A method and device for detecting the quality of a wide and thick steel plate based on data analysis and optimization

Abstract

The application discloses a kind of based on data analysis and optimization thick steel plate quality detection method and device. It is related to quality detection technical field, wherein the method comprises: obtaining the real-time production process data of the wide thick steel plate to be detected, the design standard data of producing the wide thick plate to be detected;Real-time production process data is predicted using the preset point prediction model corresponding to different mechanical property indexes, to obtain the point prediction value of each mechanical property index corresponding to real-time production process data;Based on each point prediction value, the real-time production process data is conformally predicted using a preset mechanical property interval prediction model, to obtain the prediction interval of each mechanical property index corresponding to real-time production process data;Based on the prediction interval corresponding to real-time production process data and the design standard data, the quality of the wide thick steel plate to be detected is detected, to obtain the quality detection result of the wide thick steel plate to be detected.The method of the application can improve the quality detection efficiency and improve the quality detection accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of quality detection, and particularly relates to a wide and thick steel plate quality detection method and device based on data analysis and optimization. BACKGROUND

[0002] Wide and thick plates, as a kind of key steel material, are widely used in shipbuilding, bridge construction, building structure, pressure vessel, energy pipeline and mechanical manufacturing fields. The high strength, good toughness and plasticity and other mechanical properties of the wide and thick plates make them play an irreplaceable role under heavy load, complex environment and high stress conditions. Since the wide and thick plate products are widely used in key infrastructure and high-end equipment fields, their service environment is usually accompanied by complex mechanical action and harsh external conditions. Therefore, in order to ensure the safety, reliability and long-term effectiveness of the products in actual application, the downstream enterprises put forward more strict requirements on the stability and consistency of the quality of the wide and thick plates to meet the diversified performance requirements and high-standard industry specifications. At the same time, with the continuous widening of the application field of the wide and thick plate products, the customer demand is more and more personalized and diversified, and the wide and thick plate production enterprises are all changing from the large-batch standardized production mode to the large-scale customized production mode. However, the production process of the wide and thick plate is a typical multi-stage production process, covering key processes such as steelmaking, continuous casting, multi-pass rolling and heat treatment, and the production process route is complex. The large-scale customized production mode easily leads to frequent switching of the production route, increases the error risk in the production process, and the product quality fluctuation becomes larger. At the same time, the small-batch and multi-variety production process makes it difficult for the unified quality control system to cover all the customized products, and higher requirements are put forward for the process control and product quality determination. Therefore, accurate and efficient online determination of the quality of the wide and thick plate products is a key link to realize the process optimization control and stable delivery of the products.

[0003] At present, the mechanical property quality of wide and thick plate products mainly depends on the sampling inspection method, that is, sampling the representative plate, obtaining its mechanical property indexes through laboratory testing and chemical testing to infer the quality level of the same batch of products. However, the traditional laboratory detection method has the disadvantages of high cost and lag, which delays the delivery capacity, and on the other hand, the small size of the sampling sample is difficult to represent the mechanical properties of the full size space of the large size wide and thick plate, and cannot reflect the volatility in the production process. From the physical and metallurgical principle, the microstructure of the wide and thick plate product directly determines the mechanical properties of the wide and thick plate product, and the microstructure is mainly determined by the composition ratio and rolling process. On this basis, researchers have used expert knowledge and data-driven methods to predict the mechanical properties of wide and thick plates. The expert knowledge method establishes a rule base of the domain knowledge, that is, the long-term accumulated correlation rules between material composition, process and performance, designs a specific reasoning mechanism, and infers the mechanical properties of the metal material. However, in the face of large-scale customized multi-specification products, there are problems such as insufficient knowledge acquisition and unstable reasoning results. The data-driven method trains a regression model on a large number of historical inspection data, uses machine learning methods to describe the mapping relationship between composition-process-structure-performance, and realizes the regression modeling of the mechanical properties of the wide and thick plate. However, for the wide and thick plate products with uneven mechanical property distribution, the existing data-driven method has the following problems: only a single point prediction value is provided, and the credibility of the prediction value and the quantification of the model uncertainty are lacking, and if the point prediction result is directly used as the quality judgment standard, it is easy to misjudge the unqualified products; such model often lacks the guidance of physical mechanism, and its generalization ability and interpretability are limited, and it is difficult to effectively capture the internal change law of the mechanical properties of the material under complex processing conditions; the existing wide and thick plate mechanical property data-driven model is usually based on the fixed mapping relationship constructed offline, and lacks online adaptive ability, and cannot update the model in real time to reflect the new mechanical behavior characteristics when the dynamic environment or complex process conditions change. Therefore, although the existing technology can realize the prediction of the mechanical properties of the wide and thick plate product, it cannot provide the uncertainty information of the model prediction to reduce the risk of quality judgment, and the model lacks online adaptive mechanism, and it is difficult to realize real-time correction and prediction, and it does not have the ability to cope with the frequent changes of actual production process conditions under large-scale customization, and cannot directly participate in the quality judgment process of the wide and thick plate product. The quality guarantee in the delivery process of the wide and thick plate product still depends on the sampling inspection method based on laboratory detection. SUMMARY

[0004] Therefore, the present application provides a wide and thick steel plate quality detection method and device based on data analysis and optimization, mainly to solve the problem of inaccurate wide and thick steel plate quality detection.

[0005] To solve the above problems, the application provides a wide and thick steel plate quality detection method based on data analysis and optimization, comprising:

[0006] Obtaining real-time production process data of a wide and thick steel plate to be detected and design standard data of producing the wide and thick plate to be detected;

[0007] Using a preset point prediction model corresponding to different mechanical property indexes to predict the real-time production process data, to obtain point prediction values of each of the mechanical property indexes corresponding to the real-time production process data;

[0008] Using a preset mechanical property interval prediction model based on each of the point prediction values to conformally predict the real-time production process data, to obtain a prediction interval of each of the mechanical property indexes corresponding to the real-time production process data;

[0009] Based on the prediction interval corresponding to the real-time production process data and the design standard data, performing quality detection on the wide and thick steel plate to be detected, to obtain a quality detection result of the wide and thick steel plate to be detected.

[0010] Optionally, before using a preset point prediction model corresponding to different mechanical property indexes to predict the real-time production process data, the method further comprises: constructing a preset point prediction model;

[0011] The construction of the preset point prediction model specifically comprises:

[0012] Obtaining historical experimental detection data of the wide and thick steel plate to be detected;

[0013] Performing feature extraction on the historical experimental detection data, to obtain feature data;

[0014] Using an Arrhenius rate equation to convert temperature feature data in the feature data, to obtain a target feature data set;

[0015] Labeling the target feature data set, to obtain a label data set;

[0016] Using a Lasso method to construct a model using the label data set as a training sample, to obtain a preset point prediction model corresponding to different mechanical property indexes;

[0017] Among them, each of the mechanical property indexes includes a yield strength index, a tensile strength index, an elongation rate index, and an average impact work index.

[0018] Optionally, using a Lasso method to construct a model using the label data set as a training sample, to obtain a preset point prediction model corresponding to different mechanical property indexes, specifically comprises:

[0019] The label data set is divided into sample sets according to a preset proportion, to obtain a training set, a calibration set and a test set;

[0020] The first label data in the training set is normalized to obtain a normalized first feature data set;

[0021] Based on the first feature data set and the label data set, the Lasso method is used for model training from different mechanical performance index dimensions, to obtain the preset point prediction model corresponding to each mechanical performance index.

[0022] Optionally, the method further comprises:

[0023] Obtaining experimental detection data of the wide-thick steel plate to be detected in a preset time length before the current time;

[0024] Based on the experimental detection data as a training sample, the Lasso method is used for model training again to obtain a current point prediction model corresponding to different mechanical performance indexes;

[0025] Based on the current point prediction model, the preset point prediction model is updated.

[0026] Optionally, before the real-time production process data is conformally predicted by the preset mechanical performance interval prediction model based on each point prediction value, the method further comprises: constructing a preset mechanical performance interval prediction model.

[0027] The preset mechanical performance interval prediction model is constructed, specifically comprising:

[0028] The second label data in the calibration set is normalized to obtain normalized second feature data;

[0029] The preset point prediction model is used to predict the second feature data to obtain a first point prediction value of different mechanical performance indexes corresponding to the second label data;

[0030] Based on each first point prediction value and a true value labeled by the second label data, calculation processing is performed to obtain a first non-consistency score of different mechanical performance indexes corresponding to the second label data;

[0031] Based on each first non-consistency score and a predetermined error coverage level, calculation processing is performed to obtain a first empirical quantile of different mechanical performance indexes corresponding to the second label data;

[0032] Based on each first empirical quantile and each first point prediction value, a model is constructed to obtain the preset mechanical performance interval prediction model.

[0033] Optionally, the method further comprises:

[0034] Based on the experimental detection data and the point prediction value corresponding to the experimental detection data obtained by using the preset point prediction model, a second non-consistency score corresponding to different mechanical performance indexes is obtained by calculation and processing;

[0035] Based on the second non-consistency score, a second empirical quantile at the current time is obtained by calculation and processing using a preset objective function;

[0036] Based on the second empirical quantile, a third empirical quantile at the next time is obtained by calculation and processing using a sub-gradient descent optimization method;

[0037] Based on the third empirical quantile, the preset mechanical performance interval prediction model is updated.

[0038] Optionally, the method further comprises:

[0039] The third label data in the test set is normalized to obtain normalized third feature data;

[0040] The preset point prediction model is used to predict the third feature data to obtain a second point prediction value of different mechanical performance indexes corresponding to the third label data;

[0041] Based on the second point prediction value, the updated preset mechanical performance interval prediction model is used for conformal prediction to obtain a prediction interval of each mechanical performance index corresponding to the third feature data;

[0042] Based on the second point prediction value corresponding to each mechanical performance index, the true value corresponding to each mechanical performance index, and the prediction interval corresponding to each mechanical performance index, an interval coverage rate index and an interval average width index are obtained by calculation and processing;

[0043] Based on the interval coverage rate index and the interval average width index, the preset mechanical performance interval prediction model is tested and adjusted to obtain the preset mechanical performance interval prediction model that meets the preset accuracy requirement.

[0044] Optionally, the method further comprises:

[0045] The real-time production process data is normalized to obtain normalized production process feature data;

[0046] Based on the production process characteristic data, the preset point prediction model corresponding to different mechanical property indexes is used for prediction, to obtain point prediction values of each mechanical property index corresponding to the real-time production process data.

[0047] Optionally, based on the prediction interval corresponding to the real-time production process data and the design standard data, quality detection is performed on the wide and thick steel plate to be detected, to obtain a quality detection result of the wide and thick steel plate to be detected, specifically including:

[0048] Based on the prediction interval corresponding to the real-time production process data, the upper limit value and the lower limit value of the prediction interval of each different mechanical property index are extracted.

[0049] The upper limit value and the lower limit value of the prediction interval of the same mechanical property index are compared with the design standard data to obtain a comparison result.

[0050] When the lower limit value of the average impact work mechanical property index is greater than or equal to the minimum standard impact work, the lower limit value of the yield strength mechanical property index is greater than or equal to the minimum standard yield strength, the lower limit value of the tensile strength performance index is greater than or equal to the minimum standard tensile strength, and the upper limit value of the tensile strength performance index is less than or equal to the maximum standard tensile strength, and the lower limit value of the elongation mechanical property index is greater than or equal to the minimum standard elongation, it is determined that the wide and thick steel plate to be detected is qualified.

[0051] To solve the above problems, the application provides a wide and thick steel plate quality detection device based on data analysis and optimization, comprising:

[0052] An acquisition module is configured to acquire real-time production process data of a wide and thick steel plate to be detected and design standard data of the wide and thick steel plate to be detected.

[0053] A point prediction module is configured to use a preset point prediction model corresponding to different mechanical property indexes to predict the real-time production process data, to obtain point prediction values of each mechanical property index corresponding to the real-time production process data.

[0054] A conformal prediction module is configured to use a preset mechanical property interval prediction model to perform conformal prediction on the real-time production process data based on the point prediction values, to obtain a prediction interval of each mechanical property index corresponding to the real-time production process data.

[0055] A quality detection module is configured to perform quality detection on the wide and thick steel plate to be detected based on the prediction interval corresponding to the real-time production process data and the design standard data, to obtain a quality detection result of the wide and thick steel plate to be detected.

[0056] The beneficial effects in the present application: the present application provides the confidence interval of the mechanical properties of the wide and thick plate product under a given confidence level by integrating the design standards of the wide and thick plate product and the production and manufacturing process variables in the same database, by constructing an online real-time quality judgment system, realizing the dynamic comparison of the design standards and the predicted performance of the wide and thick plate product, accelerating the product quality judgment efficiency, and reducing the possibility of product quality misjudgment.

[0057] The above description is only a summary of the technical solutions of the present application, in order to enable the technical means of the present application to be more clearly understood, and to be implemented according to the content of the specification, and in order to enable the above and other purposes, features and advantages of the present application to be more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0058] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments and are not meant to limit the present application. Moreover, the same reference numerals in the attached drawings indicate the same or similar components. In the drawings:

[0059] Figure 1 A flowchart of a wide and thick steel plate quality detection method based on data analysis and optimization provided by an embodiment of the present application is shown;

[0060] Figure 2 A flowchart of a wide and thick steel plate quality detection method based on data analysis and optimization provided by an embodiment of the present application is shown;

[0061] Figure 3 Point prediction and interval prediction results are shown, taking the mechanical performance index of tensile strength as an example, provided by an embodiment of the present application;

[0062] Figure 4 A comparison result diagram of the prediction interval of the mechanical performance index of the wide and thick plate product and the design standard is shown, provided by an embodiment of the present application;

[0063] Figure 5 A structural block diagram of a wide and thick steel plate quality detection device based on data analysis and optimization provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0064] The various schemes and features of the present application are described herein with reference to the accompanying drawings.

[0065] It should be understood that various modifications can be made to the embodiments of the present application. Therefore, the above description should not be considered as limiting, but only as an example of the embodiments. Those skilled in the art will think of other modifications within the scope and spirit of the present application.

[0066] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the application and, together with the general description of the application given above, and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0067] These and other characteristics of the present application will become apparent from the following description of the preferred forms of the application given, by way of non-limiting example, with reference to the accompanying drawings.

[0068] It should also be understood that, although the present application has been described in relation to certain specific examples, many other equivalents forms of which will be apparent to those skilled in the art.

[0069] The above and other aspects, features and advantages of the present application will become more apparent from the following detailed description, when taken in conjunction with the accompanying drawings, in which:

[0070] Specific embodiments of the present application are described hereinafter, by way of non-limiting example; however, it should be understood that the claimed embodiments are merely examples of the present application, which can be implemented in numerous ways. Well-known and / or redundant functions and structures have not been described in detail to avoid obscuring the present application unnecessarily. Therefore, specific structural and functional details disclosed herein are not intended to limit the claimed embodiments, but merely to set forth representative embodiments of the present application.

[0071] The specification can use phrases such as "in one embodiment", "in another embodiment", "in yet another embodiment", or "in other embodiments", which can refer to one or more of the same or different embodiments of the application.

[0072] The embodiments of the present application provide a wide and thick steel plate quality detection method based on data analysis and optimization, as shown in Figure 1 The method comprises the following steps:

[0073] Step S101: acquiring real-time production process data of a wide and thick steel plate to be detected and design standard data of producing the wide and thick steel plate to be detected;

[0074] In the specific implementation process, the real-time production process data includes wide and thick steel full-process production and manufacturing data, including process variable data in processes such as steelmaking, rolling, and heat treatment; and the design standard data includes data such as steel grade, internal steel grade, steel plate thickness, maximum and minimum values of weight percentages of alloying elements in continuous casting billets, maximum and minimum values of carbon equivalent, minimum value of first controlled rolling point thickness, target value of first controlled rolling point temperature, target value of open cooling temperature, target value of final cooling temperature, minimum value of impact energy, minimum value of yield strength, minimum value of tensile strength, maximum value of tensile strength, and minimum value of elongation.

[0075] Step S102: predicting the real-time production process data by using a preset point prediction model corresponding to different mechanical property indexes to obtain point prediction values of each of the mechanical property indexes corresponding to the real-time production process data;

[0076] In the specific implementation process, the real-time production process data is normalized to obtain normalized production process feature data; and the preset point prediction model corresponding to different mechanical property indexes is used to predict based on the production process feature data to obtain point prediction values of each of the mechanical property indexes corresponding to the real-time production process data. The mechanical property indexes include yield strength, tensile strength, elongation, and average impact work, etc.

[0077] Step S103: conformally predicting the real-time production process data based on each of the point prediction values by using a preset mechanical property interval prediction model to obtain a prediction interval of each of the mechanical property indexes corresponding to the real-time production process data;

[0078] In the specific implementation process, a preset mechanical property interval prediction model of a current time is obtained; and the real-time production process data is conformally predicted based on each of the point prediction values by using a preset mechanical property interval prediction model to obtain a prediction interval of each of the mechanical property indexes corresponding to the real-time production process data.

[0079] Step S104: quality detecting the wide and thick steel plate to be detected based on the prediction interval corresponding to the real-time production process data and the design standard data to obtain a quality detection result of the wide and thick steel plate to be detected.

[0080] In the specific implementation process, the upper limit value and the lower limit value of the prediction interval of each of the different mechanical property indexes are extracted based on the prediction interval corresponding to the real-time production process data; the upper limit value and the lower limit value of the prediction interval of the same mechanical property index are compared with the design standard data to obtain a comparison result; when the lower limit value of the average impact work mechanical property index is greater than or equal to the minimum standard impact work, the lower limit value of the yield strength mechanical property index is greater than or equal to the minimum standard yield strength, the lower limit value of the tensile strength performance index is greater than or equal to the minimum standard tensile strength, and the upper limit value of the tensile strength performance index is less than or equal to the maximum standard tensile strength, and the lower limit value of the elongation mechanical property index is greater than or equal to the minimum standard elongation, it is determined that the wide and thick steel plate to be detected is qualified. Otherwise, it is determined that the wide and thick steel plate to be detected is qualified.

[0081] The application provides a confidence interval of mechanical properties of a wide and thick plate product under a given confidence level by integrating design standards of the wide and thick plate product and production process variables in a same database, constructing an online real-time quality judgment system, realizing dynamic comparison of the design standards of the wide and thick plate product and predicted properties, accelerating product quality judgment efficiency, and reducing the possibility of product quality misjudgment.

[0082] Another embodiment of the application provides another wide and thick steel plate quality detection method based on data analysis and optimization, which comprises the following steps as shown in the figure: Figure 2

[0083] Step S201: constructing a wide and thick plate full-process production manufacturing database and a process design database;

[0084] In the specific implementation process, the construction of the wide and thick plate full-process production manufacturing database comprises the following steps:

[0085] Step one: determining the collection variable field and data type according to the wide and thick plate full-process production process, including full-process production process data collected in real time and production process experimental detection data representing the steel plate obtained through experimental detection. The process variable data in the steelmaking, rolling and heat treatment processes are included. The process variables in the steelmaking process include but are not limited to continuous casting blank number, furnace number, continuous casting blank width, continuous casting blank thickness, continuous casting blank length, weight percentage (%) of alloying elements in the continuous casting blank and carbon equivalent (%), wherein the alloying elements include but are not limited to carbon, silicon, manganese, phosphorus, sulfur, chromium, nickel, copper, molybdenum, vanadium, aluminum, titanium and niobium; the process variables in the rolling process include but are not limited to steel grade, internal steel type, steel plate number, steel plate width, steel plate length, steel plate thickness, continuous casting blank pre-furnace temperature, slab first heating section temperature, slab second heating section temperature, slab soaking temperature, slab casting temperature, controlled rolling code, rough rolling opening thickness, rough rolling opening temperature, rough rolling final rolling temperature, fine rolling opening temperature, fine rolling final rolling temperature, cooling opening temperature, final cooling temperature, pre-heat straightening temperature, post-heat straightening temperature and the like; the process variables in the heat treatment process include but are not limited to heat treatment furnace number, heat treatment material number, heat treatment furnace holding temperature, heat treatment furnace holding time, plate temperature, water quenching temperature, quenching speed, tempering temperature and the like; the performance variables in the detection process include but are not limited to tensile test batch number, elongation, sample width, sample thickness, tensile strength, yield strength, impact test batch number, impact test temperature, average impact energy, impact energy single value 1, impact energy single value 2, impact energy single value 3;

[0086] ​Step two, determine the process design standard field and data type according to the full-process production process of wide-thick plate, including but not limited to steel grade, internal steel grade, plate thickness, maximum and minimum values of alloy element weight percentage in continuous casting billet and maximum and minimum values of carbon equivalent, minimum value of first controlled rolling point thickness, target value of first controlled rolling point temperature, target value of open cooling temperature, target value of final cooling temperature, minimum value of impact energy, minimum value of yield strength, minimum value of tensile strength, maximum value of tensile strength, minimum value of elongation, etc.

[0087] Step three, determine the process variable field and data type for each process in step one, and design a process data table for each process, including a slab composition table, a slab rolling table, a steel plate heat treatment table, a steel plate tensile property table, and a steel plate impact property table; establish an entity-relationship model according to the production relationship, and define the main entities in each process table, including furnace number, plate number, heat treatment material number, tensile test batch number, and impact test batch number, etc.

[0088] Step four, design a process standard table structure for the process standard field and data type determined in step two.

[0089] Step five, connect with the database of the manufacturing execution system (MES), collect corresponding field data from the MES database through direct database connection, and realize automatic data collection and update through a data synchronization mechanism: specifically, full-quantity collection is used for initial collection, all historical data is collected in batches according to the time stamp and data integrity is sampled randomly for inspection; after the historical data collection is completed, from the current time τ, configure the scheduling tool Crontab to perform incremental collection regularly, realize the online synchronization update mechanism of the local database and the MES system database, and sample randomly for data integrity inspection;

[0090] Step six, realize data alignment and integration, determine the logical relationship between the main bodies in the data table according to the coupling and correlation between the production processes, design the full-process production and manufacturing according to the order of slab information-composition-rolling process-heat treatment process-detection information, and integrate each process data table designed in step 1.3 to the steel plate product level.

[0091] Step seven, clean and pretreat the data, clean and correct the data quality problems (such as missing values, abnormal values, and noise data) during collection, and assign a uniform time stamp to all records to ensure the time sequence consistency and traceability of the data.

[0092] Step S202: extracting real-time production process data of the wide-thick steel plate to be detected from the production and manufacturing database and extracting design standard data of the wide-thick plate to be detected from the process design database.

[0093] In the implementation process, the real-time production process data includes wide-thick steel full-process production and manufacturing data, including process variable data in processes such as steelmaking, rolling, heat treatment, and the like; and the design standard data includes data such as steel grade, internal steel grade, steel plate thickness, maximum and minimum values of weight percentages of alloying elements in continuous casting billets, maximum and minimum values of carbon equivalent, minimum value of first controlled rolling point thickness, target value of first controlled rolling point temperature, target value of open cooling temperature, target value of final cooling temperature, minimum value of impact energy, minimum value of yield strength, minimum value of tensile strength, maximum value of tensile strength, minimum value of elongation, and the like.

[0094] Step S203: constructing a preset point prediction model corresponding to different mechanical property indexes;

[0095] In the implementation process, the historical experimental detection data H = {(X, Y)} of the wide-thick steel plate to be detected is obtained; feature extraction is performed on the historical experimental detection data to obtain feature data N represents the number of historical data samples, x i ∈R d represents feature data composed of all process variables in step S101, d represents the number of features; y i ∈R m represents label data composed of all mechanical property indexes, m = 4 represents the number of labels; different mechanical property indexes include indexes such as yield strength, tensile strength, elongation, and average impact energy; the temperature feature data in the feature data is converted by using an Arrhenius rate equation to obtain a target feature data set; the mathematical expression of the Arrhenius rate equation can be shown in the following formula (1):

[0096]

[0097] wherein k is a process rate, A is a constant, Q is an activation energy, R is a gas constant, and T is an absolute temperature. The Arrhenius rate equation is an equation proposed by the Swedish chemist Svante Arrhenius in 1889 to describe the relationship between the rate of a chemical reaction and temperature. The above formula shows that the temperature and the process rate are in logarithmic relationship, and the temperature usually appears in the logarithmic form in the mechanism empirical formula. At the same time, in the high-temperature zone, the influence of temperature on the performance index tends to be flat, and after taking the logarithm, the high-temperature and low-temperature zone data can be processed more uniformly, and the weight imbalance can be avoided. Therefore, the feature variables related to temperature in x i are selected for logarithmic transformation, to avoid confusion, the transformed data is still denoted as x irepresenting; labeling the target feature data set to obtain a label data set; using the label data set as a training sample to construct a model by using a Lasso method (Least Absolute Shrinkage and Selection Operator), to obtain a preset point prediction model corresponding to different mechanical performance indicators; wherein each of the mechanical performance indicators includes a yield strength indicator, a tensile strength indicator, an elongation rate indicator, and an average impact work indicator. Specifically, the label data set is divided into a training set H train = {(X train ,Y train )}, a calibration set H cal = {(X cal ,Y cal )}, and a test set H test = {(X test ,Y test )} according to a preset proportion; the historical data set can be divided into the training set, the calibration set, and the test set according to a proportion of 5:4:1; the first label data in the training set is normalized to obtain a normalized first feature data set; a calculation formula for MinMax normalization processing of the obtained training set data can be shown in the following formula (2):

[0098]

[0099] wherein, represents normalized data, n train is the number of training set samples; X min ∈R d represents a minimum value vector of data, wherein each element represents the minimum value of the jth feature; X max ∈R d represents a maximum value vector of data, wherein each element represents the maximum value of the jth feature; based on the first feature data set X′ train and the label data set y train = {y i ,i = 1,…,m}, model training is performed from different mechanical performance indicator dimensions by using the Lasso method, to obtain the preset point prediction model corresponding to each of the mechanical performance indicators. The objective function of model training by using the Lasso method can be shown in the following formula (3):

[0100]

[0101] wherein, f i represents a point prediction model of the ith label, Y trainiThe label data representing the i-th label, i = 1, …, m respectively correspond to m mechanical property indexes of the wide and thick plate product; β i ∈R d is the regression coefficient of the i-th model, λ i is a regularization coefficient for controlling the sparsity of the regression coefficient. The Lasso point prediction model uses L1 norm as regularization, realizes automatic feature selection, increases the interpretability of the model, and narrows down the scope for subsequent root cause analysis of product quality anomalies.

[0102] Step S204: using the preset point prediction model corresponding to different mechanical property indexes to predict the real-time production process data, to obtain the point prediction value of each mechanical property index corresponding to the real-time production process data;

[0103] In the specific implementation process, the real-time production process data is normalized to obtain normalized production process feature data; based on the production process feature data, the preset point prediction model corresponding to different mechanical property indexes is used for prediction to obtain the point prediction value of each mechanical property index corresponding to the real-time production process data.

[0104] In the process of continuously detecting the wide and thick steel plate to be detected, the experimental detection data of the wide and thick steel plate to be detected in a preset time length before the current time is obtained; based on the experimental detection data as a training sample, the Lasso method is used to retrain the model to obtain the current point prediction model corresponding to different mechanical property indexes; and the preset point prediction model is updated based on the current point prediction model. Regularly updating the model can improve the prediction accuracy and adapt to data changes to reduce cumulative errors.

[0105] Step S205: constructing a preset mechanical property interval prediction model;

[0106] In the specific implementation process, the second label data in the calibration set is normalized to obtain normalized second feature data; the mathematical expression for normalizing the second label data in the calibration set can be represented by the following formula (4):

[0107]

[0108] wherein, indicates the normalized validation set data; n cal indicates the number of calibration set samples. The second feature data is predicted using the preset point prediction model to obtain the first point prediction value of different mechanical property indexes corresponding to the second label data; the mathematical expression can be shown in the following formula (5):

[0109]

[0110] wherein, denotes a vector composed of the first point prediction values of the i-th label in the calibration set;

[0111] Based on each of the first point prediction values and the true value labeled by the second label data, a calculation process is performed to obtain a first inconsistency score of different mechanical performance indicators corresponding to the second label data. The mathematical expression of the first inconsistency score can be shown in the following formula (6):

[0112]

[0113] wherein, a i denotes the inconsistency score vector of the Lasso point prediction model for the i-th label in the calibration set. Based on each of the first inconsistency scores and a predetermined false coverage level, a calculation process is performed to obtain a first empirical quantile of different mechanical performance indicators corresponding to the second label data. The mathematical expression of the first empirical quantile can be shown in the following formula (7):

[0114]

[0115] wherein, s i denotes the first empirical quantile of the first inconsistency score vector for the i-th label in the calibration set. Based on each of the first empirical quantiles and each of the first point prediction values, a model is constructed to obtain the preset mechanical performance interval prediction model. The mathematical expression of the interval prediction model can be shown in the following formula (8):

[0116]

[0117] wherein, denotes the prediction interval of the i-th label of the interval prediction model at the test point.

[0118] After the preset mechanical property interval prediction model is constructed, the test set data is used to test the preset mechanical property interval prediction model. Specifically, the third label data in the test set is normalized to obtain normalized third feature data. The preset point prediction model is used to predict the third feature data to obtain second point prediction values of different mechanical property indicators corresponding to the third label data. Based on the second point prediction values, the updated preset mechanical property interval prediction model is used for conformal prediction to obtain prediction intervals of each mechanical property indicator corresponding to the third feature data. Based on the second point prediction values corresponding to each mechanical property indicator, the true values corresponding to each mechanical property indicator, and the prediction intervals corresponding to each mechanical property indicator, interval coverage rate indicators and interval average width indicators are calculated. Based on the interval coverage rate indicators and the interval average width indicators, the preset mechanical property interval prediction model is tested and adjusted to obtain the preset mechanical property interval prediction model meeting the preset accuracy requirement.

[0119] Step S206: Based on the point prediction values, a preset mechanical property interval prediction model is used to conformally predict the real-time production process data to obtain prediction intervals of each mechanical property indicator corresponding to the real-time production process data.

[0120] In the specific implementation process of this step, based on the point prediction values, a preset mechanical property interval prediction model is used to conformally predict the real-time production process data to obtain prediction intervals of each mechanical property indicator corresponding to the real-time production process data.

[0121] In the process of continuously detecting the to-be-detected wide-thick steel plate, based on the experimental detection data and the point prediction values corresponding to the experimental detection data predicted by the preset point prediction model, a second non-uniformity score corresponding to each mechanical property indicator is calculated. The mathematical expression of the second non-uniformity score can be shown in the following formula (9):

[0122]

[0123] Wherein, represents the second non-uniformity score of the i-th label of the current data point, i.e. the absolute error of the i-th label. Based on the second non-uniformity score, a preset objective function is used for calculation and processing to obtain the second empirical quantile at the current time The mathematical expression of the preset objective function can be shown in the following formula (10):

[0124]

[0125] The preset target function can be shown as formula (11) by a pinball loss function:

[0126]

[0127] The loss function is a convex function, and the second differential equation thereof is shown as formula (12):

[0128]

[0129] Based on the second empirical quantile, a sub-gradient descent optimization method is used for calculation and processing to obtain a third empirical quantile at the next moment; the mathematical expression can be shown as formula (13):

[0130]

[0131] wherein, represents the length of the prediction interval radius of the i-th label at the next moment; η i represents the learning rate corresponding to the i-th label, used to control the step size when updating the parameters each time. Based on the third empirical quantile, the preset mechanical property interval prediction model is updated. The prediction interval radius is adaptively updated to realize online calibration of the conformal prediction model.

[0132] Step S207: based on the prediction interval corresponding to the real-time production process data and the design standard data, the quality of the wide and thick steel plate to be detected is detected to obtain the quality detection result of the wide and thick steel plate to be detected.

[0133] In the specific implementation process, based on the prediction interval corresponding to the real-time production process data, the upper limit value and the lower limit value of the prediction interval of each different mechanical property index are extracted;

[0134] The upper limit of the prediction interval and the lower limit of the prediction interval of the same mechanical property index are compared with the design standard data to obtain a comparison result; when the lower limit of the prediction interval of the average impact work mechanical property index is greater than or equal to the minimum standard impact work, the lower limit of the prediction interval of the yield strength mechanical property index is greater than or equal to the minimum standard yield strength, the lower limit of the prediction interval of the tensile strength performance index is greater than or equal to the minimum standard tensile strength, and the upper limit of the prediction interval of the tensile strength performance index is less than or equal to the maximum standard impact tensile strength, and the lower limit of the prediction interval of the elongation mechanical property index is greater than or equal to the minimum standard elongation, it is determined that the quality of the wide-thick steel plate to be detected is qualified. For the wide-thick plate product whose quality is determined to be unqualified, the process data and key process node standards of the data point are read according to the feature attributes automatically selected by the Lasso point prediction model in step S204, and reason analysis is realized. According to the steel grade, internal steel grade and steel plate thickness of the current wide-thick plate product, the quality design standards of the same specification product are searched in the wide-thick plate process design library, represented by IMPACT_ENERGY_MIN, YS_MIN, TS_MIN, TS_MAX and EL_MIN, corresponding to the minimum impact work, the minimum yield strength, the minimum tensile strength, the maximum tensile strength and the minimum elongation respectively; the upper limit and the lower limit of each performance index obtained in step S206 are compared with the product quality design standards, and the product quality grade is automatically determined according to the comparison result, and the quality of each wide-thick plate is automatically determined according to the preset grading rule.

[0135] The application pre-constructs a wide-thick plate full-process production and manufacturing database and a process design database, ensures the time sequence consistency and traceability of the data through data alignment and integration, cleaning and preprocessing and other data processing methods; the preset point prediction model corresponding to different mechanical property indexes can be used to predict the real-time production process data collected in real time, obtain the point prediction values of the yield strength, elongation, tensile strength, average impact work and other mechanical property parameters corresponding to the real-time production process data, and perform conformal prediction on each point prediction value through the constructed preset mechanical property interval prediction model to obtain the prediction interval of each mechanical property index corresponding to the real-time production process data; the shortcomings of the traditional point prediction model that cannot quantify the uncertainty of the model itself are effectively avoided, and the confidence level can be adjusted according to the requirements to adapt to the quality inspection requirements of different wide-thick plate products. The online calibration problem of conformal prediction is modeled as a convex optimization problem in the application, and the data distribution changes are dynamically adapted through online optimization method, which can effectively improve the accuracy and stability of model prediction.

[0136] The wide-thick steel plate quality detection method of the application will be explained and described below in combination with specific application scenarios:

[0137] The data analysis and optimization-based wide and thick steel plate quality detection method of the embodiment of the application is applied to a wide and thick plate manufacturing plant of a certain steel plant for automatic product quality determination. For four types of mechanical property indexes of yield strength, tensile strength, elongation, and average impact energy, data from January 2023 to May 2024 are selected as historical data to train a Lasso point prediction model and an interval prediction model. Online quality determination and interval prediction model adaptive updating are performed from June 1, 2024. As of August 31, 2024, RMSE and R 2 The Lasso point prediction model is evaluated, and the interval coverage rate and the interval average width are used to evaluate the conformal interval prediction model. The model performance is calculated through representative plate data as shown in Table 1:

[0138] Table 1: Results of Lasso point prediction model and conformal prediction model

[0139] Tensile strength Yield strength Elongation Average impact energy RMSE 14.37 19.77 1.35 28.33 [R 2 ]] 0.905 0.837 0.775 0.61 Interval coverage 92.16％ 92.01％ 90.32％ 91.79％ Interval width 46.01 66.19 4.58 98.26

[0140] From Table 1, it can be seen that the interval coverage rate of the conformal prediction model always remains above the set value. Taking the tensile strength performance index as an example, the point prediction and interval prediction results are as shown in Figure 3 It can be seen that the point prediction curve is very close to the true value curve, and the true value curve falls into the prediction interval with a high probability, proving that the model established in the example has high prediction accuracy. The comparison chart of the prediction interval of the wide and thick plate product mechanical property index and the design standard is as shown in Figure 4 It can be seen that product quality determination through the prediction interval is more robust, proving the effectiveness of the method proposed in the application.

[0141] Another embodiment of the application provides a data analysis and optimization-based wide and thick steel plate quality detection device, as shown in Figure 5 , which comprises:

[0142] The acquisition module 1 is configured to acquire real-time production process data of a wide and thick steel plate to be detected and design standard data of the wide and thick plate to be detected.

[0143] The point prediction module 2 is configured to predict the real-time production process data by using a preset point prediction model corresponding to different mechanical property indexes, to obtain point prediction values of each mechanical property index corresponding to the real-time production process data.

[0144] The conformal prediction module 3 is configured to perform conformal prediction on the real-time production process data by using a preset mechanical property interval prediction model based on each point prediction value, to obtain a prediction interval of each mechanical property index corresponding to the real-time production process data.

[0145] The quality detection module 4 is configured to perform quality detection on the wide and thick steel plate to be detected based on the prediction interval corresponding to the real-time production process data and the design standard data, and obtain a quality detection result of the wide and thick steel plate to be detected.

[0146] In specific implementation, the device further comprises a preset point prediction model construction module, which is specifically configured to: acquire historical experimental detection data of the wide and thick steel plate to be detected; perform feature extraction on the historical experimental detection data to obtain feature data; convert temperature feature data in the feature data into target feature data set by using an Arrhenius rate equation; label the target feature data set to obtain a label data set; and use the label data set as a training sample to construct a model by using a Lasso method, to obtain a preset point prediction model corresponding to different mechanical performance indexes. Each of the mechanical performance indexes includes a yield strength index, a tensile strength index, an elongation index, and an average impact work index.

[0147] In specific implementation, the preset point prediction model construction module is further configured to: divide the label data set into a training set, a calibration set, and a test set according to a preset proportion; perform normalization processing on first label data in the training set to obtain a first feature data set after normalization; and perform model training from different mechanical performance index dimensions by using a Lasso method based on the first feature data set and the label data set, to obtain the preset point prediction model corresponding to each of the mechanical performance indexes.

[0148] In specific implementation, the device further comprises a first updating module, which is specifically configured to: acquire experimental detection data of the wide and thick steel plate to be detected in a preset time length before a current time; perform model training again by using a Lasso method based on the experimental detection data as a training sample, to obtain a current point prediction model corresponding to different mechanical performance indexes; and update the preset point prediction model based on the current point prediction model.

[0149] In the specific implementation process, the device further comprises a preset mechanical property interval prediction model construction module, which is specifically configured to normalize the second label data in the calibration set to obtain normalized second feature data; predict the second feature data by using the preset point prediction model to obtain first point prediction values of different mechanical property indicators corresponding to the second label data; perform calculation processing based on the first point prediction values and true values labeled by the second label data to obtain first inconsistency scores of different mechanical property indicators corresponding to the second label data; perform calculation processing based on the first inconsistency scores and a predetermined error coverage level to obtain first empirical quantiles of different mechanical property indicators corresponding to the second label data; and perform model construction based on the first empirical quantiles and the first point prediction values to obtain the preset mechanical property interval prediction model.

[0150] In the specific implementation process, the device further comprises a second updating module, which is specifically configured to perform calculation processing based on the experimental detection data and point prediction values corresponding to the experimental detection data obtained by using the preset point prediction model to obtain second inconsistency scores corresponding to different mechanical property indicators; perform calculation processing based on the second inconsistency scores by using a preset objective function to obtain second empirical quantiles at the current moment; perform calculation processing based on the second empirical quantiles by using a sub-gradient descent optimization method to obtain third empirical quantiles at the next moment; and update the preset mechanical property interval prediction model based on the third empirical quantiles.

[0151] In the specific implementation process, the device further comprises a model testing module, which is specifically configured to normalize third label data in the test set to obtain normalized third feature data; predict the third feature data by using the preset point prediction model to obtain second point prediction values of different mechanical property indicators corresponding to the third label data; perform conformal prediction based on the second point prediction values by using the updated preset mechanical property interval prediction model to obtain prediction intervals of the mechanical property indicators corresponding to the third feature data; perform calculation processing based on the second point prediction values corresponding to each of the mechanical property indicators, true values corresponding to each of the mechanical property indicators, and prediction intervals corresponding to each of the mechanical property indicators to obtain an interval coverage rate index and an interval average width index; and test and adjust the preset mechanical property interval prediction model based on the interval coverage rate index and the interval average width index to obtain the preset mechanical property interval prediction model meeting a preset accuracy requirement.

[0152] In the implementation process, the point prediction module 2 is specifically configured to: perform normalization processing on the real-time production process data to obtain normalized production process feature data; and perform prediction on the preset point prediction model corresponding to different mechanical property indexes based on the production process feature data to obtain point prediction values of each mechanical property index corresponding to the real-time production process data.

[0153] In the implementation process, the quality detection module 4 is specifically configured to: extract an upper limit value and a lower limit value of the prediction interval of each different mechanical property index based on the prediction interval corresponding to the real-time production process data.

[0154] The upper limit value and the lower limit value of the prediction interval of the same mechanical property index are compared with the design standard data to obtain a comparison result; when the lower limit value of the average impact work mechanical property index is greater than or equal to the minimum value of the standard impact work, the lower limit value of the yield strength mechanical property index is greater than or equal to the minimum value of the standard yield strength, the lower limit value of the tensile strength performance index is greater than or equal to the minimum value of the standard tensile strength, and the upper limit value of the tensile strength performance index is less than or equal to the maximum value of the standard tensile strength, and the lower limit value of the elongation mechanical property index is greater than or equal to the minimum value of the standard elongation, it is determined that the quality of the wide-thick steel plate to be detected is qualified.

[0155] The present application integrates the design standard of the wide-thick plate product and the production manufacturing process variables in the same database, constructs an online real-time quality determination system, provides a confidence interval of the mechanical property of the wide-thick plate product under a given confidence level, realizes dynamic comparison of the design standard and the predicted performance of the wide-thick plate product, accelerates the product quality determination efficiency, and reduces the possibility of product quality misjudgment.

[0156] The above examples are only exemplary embodiments of the present application and are not used to limit the present application, and the protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements to the present application within the spirit and protection scope of the present application, and such modifications or equivalent replacements are also regarded as falling within the protection scope of the present application.

Claims

1. A method for detecting the quality of a wide and thick steel plate based on data analysis and optimization, characterized in that, The method comprises the following steps: obtaining real-time production process data of a wide and thick steel plate to be detected and design standard data of the wide and thick plate to be detected; using preset point prediction models corresponding to different mechanical property indexes to predict the real-time production process data, to obtain point prediction values of each of the mechanical property indexes corresponding to the real-time production process data; based on each of the point prediction values, using a preset mechanical property interval prediction model to conduct conformal prediction on the real-time production process data, to obtain a prediction interval of each of the mechanical property indexes corresponding to the real-time production process data; based on the prediction interval corresponding to the real-time production process data and the design standard data, conducting quality detection on the wide and thick steel plate to be detected, to obtain a quality detection result of the wide and thick steel plate to be detected; before using the preset point prediction models corresponding to different mechanical property indexes to predict the real-time production process data, the method further comprises constructing a preset point prediction model; the construction of the preset point prediction model specifically comprises: obtaining historical experimental detection data of the wide and thick steel plate to be detected; extracting features from the historical experimental detection data, to obtain feature data; using an Arrhenius rate equation to convert temperature feature data in the feature data, to obtain a target feature data set; labeling the target feature data set, to obtain a label data set; using the label data set as a training sample to construct a model by using a Lasso method, to obtain the preset point prediction models corresponding to different mechanical property indexes; wherein each of the mechanical property indexes comprises a yield strength index, a tensile strength index, an elongation rate index, and an average impact work index; before using the preset mechanical property interval prediction model to conduct conformal prediction on the real-time production process data based on each of the point prediction values, the method further comprises constructing a preset mechanical property interval prediction model; the construction of the preset mechanical property interval prediction model specifically comprises: normalizing second label data in a calibration set, to obtain normalized second feature data; using the preset point prediction model to predict the second feature data, to obtain first point prediction values of different mechanical property indexes corresponding to the second label data; based on each of the first point prediction values and true values labeled by the second label data, performing calculation processing, to obtain first inconsistency scores of different mechanical property indexes corresponding to the second label data; based on each of the first inconsistency scores and a predetermined error coverage level, performing calculation processing, to obtain first empirical quantiles of different mechanical property indexes corresponding to the second label data; based on each of the first empirical quantiles and each of the first point prediction values, constructing a model, to obtain the preset mechanical property interval prediction model.

2. The method of claim 1, wherein, the use of the label data set as a training sample to construct a model by using a Lasso method, to obtain the preset point prediction models corresponding to different mechanical property indexes, specifically comprises: dividing the label data set into a training set, a calibration set, and a test set according to a preset proportion; Normalizing first label data in the training set to obtain a normalized first feature data set; Based on the first feature data set and the label data set, respectively from different mechanical performance index dimensions, using Lasso method for model training, obtaining the preset point prediction model corresponding to each mechanical performance index.

3. The method of claim 2, wherein, The method further comprises: Obtaining experimental detection data of the wide and thick steel plate to be detected in a preset time length before the current time; Based on the experimental detection data as a training sample, using Lasso method to retrain the model to obtain a current point prediction model corresponding to different mechanical performance indexes; Based on the current point prediction model, updating the preset point prediction model.

4. The method of claim 3, wherein, The method further comprises: Based on the experimental detection data and the point prediction value corresponding to the experimental detection data obtained by using the preset point prediction model, calculating and processing to obtain a second non-consistency score corresponding to different mechanical performance indexes; Based on the second non-consistency score, using a preset objective function to calculate and process to obtain a second empirical quantile at the current time; Based on the second empirical quantile, using a sub-gradient descent optimization method to calculate and process to obtain a third empirical quantile at the next time; Based on the third empirical quantile, updating the preset mechanical performance interval prediction model.

5. The method of claim 4, wherein, The method further comprises: Normalizing third label data in the test set to obtain normalized third feature data; Using the preset point prediction model to predict the third feature data to obtain a second point prediction value of different mechanical performance indexes corresponding to the third label data; Based on the second point prediction value, using the updated preset mechanical performance interval prediction model to perform conformal prediction to obtain a prediction interval of each mechanical performance index corresponding to the third feature data; Based on the second point prediction value corresponding to each mechanical performance index, the true value corresponding to each mechanical performance index, and the prediction interval corresponding to each mechanical performance index, calculating and processing to obtain an interval coverage rate index and an interval average width index; Based on the interval coverage rate index and the interval average width index, testing and adjusting the preset mechanical performance interval prediction model to obtain the preset mechanical performance interval prediction model meeting the preset accuracy requirement.

6. The method of claim 1, wherein, The method further comprises: Based on the real-time production process data, using the preset point prediction model corresponding to different mechanical performance indexes to predict to obtain a point prediction value of each mechanical performance index corresponding to the real-time production process data, specifically including: Normalizing the real-time production process data to obtain normalized production process feature data; 7. The method of claim 1, wherein, Based on the production process feature data, using the preset point prediction model corresponding to different mechanical performance indexes to predict to obtain a point prediction value of each mechanical performance index corresponding to the real-time production process data. Based on the prediction interval corresponding to the real-time production process data and the design standard data, performing quality detection on the wide and thick steel plate to be detected to obtain a quality detection result of the wide and thick steel plate to be detected, specifically including: extracting an upper limit value and a lower limit value of a prediction interval of each of the mechanical performance indexes based on the prediction interval corresponding to the real-time production process data; comparing the upper limit value and the lower limit value of the prediction interval of the same mechanical performance index with the design standard data to obtain a comparison result; when the lower limit value of the average impact work mechanical performance index is greater than or equal to the minimum standard impact work, the lower limit value of the yield strength mechanical performance index is greater than or equal to the minimum standard yield strength, the lower limit value of the tensile strength performance index is greater than or equal to the minimum standard tensile strength, and the upper limit value of the tensile strength performance index is less than or equal to the maximum standard tensile strength, and the lower limit value of the elongation mechanical performance index is greater than or equal to the minimum standard elongation, it is determined that the quality of the wide and thick steel plate to be detected is qualified.

8. A wide and thick steel plate quality detection device based on data analysis and optimization, characterized in that, comprising: an acquisition module configured to acquire real-time production process data of a wide and thick steel plate to be detected and design standard data of the wide and thick steel plate to be detected; a point prediction model construction module configured to acquire historical experimental detection data of the wide and thick steel plate to be detected, perform feature extraction on the historical experimental detection data to obtain feature data, and convert temperature feature data in the feature data into target feature data sets by using an Arrhenius rate equation; performing labeling on the target feature data sets to obtain a label data set, and using the label data set as a training sample to construct a model by using a Lasso method to obtain a preset point prediction model corresponding to different mechanical performance indexes; wherein each of the mechanical performance indexes includes a yield strength index, a tensile strength index, an elongation index, and an average impact work index; a point prediction module configured to predict the real-time production process data by using the preset point prediction model corresponding to different mechanical performance indexes to obtain point prediction values of each of the mechanical performance indexes corresponding to the real-time production process data; a mechanical performance interval prediction model construction module configured to perform normalization processing on second label data in a calibration set to obtain normalized second feature data, predict the second feature data by using the preset point prediction model to obtain first point prediction values of different mechanical performance indexes corresponding to the second label data, perform calculation processing based on each of the first point prediction values and true values labeled by the second label data to obtain first inconsistency scores of different mechanical performance indexes corresponding to the second label data, perform calculation processing based on each of the first inconsistency scores and a predetermined error coverage level to obtain first empirical quantiles of different mechanical performance indexes corresponding to the second label data, and construct a model based on each of the first empirical quantiles and each of the first point prediction values to obtain a preset mechanical performance interval prediction model; a conformal prediction module configured to perform conformal prediction on the real-time production process data by using the preset mechanical performance interval prediction model based on each of the point prediction values to obtain prediction intervals of each of the mechanical performance indexes corresponding to the real-time production process data. A quality detection module is configured to perform quality detection on the wide and heavy steel plate to be detected based on the prediction interval corresponding to the real-time production process data and the design standard data, and obtain a quality detection result of the wide and heavy steel plate to be detected.

Images