A strip coil shape defect prediction and diagnosis method based on deep learning

Abstract

The application discloses a strip steel coil shape defect prediction and diagnosis method based on deep learning, comprising the following steps: collecting production data in the strip steel hot continuous rolling coiling process; calculating equipment accuracy index characteristic values and preprocessing the calculation results; constructing a strip steel coil shape quality prediction model based on a twin semi-supervised learning network, and training the model by using the preprocessed equipment accuracy index characteristic value data; sorting strip steel coil shape quality professional information, and constructing a strip steel hot continuous rolling coil shape quality knowledge base by means of knowledge graph technology; combining the strip steel coil shape quality prediction model and the strip steel hot continuous rolling coil shape quality knowledge base to perform real-time prediction and diagnosis on the to-be-tested strip steel coil shape defect, and obtaining a defect type and corresponding treatment measures. The application is based on strip steel hot continuous rolling production industrial big data, and realizes real-time judgment and auxiliary decision of the strip steel coil shape defect by means of deep learning and knowledge graph technology, so that the product quality control and production line intelligent manufacturing level can be improved, and the construction progress of a steel industry digital factory is promoted.

Description

Technical Field

[0001] This invention relates to the field of metallurgical rolling control technology, and in particular to a method for predicting and diagnosing strip coil defects based on deep learning. Background Technology

[0002] As steel manufacturing shifts from high output to high quality, customers are placing increasingly higher demands on the quality of hot-rolled products. Poor strip coil shape not only poses safety hazards during product transportation and storage but also significantly impacts the performance of downstream products. Currently, the assessment of strip defects mainly relies on manual inspection or machine identification, which fails to fully utilize on-site production data, resulting in a limited and inefficient approach. Furthermore, research on data-driven strip quality prediction and identification is relatively scarce.

[0003] Patent application CN101670372B discloses a method for eliminating shape defects in hot-rolled high-strength steel sheets. For hot-rolled high-strength steel, it combines leveling rolling, annealing in a shield, and roll straightening process into the production of hot-rolled high-strength steel coils, thereby eliminating internal stress and springback stress. This results in better sheet shape stability and significantly improved flatness. Patent application CN110197170A discloses a method for detecting and identifying coil shape defects based on target detection. It utilizes a deep convolutional neural network to label and grade coil defects, reducing human error and labor costs, quantifying evaluation criteria, and improving the automation level of production. Patent application CN110216148A discloses a method for controlling coil shape defects during hot rolling. It sends side servo valve signals based on surface depressions in the strip or wear on the pinch rolls, obtaining accurate actual deviations on both sides from a pressure control perspective. The sensor causes the hydraulic cylinder to rise, realizing a closed-loop control process that automatically or manually adjusts the deviation to ensure that the pinch rolls are relatively horizontal, making the strip move more smoothly and the coiler more stable.

[0004] Although the aforementioned existing technologies have provided specific analyses on the prevention, detection, and control of strip coil defects and formulated corresponding solutions for specific problems, there is a lack of research on real-time prediction of strip coil conditions based on field data in the current existing technologies, and there is a lack of proactive prediction of strip product quality. Summary of the Invention

[0005] This invention provides a deep learning-based method for predicting and diagnosing strip coil defects, addressing the current lack of research on real-time prediction of strip coil conditions based on field data and the absence of proactive prediction of strip product quality. It enables real-time prediction of strip coil quality and, combined with a coil knowledge base, provides corresponding solutions, thereby improving the intelligence level of hot strip rolling production lines and promoting the construction of digital factories in the steel industry.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] On the one hand, the present invention provides a deep learning-based method for predicting and diagnosing strip coil defects, the deep learning-based method for predicting and diagnosing strip coil defects includes:

[0008] Collect production data during the hot strip rolling and coiling process;

[0009] Based on the production data collected during the hot strip rolling and coiling process, the characteristic values ​​of the equipment accuracy index are calculated according to the preset equipment process evaluation standards, and the calculated characteristic values ​​of the equipment accuracy index are preprocessed.

[0010] A strip steel coil quality prediction model is constructed based on a deep learning network of a preset type, and the constructed strip steel coil quality prediction model is trained using preprocessed equipment accuracy index feature value data; wherein, the strip steel coil quality prediction model is used to reconstruct the strip steel feature vector based on the preprocessed equipment accuracy index feature value data, and combine the strip steel feature vector to predict the strip steel coil quality.

[0011] Organize professional information on strip steel coil quality and construct a knowledge base for hot strip rolling coil quality using knowledge graph technology; wherein, the professional information includes information related to coil defects, causes and diagnostic measures;

[0012] By combining a trained strip coil quality prediction model with a constructed hot strip rolling coil quality knowledge base, real-time prediction and diagnosis of strip coil defects to be detected are performed, resulting in defect types and corresponding treatment measures.

[0013] Furthermore, the production data refers to field equipment data directly related to strip coiling, including: strip length set value, strip length measured value, strip speed set value, strip speed measured value, pinch roll gap set value, pinch roll gap measured value, and the difference between the coiler tension set value and the measured value.

[0014] Furthermore, the characteristic values ​​of the equipment accuracy indicators include: the pressure difference between the head of the side guide plate, the body of the side guide plate, and the tail of the side guide plate, the deviation between the measured value and the set value of the pinch roller gap, and the plate shape quality.

[0015] Furthermore, the preprocessing of the calculated equipment accuracy index feature values ​​includes:

[0016] The calculated equipment accuracy index characteristic values ​​are checked, incomplete data in the index content are removed, and text and Boolean content are converted into floating point data. After normalization processing to remove data dimensions, the data range is compressed to the [0,1] interval to form standardized data that is easy for the strip coil quality prediction model to process.

[0017] Furthermore, the strip coil quality prediction model is a twin semi-supervised learning network.

[0018] Furthermore, the strip coil quality prediction model includes a reconstruction layer, a convolutional neural network (CNN) layer, and a prediction layer. The reconstruction layer and the CNN layer constitute a Siamese network for data reconstruction and integration, while the prediction layer is responsible for calculating the membership degree of feature information to different coil shapes based on the output of the Siamese network.

[0019] The reconstruction layer uses two identical deep autoencoders (DAEs) as a symmetric network. Gaussian noise is added to the input data before it is fed into this layer, ultimately yielding the intermediate layer feature vector and the reconstructed output vector.

[0020]

[0021] In the formula, X input Let n0 represent random Gaussian noise, and W represent the input data. i ,b i σ represents the weights and biases of each hidden layer, respectively. e H represents the hidden layer activation function. mid Output represents the intermediate layer feature vector and the reconstructed output vector, respectively. H i Let represent the feature vector of the i-th layer in the DAE network, and l represent the number of intermediate layers in the DAE network. CNN layers further extract correlation information from the feature vectors through convolution, activation, and pooling operations.

[0022]

[0023]

[0024]

[0025]

[0026] In the formula, These represent the feature maps of the (l+1)th convolutional layer. Let b represent the kernel function of the feature map convolution, b represent the convolution bias, K represent the number of feature map channels, and L represent the feature map convolution kernel function. l+1 ,L l Let f, s0, and p represent the dimensions of the feature maps of the (l+1)th and lth layers, respectively, and let f, s0, and p represent the kernel size, stride, and number of padding layers, respectively. Let (i,j) represent the feature map of the k-th channel in the l-th layer, where (i,j) represents the coordinates corresponding to the feature map values. This represents the activated feature map, using the ReLU activation function. This represents the feature map after pooling, where (x, y) represents the coordinates corresponding to the numerical values ​​of the feature map convolution kernel function.

[0027] To avoid overfitting caused by deepening the network, after obtaining the convolutional results, a residual module is constructed using skip connections to preserve DAE feature information, and the corresponding vectors are concatenated to obtain the Siamese network output:

[0028] H siamese =concate[H1 mid ,A1,,H2 mid A2]

[0029] In the formula, H1 mid H2 mid A1 and A2 represent the intermediate feature vectors obtained from the standard data and real-time data after the reconstruction layer, respectively. mid H2 mid The output vector H after the CNN layer siamese This represents the features after concatenation of symmetric networks. Here, concate[.] is used to concatenate the two features.

[0030] The prediction layer uses a fully connected classification network, and calculates the membership degree of each roll category using the softmax function.

[0031] O = σ f (MH siamese +N)

[0032] Y pred =softmax(O)

[0033] In the formula, M and N are the weight matrix and bias matrix of the fully connected layer, respectively, and the activation function σ is the weight matrix and bias matrix of the fully connected layer. f ReLU is used, and then softmax is applied to calculate the specific probability Y of the strip belonging to each coil shape. pred .

[0034] Furthermore, the process of organizing professional information on strip steel coil quality and constructing a knowledge base for hot strip rolling coil quality using knowledge graph technology includes:

[0035] Collect specialized information on the quality of strip steel coils, including professional content in the form of tables, texts, or logs related to coil defects, causes, and diagnostic measures.

[0036] Knowledge triples are extracted from structured or semi-structured data through rule matching and structure parsing, and knowledge triples are obtained from unstructured text through entity extraction and relation extraction.

[0037] The extracted knowledge triples are manually verified and then stored in the graph database to form a semantic network of strip steel coil forming process knowledge, thus obtaining a knowledge base for the quality of hot strip steel coil forming.

[0038] Furthermore, the training-prepared strip coil quality prediction model and the constructed hot strip rolling coil quality knowledge base are used to perform real-time prediction and diagnosis of the strip coil defects to be detected, obtaining the defect type and corresponding handling measures, including:

[0039] Real-time acquisition of production data during the production and coiling process of the strip steel to be inspected;

[0040] Based on the production data collected during the production and coiling process of the strip steel to be inspected, the characteristic values ​​of the equipment accuracy index are calculated according to the preset equipment process evaluation standards, and the obtained characteristic values ​​of the equipment accuracy index are preprocessed.

[0041] Using a trained strip coil shape quality prediction model, the strip coil shape is predicted in real time based on the pre-processed equipment accuracy index feature values, and the prediction results are fed back to the strip hot continuous rolling coil shape quality knowledge base.

[0042] Based on the analysis and prediction results of relevant process knowledge in the hot strip rolling coil quality knowledge base, corresponding quality analysis results are provided, along with the causes and remedial measures for the corresponding situations.

[0043] In another aspect, the present invention also provides an electronic device comprising a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-described method.

[0044] In another aspect, the present invention also provides a computer-readable storage medium storing at least one instruction that is loaded and executed by a processor to implement the above-described method.

[0045] The beneficial effects of the technical solution provided by this invention include at least the following:

[0046] The method for predicting and diagnosing strip steel coil defects provided by this invention predicts the strip steel coil shape based on a data-driven approach. It uses a semi-supervised learning network to reconstruct the strip steel feature vector and combines a CNN network and a residual network to fully explore the index factors affecting coil quality. This provides an effective solution for predicting strip steel coil quality. At the same time, it combines a strip steel coil knowledge base to conduct professional analysis of the prediction results and provides corresponding improvement measures to assist professionals in improving the reliability of production processes and ensuring smooth and efficient strip steel production. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a schematic diagram of the execution flow of the deep learning-based strip coil defect prediction and diagnosis method provided in the embodiments of the present invention;

[0049] Figure 2 This is a schematic diagram of the strip coil quality knowledge network provided in an embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0051] First Embodiment

[0052] To address the current lack of research on real-time prediction of strip steel coil shape based on on-site data and the absence of proactive prediction of strip steel product quality, and considering the lack of a systematic diagnostic network for strip steel coil defects, this embodiment provides a deep learning-based method for predicting and diagnosing strip steel coil defects. Deep learning uncovers the potential correlation between equipment process feature values ​​and strip steel coil quality, enabling real-time prediction of strip steel coil quality. Simultaneously, it integrates a strip steel coil quality knowledge base to achieve real-time judgment and decision support for strip steel product quality, significantly improving the level of intelligent manufacturing for enterprises. This method can be implemented by electronic devices, such as terminals or servers. The execution flow of this method is as follows: Figure 1 As shown, it includes the following steps:

[0053] S1, collects production data during the hot strip rolling and coiling process;

[0054] It should be noted that the production data that S1 needs to collect during the strip coiling process includes the strip length set value and measured value, the strip speed set value and measured value, the pinch roll gap set value and measured value, and the difference between the coiler tension set value and measured value, as well as other field equipment data directly related to strip coiling.

[0055] Specifically, this embodiment collects production data from 19,058 strip steel coils from a certain steel 1580 production line.

[0056] S2, based on the production data collected during the hot strip rolling and coiling process, calculate the characteristic value of the equipment accuracy index according to the preset equipment process evaluation standard, and preprocess the obtained characteristic value of the equipment accuracy index.

[0057] It should be noted that the above S2 needs to be combined with the preset strip hot rolling equipment process accuracy evaluation system and the on-site production data from a certain steel coiling process to calculate various equipment process accuracy evaluation indicators related to the coiling process, including 114 equipment process characteristic values ​​such as the pressure difference between the head, body and tail of the side guide plate, the measured and set deviation of the pinch roll gap, and the plate shape quality. Some indicator information is shown in Table 1.

[0058] Table 1. Information on Partial Accuracy Characteristics of Winding Process

[0059]

[0060] After calculating the characteristic values ​​of the equipment accuracy index, the calculated characteristic values ​​of the equipment process accuracy are checked, incomplete strip steel data in the index content are removed, and text and Boolean type content are converted into floating point data. After normalization processing to remove the data dimensions, the data range is compressed to the [0,1] interval to form standardized data that is easy to process in the subsequent strip steel coil quality prediction model.

[0061] S3. Construct a strip steel coil quality prediction model based on a deep learning network of a preset type, and train the constructed strip steel coil quality prediction model using preprocessed equipment accuracy index feature value data.

[0062] The strip coil quality prediction model is used to reconstruct the strip feature vector based on the preprocessed equipment accuracy index feature value data, and to predict the strip coil quality in combination with the strip feature vector.

[0063] Specifically, in this embodiment, the strip coil quality prediction model is a Siamese semi-supervised learning network, which includes: a reconstruction layer, a convolutional neural network (CNN) layer, and a prediction layer. The reconstruction layer and the CNN layer constitute a Siamese network for data reconstruction and integration, and the prediction layer is responsible for calculating the membership degree of feature information to different coil shapes based on the output of the Siamese network. Specifically, the calculation process of each model is as follows:

[0064] The reconstruction layer uses two identical deep autoencoders (DAEs) as symmetric networks. Gaussian noise is added to the input data before it is fed into the model, ultimately yielding the intermediate layer feature vector and the reconstructed output vector.

[0065]

[0066] In the formula, X input Let n0 represent random Gaussian noise, and W represent the input data. i ,b i σ represents the weights and biases of each hidden layer, respectively. e H represents the hidden layer activation function. midOutput represents the intermediate layer feature vector and the reconstructed output vector, respectively. H i Let represent the feature vector of the i-th layer in the DAE network, and l represent the number of intermediate layers in the DAE network;

[0067] Then, the CNN layer further extracts the correlation information in the feature vectors through convolution, activation, and pooling operations:

[0068]

[0069]

[0070]

[0071]

[0072] In the formula, These represent the feature maps of the (l+1)th convolutional layer. Let b represent the kernel function of the feature map convolution, b represent the convolution bias, K represent the number of feature map channels, and L represent the feature map convolution kernel function. l+1 ,L l Let f, s0, and p represent the dimensions of the feature maps of the (l+1)th and lth layers, respectively, and let f, s0, and p represent the kernel size, stride, and number of padding layers, respectively. Let (i,j) represent the feature map of the k-th channel in the l-th layer, where (i,j) represents the coordinates corresponding to the feature map values. This represents the activated feature map, using the ReLU activation function. This represents the feature map after pooling, where (x, y) represents the coordinates corresponding to the numerical values ​​of the feature map convolution kernel function.

[0073] To avoid overfitting caused by deepening the network, after obtaining the convolutional results, a residual module is constructed using skip connections to preserve DAE feature information, and the corresponding vectors are concatenated to obtain the Siamese network output.

[0074] H siamese =concate[H1 mid ,A1,,H2 mid A2]

[0075] In the formula, H1 mid H2 mid A1 and A2 represent the intermediate feature vectors obtained from the standard data and real-time data after the reconstruction layer, respectively. mid H2 mid The output vector H after the CNN layer siamese This represents the features after concatenation of symmetric networks. Here, concate[.] is used to concatenate the two features.

[0076] Finally, the prediction layer uses a fully connected classification network, and the softmax function is used to calculate the membership degree of each roll category:

[0077] O = σ f (MH siamese +N)

[0078] Y pred =softmax(O)

[0079] In the formula, M and N are the weight matrix and bias matrix of the fully connected layer, respectively, and the activation function σ is the weight matrix and bias matrix of the fully connected layer. f ReLU is used, and then softmax is applied to calculate the specific probability Y of the strip belonging to each coil shape. pred .

[0080] After the model was built, the process feature data of the strip steel coiling equipment were divided into training set and test set in a ratio of 7:3. The prediction effect of some coil shapes of the final model is shown in Table 2. As can be seen from Table 2, the model can basically accurately and efficiently predict the quality of common coil shapes based on strip steel production data.

[0081] Table 2 Predicted Results for Different Roll Shapes

[0082] category Accuracy Recall rate F measurement Support normal 93.49％ 93.33％ 93.41％ 600 Folding Print 88.84％ 93.56％ 91.14％ 621 Head and tail intertwined 87.32％ 93.30％ 90.21％ 1314 Outer ring overflow 82.44％ 77.59％ 79.94％ 696 Tower type 84.33％ 67.05％ 74.70％ 610 Interlaced throughout 91.46％ 96.62％ 93.97％ 621 Inner ring overflow 96.40％ 97.23％ 96.81％ 1046

[0083] S4. Organize professional information on strip steel coil quality and construct a knowledge base for hot strip rolling coil quality using knowledge graph technology; the professional information includes information related to coil defects, causes, and diagnostic measures.

[0084] Specifically, in this embodiment, the implementation process of S4 is as follows: Collect relevant data on hot-rolled strip steel coil shape from the hot-rolled production site or the internet, including professional domain content in the form of tables, texts, or logs related to coil defects, causes, and diagnostic measures. Extract knowledge triples from the structured or semi-structured data using rule matching and structure parsing. Obtain knowledge triples from the unstructured text through entity extraction and relation extraction. After manual verification, store the aforementioned strip steel coil process knowledge triples into a graph database, forming a graph database as shown below. Figure 2 The shown is a semantic network of knowledge about the strip coil process.

[0085] S5 combines the trained strip coil quality prediction model and the strip hot rolling coil quality knowledge base to perform real-time prediction and diagnosis of strip coil defects to be detected, and obtain the defect type and corresponding treatment measures.

[0086] Specifically, in this embodiment, the implementation process of S5 is as follows: Real-time acquisition of production data during the coiling process of the strip steel to be tested; based on the acquired production data during the coiling process, calculation of the corresponding equipment precision index feature value according to the preset equipment process evaluation standard, and preprocessing of the obtained equipment precision index feature value; then, using the trained strip steel coil shape quality prediction model, real-time prediction of the strip steel coil shape based on the preprocessed equipment precision index feature value, and feeding the prediction result back to the hot strip steel continuous rolling coil shape quality knowledge base; finally, combining the relevant process knowledge in the hot strip steel continuous rolling coil shape quality knowledge base to analyze the prediction result, providing corresponding quality analysis results, and simultaneously providing the inducing causes and remedial measures for the corresponding situation. This improves the reliability of strip steel production and provides a new method for optimizing the hot continuous rolling coiling process.

[0087] In summary, this embodiment provides a deep learning-based method for predicting and diagnosing strip steel coil defects. By combining real-time predictions of strip steel coil conditions with on-site data, it offers a method for data-driven optimization of production processes. Furthermore, the system built using this method can specify corresponding coil quality improvement schemes based on the prediction results, thereby significantly ensuring strip steel production efficiency and quality, and improving the level of intelligent manufacturing in steel enterprises.

[0088] Second Embodiment

[0089] This embodiment provides an electronic device, which includes a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment.

[0090] The electronic device can vary considerably depending on its configuration or performance, and may include one or more processors (central processing units, CPUs) and one or more memories, wherein the memories store at least one instruction that is loaded by the processor and executed in accordance with the above method.

[0091] Third Embodiment

[0092] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc. The instruction stored therein can be loaded and executed by a processor in a terminal.

[0093] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0094] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0095] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0096] It should also be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0097] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A method for predicting and diagnosing strip coil defects based on deep learning, characterized in that, include: Collect production data during the hot strip rolling and coiling process; Based on the production data collected during the hot strip rolling and coiling process, the characteristic values ​​of the equipment accuracy index are calculated according to the preset equipment process evaluation standards, and the calculated characteristic values ​​of the equipment accuracy index are preprocessed. A strip steel coil quality prediction model is constructed based on a deep learning network of a preset type, and the constructed strip steel coil quality prediction model is trained using preprocessed equipment accuracy index feature value data; wherein, the strip steel coil quality prediction model is used to reconstruct the strip steel feature vector based on the preprocessed equipment accuracy index feature value data, and combine the strip steel feature vector to predict the strip steel coil quality. Organize professional information on strip steel coil quality and construct a knowledge base for hot strip rolling coil quality using knowledge graph technology; wherein, the professional information includes information related to coil defects, causes and diagnostic measures; By combining the trained strip coil quality prediction model and the constructed strip hot continuous rolling coil quality knowledge base, the strip coil defects to be detected are predicted and diagnosed in real time, and the defect types and corresponding treatment measures are obtained. The characteristic values ​​of the equipment accuracy indicators include: the pressure difference between the head of the side guide plate, the body of the side guide plate and the tail of the side guide plate, the deviation between the measured value and the set value of the pinch roller gap, and the plate shape quality; The strip coil quality prediction model is a twin semi-supervised learning network; The strip coil quality prediction model includes a reconstruction layer, a convolutional neural network (CNN) layer, and a prediction layer. The reconstruction layer and the CNN layer constitute a Siamese network for data reconstruction and integration, while the prediction layer is responsible for calculating the membership degree of feature information to different coil shapes based on the output of the Siamese network. The reconstruction layer uses two identical deep autoencoders (DAEs) as a symmetric network. To avoid overfitting caused by deepening the network, after obtaining the output of the CNN layer, a residual module is constructed using skip connections to preserve the DAE feature information, and the corresponding vectors are concatenated to obtain the Siamese network output. In the formula, These represent the intermediate feature vectors obtained from the standard data and real-time data after the reconstruction layer, respectively. They represent The output vector after passing through the CNN layer This represents the features of the symmetric network after splicing, and is applied here. The two features are concatenated.

2. The method for predicting and diagnosing strip coil defects based on deep learning as described in claim 1, characterized in that, The production data refers to the field equipment data directly related to strip coiling, including: strip length set value, strip length measured value, strip speed set value, strip speed measured value, pinch roll gap set value, pinch roll gap measured value, and the difference between the coiler tension set value and the measured value.

3. The method for predicting and diagnosing strip coil defects based on deep learning as described in claim 1, characterized in that, The preprocessing of the calculated equipment accuracy index feature values ​​includes: The calculated equipment accuracy index characteristic values ​​are checked, incomplete data in the index content are removed, and text and Boolean content are converted into floating point data. After normalization processing to remove data dimensions, the data range is compressed to the [0,1] interval to form standardized data that is easy for the strip coil quality prediction model to process.

4. The method for predicting and diagnosing strip coil defects based on deep learning as described in claim 1, characterized in that, The input data, after Gaussian noise is added, is fed into the reconstruction layer, ultimately yielding the intermediate layer feature vector and the reconstructed output vector: ； In the formula, Indicates input data, Represents random Gaussian noise. These represent the weights and biases of each hidden layer, respectively. This represents the activation function of the hidden layer. These represent the intermediate layer feature vector and the reconstructed output vector, respectively. Indicates the first in the DAE network i The feature vector of the layer, Indicates the number of intermediate layers in the DAE network; CNN layers further extract correlation information from feature vectors through convolution, activation, and pooling operations; ； ； ； ； In the formula, Indicates the first Layer convolutional feature maps This represents the kernel function of the feature map convolution. This indicates the convolution bias. Indicates the number of channels in the feature map. They represent the first l+1 Layer and first l The size of the layer feature map and These represent the kernel size, stride, and number of padding layers, respectively. Indicates the first l Layer k Channel feature map, This represents the coordinates corresponding to the feature map values. Representing the activated feature map, using relu Activation function This represents the feature map after pooling. This represents the coordinates corresponding to the numerical values ​​of the feature map convolution kernel function; The prediction layer uses a fully connected classification network, through The function calculates the membership degree of each volume category: ； ； In the formula, These are the weight matrix and bias matrix of the fully connected layer, and the activation function. Using ReLU, and then through Calculate the specific probability of the strip steel belonging to each coil shape. .

5. The method for predicting and diagnosing strip coil defects based on deep learning as described in claim 1, characterized in that, The aforementioned process involves compiling specialized information on the quality of hot-rolled strip coils, and constructing a knowledge base for the quality of hot-rolled strip coils using knowledge graph technology. This includes: Collect specialized information on the quality of strip steel coils, including professional content in the form of tables, texts, or logs related to coil defects, causes, and diagnostic measures. Knowledge triples are extracted from structured or semi-structured data through rule matching and structure parsing, and knowledge triples are obtained from unstructured text through entity extraction and relation extraction. The extracted knowledge triples are manually verified and then stored in the graph database to form a semantic network of strip steel coil forming process knowledge, thus obtaining a knowledge base for the quality of hot strip steel coil forming.

6. The method for predicting and diagnosing strip coil defects based on deep learning as described in claim 1, characterized in that, The method combines a trained strip coil quality prediction model with a constructed hot strip rolling coil quality knowledge base to perform real-time prediction and diagnosis of strip coil defects to be detected, obtaining defect types and corresponding handling measures, including: Real-time acquisition of production data during the production and coiling process of the strip steel to be inspected; Based on the production data collected during the production and coiling process of the strip steel to be inspected, the characteristic values ​​of the equipment accuracy index are calculated according to the preset equipment process evaluation standards, and the obtained characteristic values ​​of the equipment accuracy index are preprocessed. Using a trained strip coil shape quality prediction model, the strip coil shape is predicted in real time based on the pre-processed equipment accuracy index feature values, and the prediction results are fed back to the strip hot continuous rolling coil shape quality knowledge base. Based on the analysis and prediction results of relevant process knowledge in the hot strip rolling coil quality knowledge base, corresponding quality analysis results are provided, along with the causes and remedial measures for the corresponding situations.

Images