efficiency and insufficient accuracy in the current core defect classification method of continuous casting round billets, this embodiment provides a time-domain ultrasonic data classification method for core defects of continuous casting round billets based on a genetic algorithm optimized BP neural network, constructs a BP neural network and uses a genetic algorithm for optimization, collects the original time-domain signals of the core of the continuous casting round billet and extracts multiple waveform time-domain features, then uses the multiple waveform time-domain features as the input of the optimized BP neural network for network training, and finally uses the trained optimized BP neural network to classify the core defects of the continuous casting round billet. This method combines the global search ability of the genetic algorithm and the non-linear mapping ability of the BP neural network, and shows good performance in multi-feature classification prediction, so as to realize the data classification of typical core defects of continuous casting round billets, solve the problems of low efficiency and accuracy relying on manual detection, and classify based on the time-domain signals of the actual core of the continuous casting round billet, which can have higher classification accuracy compared with predicting using production parameters.

[0032] Figure 1Schematic flowchart of a method for classifying core defects of continuous casting round billets provided by an embodiment of the present invention. Among them, Figure 1 The execution subject can be a system for classifying core defects of continuous casting round billets. The method for classifying core defects of continuous casting round billets provided by the embodiment of the present invention is executed by a computer device. Correspondingly, the system for classifying core defects of continuous casting round billets runs in the computer device. According to different requirements, the order of steps in this flowchart can be changed, and some can be omitted.

[0033] As Figure 1 shown, the method includes the following steps.

[0034] S1. Collect multiple groups of original time-domain signals of the core of continuous casting round billets, extract multiple waveform time-domain features from the original time-domain signals, and construct multiple multi-dimensional time-domain feature vectors.

[0035] In this step, multiple groups of original time-domain signals of the core of continuous casting round billets are collected, waveform time-domain features are extracted from the original time-domain signals, and subsequently, the waveform time-domain features are used as the basis for defect classification to train a classification model. Among them, the waveform time-domain features can include various time-domain features such as average value and standard deviation, and then a multi-dimensional time-domain feature vector is constructed as the input of the classification model.

[0036] The original time-domain signals of the core of continuous casting round billets itself contain core defect information of continuous casting round billets. Extracting various time-domain features can comprehensively capture the signal characteristics, and the multi-dimensional time-domain feature vectors integrate these features, providing rich and valuable data for subsequent model training, improving the model's ability to distinguish different defect types, and thus improving the accuracy of defect classification.

[0037] S2. Construct a BP neural network and optimize the BP neural network using a genetic algorithm.

[0038] In this step, a BP neural network including an input layer, a hidden layer, and an output layer is built, and it is optimized using a genetic algorithm. The BP neural network has a powerful non-linear mapping ability but is prone to falling into a local optimum. The genetic algorithm has strong global search ability. The combination of the two can find the optimal parameters of the BP neural network (such as the weights and biases of each layer), which can avoid the problem of the BP neural network falling into a local optimum solution during training, making the model have higher accuracy when predicting the time-domain defect data of the core of continuous casting round billets. At the same time, the global search ability of the genetic algorithm makes the optimized BP neural network model have stronger stability and robustness when facing different time-domain ultrasonic data sets. Therefore, the model can maintain good prediction performance under different working conditions and conditions, improving the applicability and reliability of the model. In addition, the BP neural network model optimized by the genetic algorithm can better learn the internal laws and characteristics of time-domain ultrasonic data during the training process, thereby improving the generalization ability of the model.

[0039] S3. Configure sample labels for each multi-dimensional time-domain feature vector according to the core defect types of the continuous casting round billet, form a sample set with multiple multi-dimensional time-domain feature vectors and sample labels, and use the sample set to train the optimized BP neural network.

[0040] In this step, according to the defect types such as cracks, shrinkage cavities, porosity, and normal morphology of the continuous casting round billet, sample labels are configured for the multi-dimensional time-domain feature vectors to form a sample set, and the optimized BP neural network is trained with it. The sample labels clarify the corresponding relationship between the data and the defect types, and the sample set provides a large amount of supervised learning data for model training. During the training process, the BP neural network learns the time-domain feature patterns corresponding to different defect types, and thus has the ability to accurately classify unknown data.

[0041] S4. Collect the target time-domain signal of the core of the target continuous casting round billet, and input the target time-domain signal into the trained BP neural network to obtain the defect classification of the core of the target continuous casting round billet.

[0042] In the subsequent actual use process, collect the target time-domain signal of the core of the target continuous casting round billet, input it into the trained BP neural network, and obtain the defect classification result. Realize the automatic and rapid classification of the core defects of the target continuous casting round billet, replace manual inspection, which can improve the detection efficiency, and based on the optimized model and the actual time-domain signal, the classification accuracy is much higher than that of manual inspection and the prediction method based on production parameters, can classify the core defects of the continuous casting round billet in time, provide an accurate basis for the subsequent rolling process adjustment, and improve the qualified rate of rolled products.

[0043] Furthermore, as a refinement and extension of the specific implementation manner of the above embodiment, in order to completely illustrate the specific implementation process in this embodiment, another method for classifying the core defects of the continuous casting round billet is provided, and this method includes the following steps.

[0044] SS1. Data preparation and preprocessing before model training.

[0045] In this embodiment, the original time-domain signal of the core of the continuous casting round billet is used to extract waveform time-domain features, and the waveform time-domain features are used as the basis for defect classification and input into the BP neural network. Specifically, the original time-domain signal of the core of the continuous casting round billet is obtained through an ultrasonic flaw detector, the waveform statistical features are calculated from the original signal, and a multi-dimensional time-domain feature vector is constructed. And the features are Min-Max standardized to eliminate the dimension difference, and the training set and the test set are divided proportionally to ensure the independence of the time-domain data.

[0046] In some alternative implementation manners, a handheld flaw detector is used to collect data along the circumferential direction of the macro sample at intervals of 0°, 120°, and 240°, and the schematic diagram of the sampling positions is as Figure 2 shown, and the corresponding time-domain sequence waveform is generated from the CSV data file generated by the instrument through MATLAB, and the schematic diagram of the waveform is asFigure 3 As shown, the front section in the figure is the surface wave, the middle section is the defect wave, and the rear section is the bottom wave, thus constituting the training set and the test set.

[0047] A fragment of the CSV file is shown in Table 1. The first row represents the number of sampling points in the sampling depth direction of the A-scan waveform, the second row represents the acoustic time corresponding to each point, the third row represents the acoustic distance corresponding to each point, and the fourth row represents the waveform amplitude corresponding to each point.

[0048] Table 1: Example of a CSV file fragment

[0049] In some alternative embodiments, a total of 13 time-domain eigenvalue are defined, specifically including the mean value, standard deviation, skewness, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, amplitude index, waveform index, impulse index, margin index, and energy. When extracting these 13 waveform time-domain features from the original time-domain signal, these time-domain features of the original time-domain signal are specifically calculated through the waveform amplitude. The specific calculation formulas are as follows.

[0050] Mean value:

[0051] Standard deviation:

[0052] Skewness:

[0053] Kurtosis:

[0054] Maximum value:

[0055] Minimum value:

[0056] Peak-to-peak value:

[0057] Root mean square:

[0058] Amplitude index:

[0059] Waveform index:

[0060] Impulse index:

[0061] Margin index:

[0062] Energy:

[0063] Among them, is the amplitude of the sampled data points of the signal, and P is the number of sampled data points for each sample.

[0064] In this embodiment, 13 waveform time-domain features are extracted from the original time-domain signal of the core of the continuous casting round billet, which can comprehensively capture the information related to defects in the signal. The average value reflects the overall level of the signal, the standard deviation reflects the degree of data dispersion, and the skewness and kurtosis reveal the distribution characteristics of the signal. These features describe the signal from different angles, provide various input parameters for the model, classify defects based on the 13 waveform time-domain features, comprehensively reflect the defect information, and then enable the model to accurately identify the defect type and improve the classification accuracy.

[0065] SS2, neural network model construction and training.

[0066] First, data analysis is carried out, including defining the "number of samples" and the "delay step" (N historical data as independent variables), that is, using N historical data to predict the next data, and predicting across M time points, that is, running across M interval points.

[0067] Construct a data set, transform the single-row time-domain data into multi-row data, with a total of 922 rows (i.e., a total of 922 samples) and 14 columns of data. That is, use the 13 columns of data in the 922nd row to predict the 14th column of data in the 922nd row. The first 13 columns are inputs, including 13 types of feature values. The 14th column is the output, including sample labels. The sample labels are the sample labels configured for each multi-dimensional time-domain feature vector according to the defect type of the core of the continuous casting round billet. In some alternative embodiments, the defect types of the core of the continuous casting round billet include crack defects, shrinkage cavity defects, porosity defects, and normal morphology, and the corresponding sample labels include 4 categories, each category corresponding to a defect type.

[0068] Divide the training set and the test set. The first 700 rows and 13 columns are used as the input of the training set; the first 700 rows and 14 columns are used as the output of the training set; from the 701st row to the 922nd row, the 13th column is used as the input of the test set; from the 701st row to the 922nd row, the 14th column is used as the output of the test set. There are a total of 700 training set samples and 222 test set samples.

[0069] Normalize the data before training. Specifically, normalize the input and output of the training set to the range of [0, 1] to obtain the normalized data and the normalization criterion, and determine the input and output of the test set according to the normalization criterion of the training set. The normalization process eliminates the dimension difference, enables different features to be compared and analyzed on the same scale, avoids model training deviation caused by dimension problems, and enhances the stability and reliability of the model.

[0070] Construct a BP neural network model, as Figure 4As shown, in some alternative embodiments, the BP neural network includes an input layer, a hidden layer, and an output layer. The input layer includes 13 input neurons, corresponding to 13 feature inputs respectively. The hidden layer includes 8 hidden layer neurons, and the output layer includes 4 output neurons, corresponding to 4 type outputs respectively. The activation function of the BP neural network is the ReLU function, the loss function adopts cross-entropy loss, the optimizer is Adam, and the output layer adopts the Softmax function.

[0071] After that, the genetic algorithm is used to optimize the BP neural network, which specifically includes the following steps.

[0072] Step 1, set the training parameters of the BP neural network, including the maximum number of iterations, error threshold, and learning rate for the training of the BP neural network.

[0073] Specifically, set the maximum number of iterations to 1000 and the error threshold to 1e-6, that is, when the training error is lower than 1e-6, the training terminates prematurely.

[0074] Step 2, set the optimization parameters, including the number of genetic generations, population size, number of optimization parameters, and optimization variable boundaries.

[0075] The number of iterations (i.e., the number of generations of evolution) of the genetic algorithm is 50 generations, that is, the algorithm will terminate after 50 iterations; set the population size to 5, that is, each generation contains 5 individuals, and each individual represents a possible neural network parameter configuration; the number of optimization parameters specifically includes the following parts: the number of weights from the input layer to the hidden layer, the number of weights from the hidden layer to the output layer, the number of biases in the hidden layer, and the number of biases in the output layer. The total parameter S is the sum of the above four parts, that is: S = (input layer dimension × hidden layer dimension) + (hidden layer dimension × output layer dimension) + hidden layer dimension + output layer dimension; the optimization variable boundaries define the value range (boundaries) for each optimization parameter, generating a matrix of size S×2, with each row being [-1, 1], indicating that the search space for each parameter is restricted to the interval [-1, 1].

[0076] Step 3, initialize the population, including initializing the real number encoding, parameters of the selection function, parameters of the crossover function, and parameters of the mutation function.

[0077] Specifically, real - number coding includes setting the minimum precision of real - number coding to 1e - 6, which is used to control the error tolerance of floating - point operations and is applicable to continuous parameter optimization; the parameter of the selection function is 0.09, which is used to configure the parameters of the normalized geometric selection strategy; the parameter of arithmetic crossover is set to 2, representing the weight coefficient of the crossover operation; the parameter of the mutation function is set to [2gen 3]. Here, 2 represents the mutation intensity, gen represents the previously defined genetic algebra (gen = 50), which is used to control the attenuation of the mutation amplitude with the number of generations, and 3 represents the shape parameter, representing the attenuation speed of adjusting the mutation amplitude.

[0078] Step 4, run the genetic algorithm to optimize the BP neural network to obtain the optimal population.

[0079] That is, use the genetic algorithm to optimize the BP neural network model, including "fitness function", "initial population", "precision", "hyperparameters", etc., to obtain the optimal population.

[0080] Step 5, decode the optimal population to obtain the optimal parameters of the BP neural network.

[0081] Specifically, obtain the optimal parameters through the decoding operation, including parameters such as "weights from input to hidden", "biases from input to hidden", "weights from hidden to output", "biases from hidden to output", etc. Finally, assign the parameters to the network to obtain the optimal initial weights and biases.

[0082] In this embodiment, a BP neural network is constructed according to the characteristics of the continuous casting round - billet time - domain signal. The input layer includes 13 input neurons, corresponding to 13 feature inputs respectively, the hidden layer includes 8 hidden - layer neurons, and the output layer includes 4 output neurons, directly corresponding to the classification task. At the same time, the ReLU activation function, cross - entropy loss function, and Adam optimizer are adopted, as well as the Softmax function in the output layer to improve the accuracy of continuous casting round - billet defect classification. At the same time, use the genetic algorithm to optimize the BP neural network, realize searching for the optimal network parameters globally, avoid the BP neural network falling into local optimal solutions by setting training parameters, optimization parameters, and initial population adapted to continuous casting round - billet defect classification, improve the generalization ability and robustness of the model, and finally obtain the optimal initial weights and biases, further improving the accuracy and reliability of continuous casting round - billet core defect classification.

[0083] SS3, simulation test and model evaluation.

[0084] Put the training - set input and test - set input into the network to obtain the predicted values of the training set and test set, and denormalize the predicted values of the training set and test set. Compare the predicted values with the true values to obtain the root - mean - square error.

[0085] The performance of the model is evaluated using the test set. The overall defect recognition ability of the model is analyzed through the fitness change curve, the comparison curve of the prediction results of the training set, the comparison curve of the prediction results of the test set, and the scatter plot. The results shown include the fitness change curve, the comparison chart of the prediction results of the training set, the comparison chart of the prediction results of the test set, the scatter plot of the predicted values of the training set and the true values of the comparison training set, and the comparison of the scatter plot of the predicted values of the test set and the true values of the test set.

[0086] Figure 5 For the fitness change curve, the genetic optimization algorithm is used to optimize the initial weights and biases of the neural network. It can be seen that during the optimization process, the fitness value continuously decreases, and the final value decreases to between 0.012 and 0.0125. The fitness is the normalized fitness, representing the root mean square error.

[0087] Figure 6 and Figure 7 are respectively the comparison of the prediction results of the training set and the comparison of the prediction results of the test set. The blue line represents the predicted value of the data, and the red line represents the true value of the data. The root mean square errors of the training set and the test set are 0.0082194 and 0.0092604 respectively. It can be seen from the figure that the fitting of the two colored lines is relatively ideal, indicating good prediction results.

[0088] Figure 8 is the scatter plot of the comparison between the predicted values of the training set and the true values of the training set, Figure 9 is the scatter plot of the comparison between the predicted values of the test set and the true values of the test set. When the predicted values are completely accurate, all data points will be closely distributed on the line Y = X; if there is a deviation between the predicted values and the true values, the data points will deviate from the line Y = X, and the greater the degree of deviation, the greater the prediction deviation. By observing the degree of deviation of the data points from the diagonal line, it can be evaluated that this method has high prediction accuracy.

[0089] SS4, the trained model is applied to the prediction of the core defects of the continuous casting round billet in actual production.

[0090] The target time-domain signal of the core of the target continuous casting round billet is collected, and the target time-domain signal is input into the trained BP neural network to obtain the defect classification of the core of the target continuous casting round billet. The production process is adjusted according to the model output to improve the production quality.

[0091] In some alternative embodiments, according to the feedback of the actual application effect, the latest data is collected, and the model parameters or training data are further adjusted to improve the prediction accuracy and robustness.

[0092] In the above text, the embodiments of a method for classifying the core defects of a continuous casting round billet are described in detail. Based on the method for classifying the core defects of a continuous casting round billet described in the above embodiments, the embodiments of the present invention also provide a system for classifying the core defects of a continuous casting round billet corresponding to this method.

[0093] Figure 10 This is a schematic block diagram of the structure of a continuous casting round billet core defect classification system provided by an embodiment of the present invention. In this embodiment, the continuous casting round billet core defect classification system 1000 can be divided into multiple functional modules according to the functions it performs. The module referred to in the present invention means a series of computer program segments that can be executed by at least one processor and can complete fixed functions, and are stored in the memory.

[0094] The original feature collection module 1010 is used to collect multiple groups of original time-domain signals of the continuous casting round billet core, extract multiple waveform time-domain features from the original time-domain signals, and construct multiple multi-dimensional time-domain feature vectors.

[0095] The classification network model optimization module 1020 is used to construct a BP neural network and optimize the BP neural network using a genetic algorithm.

[0096] The classification network model training module 1030 is used to configure sample labels for each multi-dimensional time-domain feature vector according to the continuous casting round billet core defect type, form a sample set with multiple multi-dimensional time-domain feature vectors and sample labels, and use the sample set to train the optimized BP neural network.

[0097] The defect classification module 1040 is used to collect the target time-domain signal of the target continuous casting round billet core, input the target time-domain signal into the trained BP neural network to obtain the defect classification of the target continuous casting round billet core.

[0098] In an optional implementation manner, the original feature collection module 1010 extracts multiple waveform time-domain features from the original time-domain signal, specifically including: calculating the average value, standard deviation, skewness, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, amplitude index, waveform index, impact index, margin index, and energy of the original time-domain signal through the waveform amplitude of the original time-domain signal.

[0099] In an optional implementation manner, the BP neural network constructed by the classification network model optimization module 1020 includes an input layer, a hidden layer, and an output layer. The input layer includes 13 input neurons, the hidden layer includes 8 hidden layer neurons, and the output layer includes 4 output neurons; the activation function of the BP neural network is the ReLU function, the loss function adopts cross-entropy loss, the optimizer is Adam, and the output layer adopts the Softmax function.

[0100] In an alternative embodiment, the classification network model optimization module 1020 optimizes the BP neural network using a genetic algorithm, specifically including: setting the training parameters of the BP neural network, including the maximum number of iterations, error threshold, and learning rate for the BP neural network training; setting the optimization parameters, including the number of genetic generations, population size, number of optimization parameters, and optimization variable boundaries; initializing the population, including initializing the real number encoding, parameters of the selection function, parameters of the crossover function, and parameters of the mutation function; running the genetic algorithm to optimize the BP neural network to obtain the optimal population; and decoding the optimal population to obtain the optimal parameters of the BP neural network.

[0101] In an alternative embodiment, the optimal parameters of the BP neural network include the weights from input to hidden, biases from input to hidden, weights from hidden to output, and biases from hidden to output.

[0102] In an alternative embodiment, the types of core defects in continuous casting round billets include crack defects, shrinkage cavity defects, porosity defects, and normal morphology.

[0103] The core defect classification system for continuous casting round billets in this embodiment is used to implement the foregoing core defect classification method for continuous casting round billets. Therefore, the specific implementation manners in this system can be seen in the embodiment part of the core defect classification method for continuous casting round billets in the foregoing text. Therefore, its specific implementation manners can be referred to the descriptions of the corresponding individual part embodiments and will not be elaborated here.

[0104] In addition, since the core defect classification system for continuous casting round billets in this embodiment is used to implement the foregoing core defect classification method for continuous casting round billets, its functions correspond to those of the above method and will not be elaborated here.

[0105] Figure 11 The structure diagram of a terminal 1100 provided by an embodiment of the present invention includes: a processor 1110, a memory 1120, and a communication unit 1130. When the processor 1110 implements the core defect classification program for continuous casting round billets saved in the memory 1120, the following steps are implemented: Collect multiple groups of original time-domain signals of the core of continuous casting round billets, extract multiple waveform time-domain features from the original time-domain signals, and construct multiple multi-dimensional time-domain feature vectors; Construct a BP neural network and optimize the BP neural network using a genetic algorithm; Configure sample labels for each multi-dimensional time-domain feature vector according to the types of core defects in continuous casting round billets, form a sample set with multiple multi-dimensional time-domain feature vectors and sample labels, and use the sample set to train the optimized BP neural network; Collect the target time-domain signal of the core of the target continuous casting round billet, and input the target time-domain signal into the trained BP neural network to obtain the defect classification of the core of the target continuous casting round billet. The terminal 1100 includes a processor 1110, a memory 1120, and a communication unit 1130. These components communicate via one or more buses. Those skilled in the art can understand that the structure of the server shown in the figure does not limit the present invention. It can be a bus structure, a star structure, and can also include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0106] Among them, the memory 1120 can be used to store the execution instructions of the processor 1110. The memory 1120 can be implemented by any type of volatile or non-volatile storage terminal or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disc. When the execution instructions in the memory 1120 are executed by the processor 1110, the terminal 1100 can execute some or all of the steps in the above method embodiments.

[0107] The processor 1110 is the control center of the storage terminal, connecting various parts of the entire electronic terminal through various interfaces and lines. By running or executing software programs and / or modules stored in the memory 1120, and calling the data stored in the memory, it executes various functions of the electronic terminal and / or processes data. The processor can be composed of an integrated circuit (IC). For example, it can be composed of a single-packaged IC, or composed of multiple packaged ICs with the same or different functions connected. For example, the processor 1110 can only include a central processing unit (CPU). In the embodiment of the present invention, the CPU can be a single arithmetic core or include multiple arithmetic cores.

[0108] The communication unit 1130 is used to establish a communication channel so that the storage terminal can communicate with other terminals. Receive user data sent by other terminals or send user data to other terminals.

[0109] The present invention also provides a computer storage medium. The storage medium here can be a magnetic disk, an optical disc, a read-only memory (ROM), a random access memory (RAM), etc.

[0110] The computer storage medium stores a continuous casting round billet core defect classification program. When the continuous casting round billet core defect classification program is executed by the processor, the following steps are implemented: Collect multiple groups of original time-domain signals of the core of continuous casting round billets, extract multiple waveform time-domain features from the original time-domain signals, and construct multiple multi-dimensional time-domain feature vectors; Construct a BP neural network and optimize the BP neural network using a genetic algorithm; Configure sample labels for each multi-dimensional time-domain feature vector according to the core defect type of the continuous casting round billet, form a sample set with multiple multi-dimensional time-domain feature vectors and sample labels, and use the sample set to train the optimized BP neural network; Collect the target time-domain signal of the core of the target continuous casting round billet, and input the target time-domain signal into the trained BP neural network to obtain the defect classification of the core of the target continuous casting round billet.

[0111] Those skilled in the art can clearly understand that the technologies in the embodiments of the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solutions in the embodiments of the present invention, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc, etc., various media that can store program codes, including several instructions for causing a computer terminal (which can be a personal computer, a server, or a second terminal, a network terminal, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention.

[0112] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in an electrical, mechanical, or other form.

[0113] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0114] In addition, in each embodiment of the present invention, each functional unit may be integrated into one processing unit, may exist separately as individual physical units, or two or more units may be integrated into one unit.

[0115] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A classification method for core defects of continuous casting round billets, characterized in that, It includes the following steps: Collect multiple groups of original time-domain signals of the core of continuous casting round billets, extract multiple waveform time-domain features from the original time-domain signals, and construct multiple multi-dimensional time-domain feature vectors; Construct a BP neural network and optimize the BP neural network using a genetic algorithm; Configure sample labels for each multi-dimensional time-domain feature vector according to the core defect types of continuous casting round billets, form a sample set with multiple multi-dimensional time-domain feature vectors and sample labels, and use the sample set to train the optimized BP neural network; Collect the target time-domain signal of the core of the target continuous casting round billet, and input the target time-domain signal into the trained BP neural network to obtain the defect classification of the core of the target continuous casting round billet.

2. The method for classifying core defects of continuously cast round billets according to claim 1, characterized in that Extract multiple waveform time-domain features from the original time-domain signal, specifically including: Calculate the average value, standard deviation, skewness, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, amplitude index, waveform index, impact index, margin index, and energy of the original time-domain signal through the waveform amplitude of the original time-domain signal.

3. The method for classifying core defects of continuously cast round billets according to claim 2, characterized in that, The constructed BP neural network includes an input layer, a hidden layer, and an output layer. The input layer includes 13 input neurons, the hidden layer includes 8 hidden layer neurons, and the output layer includes 4 output neurons; the activation function of the BP neural network is the ReLU function, the loss function uses cross-entropy loss, the optimizer is Adam, and the output layer uses the Softmax function.

4. The method for classifying core defects of continuous casting round billets according to claim 3, characterized in that Use a genetic algorithm to optimize the BP neural network, specifically including: Set the training parameters of the BP neural network, including the maximum number of iterations, error threshold, and learning rate for BP neural network training; Set the optimization parameters, including the number of genetic generations, population size, number of optimization parameters, and optimization variable boundaries; Initialize the population, including initializing the real number coding, parameters of the selection function, parameters of the crossover function, and parameters of the mutation function; Run the genetic algorithm to optimize the BP neural network to obtain the optimal population; Decode the optimal population to obtain the optimal parameters of the BP neural network.

5. The method for classifying core defects of continuously cast round billets according to claim 4, characterized in that, The optimal parameters of the BP neural network include the weight from input to hidden, bias from input to hidden, weight from hidden to output, and bias from hidden to output.

6. The method for classifying core defects of continuously cast round billets according to claim 5, characterized in that, The core defect types of continuous casting round billets include crack defects, shrinkage cavity defects, porosity defects, and normal morphology.

7. A classification system for core defects of continuously cast round billets, characterized in that, It includes: An original feature collection module for collecting multiple groups of original time-domain signals of the core of continuous casting round billets, extracting multiple waveform time-domain features from the original time-domain signals, and constructing multiple multi-dimensional time-domain feature vectors; A classification network model optimization module for constructing a BP neural network and optimizing the BP neural network using a genetic algorithm; A classification network model training module for configuring sample labels for each multi-dimensional time-domain feature vector according to the core defect types of continuous casting round billets, forming a sample set with multiple multi-dimensional time-domain feature vectors and sample labels, and using the sample set to train the optimized BP neural network; A defect classification module for collecting the target time-domain signal of the core of the target continuous casting round billet, and inputting the target time-domain signal into the trained BP neural network to obtain the defect classification of the core of the target continuous casting round billet.

8. The continuous casting round billet core defect classification system according to claim 7, characterized in that Extract multiple waveform time-domain features from the original time-domain signal, specifically including: calculating the mean value, standard deviation, skewness, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, amplitude index, waveform index, impact index, margin index, and energy of the original time-domain signal through the waveform amplitude of the original time-domain signal.

9. A terminal, characterized in that, Including: A memory for storing a continuous casting round billet core defect classification program; A processor for implementing the steps of the continuous casting round billet core defect classification method as described in any one of claims 1 to 6 when executing the continuous casting round billet core defect classification program.

10. A computer-readable storage medium, characterized in that, The continuous casting round billet core defect classification program is stored on the readable storage medium, and when the continuous casting round billet core defect classification program is executed by the processor, the steps of the continuous casting round billet core defect classification method as described in any one of claims 1 to 6 are implemented.

Images