A material quality prediction method, device, equipment and medium

Abstract

The present application discloses a material quality prediction method, device, equipment and medium, and the present application belongs to the field of material production management technology. The method includes: obtaining a production data set of a material production equipment, inputting the production data set into a pre-constructed material quality prediction network, obtaining at least two variable sets related to the production data set and the causal relationship between each variable set; constructing a causal relationship matrix based on the at least two variable sets and the causal relationship between each variable set, and determining the quality-related variables of the produced material based on the values ​​in the causal relationship matrix; assigning time weights to the quality-related variables through a time attention mechanism, and obtaining a real-time prediction result of the quality of the produced material. Through this technical solution, the purpose of automatically detecting material quality can be achieved, and the intelligence, efficiency and reliability of material quality detection can be improved.

Description

Technical Field

[0001] The present application belongs to the technical field of material production management, and specifically relates to a material quality prediction method, device, equipment and medium. Background Art

[0002] With the development of society and economy, more and more manufacturing industries need to use a large number of materials of different types, models and suppliers as raw materials for production, and the requirements for material quality are also constantly increasing. It is very important for suppliers to test the quality of materials during the production process.

[0003] In related technologies, the main method of testing material quality is for testing personnel to use testing tools to measure and record the production data of the produced materials one by one, analyze and judge the quality of the produced materials one by one based on the recorded production data and prior information, and then control and adjust various situations in the production process.

[0004] Since existing technologies require a large amount of manpower and require inspectors to have extremely rich inspection experience, there are problems of low inspection efficiency and poor reliability of inspection results when using existing technologies for material quality inspection. Summary of the Invention

[0005] The purpose of the embodiments of the present application is to provide a material quality prediction method, device, equipment and medium, which solves the problems of high human resource consumption, low material quality detection efficiency and inaccurate detection results in the prior art. By inputting the production data set of the material production equipment into a pre-built material quality prediction network, determining the quality-related variables of the produced material, and assigning time weights to the quality-related variables, a real-time prediction result of the quality of the produced material is obtained, which can achieve the purpose of automatically detecting material quality and improve the intelligence, efficiency and reliability of material quality detection.

[0006] In a first aspect, an embodiment of the present application provides a method for predicting material quality, the method comprising:

[0007] Obtaining a production data set of a material production device, inputting the production data set into a pre-built material quality prediction network, and obtaining at least two variable sets related to the production data set and a causal relationship between the variable sets;

[0008] Constructing a causal relationship matrix based on the at least two variable sets and the causal relationship between the variable sets, and determining quality-related variables of the produced material based on values ​​in the causal relationship matrix;

[0009] A time weight is assigned to the quality-related variables through a time attention mechanism to obtain a real-time prediction result of the quality of the produced material.

[0010] Furthermore, the process of constructing the material quality prediction network includes:

[0011] Constructing a material quality prediction network model; wherein the material quality prediction network is a neural network that combines a multi-scale feature fusion attention mechanism, a gated recurrent unit, and a nonlinear Granger causality function;

[0012] The material quality prediction network model is trained using a pre-configured training data set, and a prediction result for the training data set is obtained using the material quality prediction network model; wherein the training data set includes: input variables and output variables;

[0013] Calculating residual data between the prediction result and the output variable, and adjusting the attention weight of the material quality prediction network according to the residual data;

[0014] A mean square error function is introduced as the loss function of the material quality prediction network, and the network parameters of the material quality prediction network are updated by the mean square error function until the loss value converges or the target training number is reached.

[0015] Furthermore, the material quality prediction network model is trained using a pre-configured training data set, including:

[0016] The input variables and output variables in the training data set are aggregated into a variable set, and the causal relationship between the input variables and the causal relationship between the input variables and the output variables are extracted by combining a nonlinear Granger causality function with a gated recurrent unit in the material quality prediction network model;

[0017] The material quality prediction network model is trained according to the causal relationship between the input variables and the causal relationship between the input variables and the output variables.

[0018] Furthermore, the production data set is input into a pre-built material quality prediction network to obtain at least two variable sets related to the production data set and the causal relationship between the variable sets, including:

[0019] Inputting the production data set into a pre-built material quality prediction network, and obtaining a prediction result for the production data set through the material quality prediction network; the prediction result is a variable set related to the production data set;

[0020] A residual between the prediction result of the production data set and the production data set is calculated, and a causal relationship between the variable sets is calculated based on the residual.

[0021] Furthermore, determining the quality-related variables of the produced material according to the values ​​in the causal relationship matrix includes:

[0022] Comparing the values ​​in the causal relationship matrix with a preset quality-related threshold, and updating the values ​​in the causal relationship matrix according to the comparison result to obtain an updated matrix;

[0023] Identify whether the value of the last column in the update matrix is ​​0;

[0024] If so, the variables corresponding to the non-zero elements in the update matrix are determined as indirect related variables of the quality of the produced materials;

[0025] If not, the variables corresponding to the elements with non-zero values ​​in the last column of the update matrix are determined as the variables directly related to the quality of the produced materials.

[0026] Furthermore, updating the values ​​in the causal relationship matrix according to the comparison result to obtain an updated matrix includes:

[0027] According to the comparison result, the values ​​in the causal relationship matrix that are greater than the preset quality-related threshold are retained, and the values ​​in the causal relationship matrix that are smaller than the preset quality-related threshold are set to 0.

[0028] In a second aspect, an embodiment of the present application provides a device for predicting material quality, the device comprising:

[0029] a variable set acquisition module, configured to acquire a production data set of a material production device, input the production data set into a pre-built material quality prediction network, and obtain at least two variable sets related to the production data set and a causal relationship between the variable sets;

[0030] a quality-related variable determination module, configured to construct a causal relationship matrix based on the at least two variable sets and the causal relationship between the variable sets, and determine the quality-related variables of the produced material based on the values ​​in the causal relationship matrix;

[0031] The material quality prediction module is used to assign time weights to the quality-related variables through a time attention mechanism to obtain real-time prediction results of the quality of the produced materials.

[0032] Furthermore, the material quality prediction network includes:

[0033] A model building module is used to build a material quality prediction network model; wherein the material quality prediction network is a neural network that combines a multi-scale feature fusion attention mechanism, a gated recurrent unit, and a nonlinear Granger causality function;

[0034] A model training module is used to train the material quality prediction network model using a pre-configured training data set, and obtain prediction results for the training data set through the material quality prediction network model; wherein the training data set includes: input variables and output variables;

[0035] a weight adjustment module, configured to calculate residual data between the prediction result and the output variable, and adjust the attention weight of the material quality prediction network according to the residual data;

[0036] The parameter updating module is used to introduce a mean square error function as the loss function of the material quality prediction network, and update the network parameters of the material quality prediction network by the mean square error function until the loss value converges or the target training number is reached.

[0037] In a third aspect, an embodiment of the present application provides an electronic device comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method described in the first aspect.

[0038] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0039] In a fifth aspect, an embodiment of the present application provides a chip, which includes a processor and a communication interface, the communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the method described in the first aspect.

[0040] In an embodiment of the present application, a production data set of a material production device is obtained, and the production data set is input into a pre-constructed material quality prediction network to obtain at least two variable sets related to the production data set and the causal relationship between each variable set; a causal relationship matrix is ​​constructed based on the at least two variable sets and the causal relationship between each variable set, and the quality-related variables of the produced material are determined based on the values ​​in the causal relationship matrix; time weights are assigned to the quality-related variables through a time attention mechanism to obtain a real-time prediction result of the quality of the produced material. The above-mentioned material quality prediction method solves the problems of high human resource consumption, low material quality detection efficiency and inaccurate detection results in the prior art. By inputting the production data set of the material production device into a pre-constructed material quality prediction network, determining the quality-related variables of the produced material, and assigning time weights to the quality-related variables, a real-time prediction result of the quality of the produced material is obtained, which can achieve the purpose of automatically detecting material quality and improve the intelligence, efficiency and reliability of material quality detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of the process of predicting material quality provided in Example 1 of the present application;

[0042] Figure 2 This is a schematic diagram of the construction process of the material quality prediction network provided in Example 2 of this application;

[0043] Figure 3 Schematic diagram of the process of predicting material quality provided in Example 3 of the present application;

[0044] Figure 4 A schematic diagram of the structure of the material quality prediction system provided in an embodiment of the present application;

[0045] Figure 5 This is a diagram of an actual scenario of the material quality prediction system provided in an embodiment of the present application;

[0046] Figure 6 Schematic diagram of the structure of the device for predicting material quality provided in Example 4 of the present application;

[0047] Figure 7 This is a structural diagram of the electronic device provided in Example 5 of the present application. DETAILED DESCRIPTION

[0048] To further clarify the objectives, technical solutions, and advantages of this application, specific embodiments of the present application are described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are intended only to illustrate this application and are not intended to limit it. It should also be noted that, for ease of description, the drawings only illustrate portions relevant to this application, not all of them. Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts depict the various operations (or steps) as sequential processes, many of the operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process may terminate upon completion of its operations, but may also include additional steps not shown in the accompanying drawings. The process may correspond to a method, function, procedure, subroutine, subprogram, or the like.

[0049] The following will be combined with the accompanying drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0050] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0051] The following describes in detail the material quality prediction method, device, equipment and medium provided in the embodiments of the present application through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0052] Example 1

[0053] Figure 1 This is a flow chart of the method for predicting material quality provided in Example 1 of this application. Figure 1 As shown, the specific steps include:

[0054] S101, obtaining a production data set of a material production equipment, inputting the production data set into a pre-built material quality prediction network, and obtaining at least two variable sets related to the production data set and a causal relationship between the variable sets;

[0055] First, the application scenario of this solution can be the scenario of quality inspection of produced materials, or the scenario of quality prediction of production materials during the material production process.

[0056] Based on the above usage scenarios, it can be understood that the executor of this solution can be an electronic device with data receiving and data processing capabilities, such as smart terminals such as mobile phones, tablet computers and desktop computers, and no further restrictions are made here.

[0057] The production dataset may include a set of operating parameters for material production equipment, an input dataset for the material production process, and a production output dataset. The production dataset includes parameters such as material size, weight, and shape used to characterize material quality. The material quality prediction network may be a network used to predict the quality of materials that have been produced or are in the production process based on the data in the production dataset. The at least two variable sets associated with the production dataset may be an original dataset in the production dataset and a predicted dataset corresponding to the data in the original dataset. The causal relationships between the variable sets may be causal relationships between input variable sets and between input variable sets and output variable sets in the production dataset.

[0058] In one embodiment, a production data set of a material production equipment can be obtained through data acquisition equipment such as on-site sensors, and the production data set is input into a pre-built material quality prediction network. The material quality prediction network is used to calculate at least two variable sets related to the production data set and the causal relationship between the variable sets.

[0059] In a feasible embodiment, optionally, inputting the production data set into a pre-built material quality prediction network to obtain at least two variable sets related to the production data set and the causal relationship between the variable sets includes:

[0060] Inputting the production data set into a pre-built material quality prediction network, and obtaining a prediction result for the production data set through the material quality prediction network; the prediction result is a variable set related to the production data set;

[0061] A residual between the prediction result of the production data set and the production data set is calculated, and a causal relationship between the variable sets is calculated based on the residual.

[0062] The prediction result for the production dataset can be data that predicts the production outcome of the material to be produced based on the data in the production dataset. The prediction result is a set of predicted output variables associated with the production dataset. The residual can be the difference between the predicted output variable set in the prediction result and the original output variable set in the production dataset. A causal relationship can be the interaction between one event (i.e., the "cause") and a second event (i.e., the "effect"), where the latter event is considered to be the result of the former event.

[0063] In one embodiment, the production data set can be input into a pre-built material quality prediction network, and a prediction result for the production data set can be obtained through the material quality prediction network. The residual between the prediction result of the production data set and the production data set is calculated, and the causal relationship between the variable sets is calculated based on the residual.

[0064] This solution, by inputting the production data set into a pre-built material quality prediction network to obtain the prediction result for the production data set, and calculating the residual between the prediction result of the production data set and the production data set, and then calculating the causal relationship between the variable sets based on the residual, can simplify the steps of calculating the causal relationship between the variable sets and improve the efficiency of calculating the causal relationship between the variables.

[0065] S102, constructing a causal relationship matrix based on the at least two variable sets and the causal relationship between the variable sets, and determining quality-related variables of the produced material based on values ​​in the causal relationship matrix;

[0066] The causal relationship matrix may be a Granger causality matrix. Granger causality may be a causal relationship based on prediction. Granger causality is used to determine whether Y has a causal relationship with X based on the probability distribution of X at a given moment given all information at the previous moment and the probability distribution of X at a given moment given all information except Y at the previous moment. If A is Granger-caused by B, then changes in A are one of the causes of changes in B.

[0067] In one embodiment, a causal relationship matrix may be constructed based on the at least two variable sets and the causal relationship between the variable sets, and quality-related variables of the produced material may be determined based on values ​​in the causal relationship matrix.

[0068] S103, assigning time weights to the quality-related variables through a time attention mechanism to obtain a real-time prediction result of the quality of the produced material.

[0069] The attention mechanism, a method that mimics the human visual and cognitive system, allows neural networks to focus on relevant parts of input data. By introducing the attention mechanism, neural networks can automatically learn and selectively focus on important information in the input, improving model performance and generalization. The temporal attention mechanism can be a dynamic time selection mechanism used to focus on temporal components.

[0070] In one embodiment, a time weight may be assigned to the quality-related variables through a time attention mechanism to obtain a real-time prediction result of the quality of the produced material.

[0071] The technical solution provided in the embodiment of the present application obtains a production data set of a material production equipment, inputs the production data set into a pre-built material quality prediction network, obtains at least two variable sets related to the production data set and the causal relationship between each variable set; constructs a causal relationship matrix based on the at least two variable sets and the causal relationship between each variable set, and determines the quality-related variables of the produced material based on the values ​​in the causal relationship matrix; assigns time weights to the quality-related variables through a time attention mechanism, and obtains a real-time prediction result of the quality of the produced material. Through the above-mentioned material quality prediction method, the problems of high human resource consumption, low material quality detection efficiency and inaccurate detection results in the prior art are solved. By inputting the production data set of the material production equipment into a pre-built material quality prediction network, determining the quality-related variables of the produced material, and assigning time weights to the quality-related variables, a real-time prediction result of the quality of the produced material is obtained, which can achieve the purpose of automatically detecting material quality and improve the intelligence, efficiency and reliability of material quality detection.

[0072] Example 2

[0073] Figure 2 This is a schematic diagram of the construction process of the material quality prediction network provided in Example 2 of this application. Figure 2 As shown, the specific steps include:

[0074] S201, constructing a material quality prediction network model; wherein the material quality prediction network is a neural network that combines a multi-scale feature fusion attention mechanism, a gated recurrent unit, and a nonlinear Granger causality function;

[0075] The gated recurrent unit can be a gated mechanism unit. The gated mechanism has obvious advantages in processing long sequence data and solving long-term dependency problems: the forget gate can selectively forget past information; the input gate can selectively update current information; the output gate can selectively output new information; the update gate determines how much previous state information is retained at the current time step; and the reset gate is used to control how to use past information.

[0076] In one embodiment, according to the material quality prediction requirements, a material quality prediction network model is constructed by combining a multi-scale feature fusion attention mechanism, a gated recurrent unit, and a nonlinear Granger causality function. The material quality prediction network is a neural network.

[0077] S202, training the material quality prediction network model using a pre-configured training data set, and obtaining a prediction result for the training data set using the material quality prediction network model; wherein the training data set includes: input variables and output variables;

[0078] The training data set may be a production data set used to train the material quality prediction network model. The training data set includes input variables and output variables. The input variables are material production input variables. The output variables are material production output variables.

[0079] In one embodiment, the material quality prediction network model can be trained based on a preconfigured training data set to extract the features of the input variables in the training data set, and learn the causal relationship between the input variables and the causal relationship between the input and output variables in the training data set, and obtain the prediction results for the training data set through the material quality prediction network model.

[0080] In a feasible embodiment, optionally, the training of the material quality prediction network model using a preconfigured training data set includes:

[0081] The input variables and output variables in the training data set are aggregated into a variable set, and the causal relationship between the input variables and the causal relationship between the input variables and the output variables are extracted by combining a nonlinear Granger causality function with a gated recurrent unit in the material quality prediction network model;

[0082] The material quality prediction network model is trained according to the causal relationship between the input variables and the causal relationship between the input variables and the output variables.

[0083] In one embodiment, the input variables and output variables in the training data set can be aggregated into a variable set. The causal relationships between the input variables and between the input variables and the output variables can be extracted using a nonlinear Granger causality function combined with a gated recurrent unit in the material quality prediction network model. The material quality prediction network model is trained based on the causal relationships between the input variables and between the input variables and the output variables.

[0084] This solution can improve the reliability of material quality prediction network model training by aggregating the input variables and output variables in the training data set into a variable set, extracting the variables and the causal relationship between them, and training the material quality prediction network model based on the causal relationship.

[0085] S203, calculating residual data between the prediction result and the output variable, and adjusting the attention weight of the material quality prediction network according to the residual data;

[0086] In one embodiment, the data in the prediction result can be subtracted from the data corresponding to the output variable to calculate the residual data between the prediction result and the output variable, and the attention weight of the material quality prediction network can be adjusted according to the residual data.

[0087] S204 , introducing a mean square error function as a loss function of the material quality prediction network, and updating the network parameters of the material quality prediction network by using the mean square error function until the loss value converges or the target number of training times is reached.

[0088] Among them, the loss function can be an operational function used to measure the degree of difference between the predicted value of the model and the true value. It is a non-negative real-valued function. The smaller the loss function, the better the robustness of the model.

[0089] In one embodiment, a mean square error function may be introduced as the loss function of the material quality prediction network, and the weight parameters of the material quality prediction network may be updated by the mean square error function until the loss value converges or the target number of training times is reached.

[0090] In a feasible embodiment, optionally, determining the quality-related variables of the produced material according to the values ​​in the causal relationship matrix includes:

[0091] Comparing the values ​​in the causal relationship matrix with a preset quality-related threshold, and updating the values ​​in the causal relationship matrix according to the comparison result to obtain an updated matrix;

[0092] Identify whether the value of the last column in the update matrix is ​​0;

[0093] If so, the variables corresponding to the non-zero elements in the update matrix are determined as indirect related variables of the quality of the produced materials;

[0094] If not, the variables corresponding to the elements with non-zero values ​​in the last column of the update matrix are determined as the variables directly related to the quality of the produced materials.

[0095] The preset quality-related threshold may be a preset minimum value of quality-related data.

[0096] In one embodiment, the values ​​in the causal relationship matrix can be compared with a preset quality-related threshold, and values ​​in the causal relationship matrix that are smaller than the preset quality-related threshold can be set to 0. The values ​​in the causal relationship matrix can be updated based on the comparison result to obtain an updated matrix. It is determined whether the value of the last column in the updated matrix is ​​0. If so, the variables corresponding to the non-zero elements in the remaining columns of the updated matrix are determined as indirect related variables of the quality of the produced material; if not, the variables corresponding to the non-zero elements in the last column of the updated matrix are determined as direct related variables of the quality of the produced material.

[0097] This solution updates the values ​​in the causal relationship matrix by comparing them with preset quality-related thresholds, and determines the direct and indirect related variables of the quality of the produced materials based on the values ​​in the updated matrix, thereby improving the comprehensiveness of the acquisition of material quality-related variables.

[0098] In a feasible embodiment, optionally, updating the values ​​in the causal relationship matrix according to the comparison result to obtain an updated matrix includes:

[0099] According to the comparison result, the values ​​in the causal relationship matrix that are greater than the preset quality-related threshold are retained, and the values ​​in the causal relationship matrix that are smaller than the preset quality-related threshold are set to 0.

[0100] In one embodiment, based on the comparison result of the values ​​in the causality matrix with the preset quality-related threshold, the values ​​in the causality matrix that are larger than the preset quality-related threshold can be retained, and the values ​​in the causality matrix that are smaller than the preset quality-related threshold can be set to 0.

[0101] This solution can improve the efficiency of determining directly related variables and indirectly related variables of the quality of produced materials by updating the values ​​in the causal relationship matrix according to the comparison results.

[0102] The technical solution provided in the embodiment of the present application, by constructing a material quality prediction network model, using a pre-configured training data set to train the material quality prediction network model, and introducing a mean square error function as the loss function of the material quality prediction network, can improve the reliability of the construction of the material quality prediction network, and thus improve the reliability of the quality prediction of the produced materials.

[0103] Example 3

[0104] Figure 3 This is a flow chart of the method for predicting material quality provided in Example 3 of this application. Figure 3 As shown, the specific steps include:

[0105] S301, constructing a data acquisition module to transmit and store data of the disc material production equipment through on-site sensors;

[0106] S302: Establish a digital management system to pre-process and analyze the production data of disk-shaped materials in the array warehouse, and achieve full traceability from warehousing, dispatching, reporting and equipment status;

[0107] S303, constructs a gated recurrent unit network based on multi-scale feature fusion attention, introduces nonlinear Granger causality function, and extracts causal relationships between production variables with long time lags;

[0108] S304, constructing a Granger causality enhancement module, combining the Granger causality matrix and the proposed input variable selection algorithm to effectively calculate process variables that have direct and indirect relationships with product quality;

[0109] S305, using the selected process variable set as the input variable set of the gated recurrent unit network based on multi-scale feature fusion attention to achieve the task of disk-shaped material product management and production process quality prediction.

[0110] Furthermore, the data acquisition module reads the disc material production data through data acquisition equipment such as on-site sensors, and is responsible for data transmission and storage.

[0111] Furthermore, a digital management system is built to realize centralized management of historical work data, while providing data preprocessing and data analysis functions.

[0112] Furthermore, the input variables and output variables in the training data set are aggregated into a variable set, which serves as the causal relationship between the input variables and the causal relationship between the input and output variables.

[0113] Furthermore, a gated recurrent unit network based on multi-scale feature fusion attention is constructed. First, a multi-scale feature fusion attention mechanism is introduced to increase the importance of relationships between variables by assigning weights. Then, by combining a constructed nonlinear Granger causality function with a gated recurrent unit, causal relationships between production variables can be extracted and improved. Finally, a mean square error function is introduced as the loss function of the gated recurrent unit network based on multi-scale feature fusion attention.

[0114] Furthermore, a Granger causality enhancement module is constructed. Combining the Granger causality matrix with the proposed input variable selection algorithm, this module effectively selects process variables that have direct and indirect relationships with product quality and calculates the causal relationships between them. Variables corresponding to the nonzero values ​​in the last column of the Granger causality matrix are considered direct Granger causal variables of the quality variables. Variables that have a causal relationship with the direct causes of the quality variables are considered indirect Granger causes and are selected as input variables for the quality prediction model.

[0115] Furthermore, the selected set of process variables is used as the input variable set of the gated recurrent unit network based on multi-scale feature fusion attention (direct and indirect causal relationships between variables), and a temporal attention mechanism is implemented on the input variable set to assign weights at each moment in the sequence to solve the time lag problem between process variables and quality variables, thereby performing the task of disc material product management and production process quality prediction.

[0116] This application provides a material quality prediction system. Through this intelligent industrial process quality prediction system, intelligent prediction and optimized control of quality in the industrial production process can be effectively achieved, thereby significantly improving product quality and production efficiency, and providing intelligent decision-making support for factories.

[0117] In order to achieve the above-mentioned purpose, an embodiment of the present application provides a material quality prediction system, including: constructing a data set through data collected by disc material production equipment; constructing a digital management system to perform data preprocessing and data analysis on the production data of disc materials; constructing a gated recurrent unit network based on multi-scale feature fusion attention, introducing a nonlinear Granger causality function, and extracting the causal relationship between long-time-delay production variables; constructing a Granger causality enhancement module, combining the Granger causality matrix and the proposed input variable selection algorithm to effectively calculate process variables that have direct and indirect relationships with product quality; using the selected process variable set as the input variable set of the gated recurrent unit network based on multi-scale feature fusion attention to realize the prediction of the disc material production process quality.

[0118] Based on the above scheme, the disc material production data is read through on-site sensors and other data acquisition equipment, and the data is transmitted and stored, and summarized into a data set.

[0119] Furthermore, a digital management system will be built to centrally manage historical work data and provide data preprocessing and data analysis functions.

[0120] Based on the above scheme, the input variables and output variables in the training data set are aggregated into a variable set as the causal relationship between the input variables and the causal relationship between the input and output variables.

[0121] Furthermore, the construction of a gated recurrent unit network based on multi-scale feature fusion attention introduces a nonlinear Granger causality function to extract the causal relationship between production variables with long time lags, including:

[0122] Through the multi-scale feature fusion attention mechanism, gated recurrent unit and nonlinear Granger causality function, a gated recurrent unit network based on multi-scale feature fusion attention is constructed;

[0123] Extract the features of the input variables and learn the causal relationship between the input variables and the causal relationship between the input and output variables. Then, calculate the residual between the predicted variable set and the original variable set, and adjust the attention weight to better capture important features.

[0124] The mean square error function is introduced as the loss function of the gated recurrent unit network based on multi-scale feature fusion attention, and the network weight parameters are continuously updated until the loss value converges or the target number of training times is reached;

[0125] The stochastic gradient descent algorithm and the multi-scale fusion attention mechanism are used to adjust the attention weight matrix. The relationship between the predicted variable set and the original variable set is calculated through the trained multi-scale feature fusion attention gated recurrent unit network to obtain the residual, and then the residual is used to calculate the causal relationship between the variable sets.

[0126] Furthermore, the Granger causality enhancement module is constructed, combining the Granger causality matrix and the proposed input variable selection algorithm to effectively select and calculate process variables that have direct and indirect relationships with product quality, including:

[0127] Set a threshold. Keep the values ​​of the causal relationship matrix that are larger than the threshold, and set any value smaller than the threshold to 0;

[0128] The variables corresponding to the non-zero elements in the last column of the Granger causality matrix are determined to be variables directly related to product quality;

[0129] For variables that are zero in the last column of the Granger causality matrix, the remaining columns of the Granger causality matrix can be used to determine whether there are variables that are indirectly related to the quality variable. If there is a non-zero element, and the variable corresponding to the non-zero element is not a variable directly related to product quality, it is determined to be a variable indirectly related to product quality, thereby achieving the selection of variables that are directly and indirectly related to product quality;

[0130] Furthermore, the selected process variable set is used as the input variable set (direct and indirect causal relationships) of the gated recurrent unit network based on multi-scale feature fusion attention, and a temporal attention mechanism is implemented on the input variable set to assign weights at each moment of the sequence to solve the time lag between process variables and quality variables.

[0131] Based on the above scheme, the predicted value of product quality calculated by the gated recurrent unit network based on multi-scale feature fusion attention during the production process of disc materials can be fed back to the digital management system in real time, realizing the traceability of the entire process of disc material production.

[0132] Beneficial effects of the present application: The present invention takes the production data collected by intelligent equipment such as industrial field sensors as the research object, combines the multi-scale fusion attention mechanism with the Granger causality matrix, and analyzes the time series of the collected production data through a gated recurrent unit network. The gated recurrent unit network based on multi-scale feature fusion attention extracts the causal relationship between process variables and learns the dynamic nonlinear features contained in the process data, thereby forming a more reliable causal relationship structure. Through real-time supervision and all-round traceability of the digital management system, technicians can improve the accuracy, real-time and adaptability of quality predictions, thereby optimizing the disc material production process and improving product quality and production efficiency.

[0133] Figure 4 This is a schematic diagram of the structure of the material quality prediction system provided in an embodiment of the present application.

[0134] like Figure 4 As shown, the material quality prediction system shown includes: a data acquisition module, a Granger causality enhancement module, a gated recurrent unit network based on multi-scale feature fusion attention, and a digital management system.

[0135] The data acquisition module primarily uses a wireless network to read historical data from the disc production process. The data collector is a PLC device that records processing parameters during the product production process. The data acquisition module aggregates data related to the production process into a data set.

[0136] Build a digital management system and perform pre-processing operations on historical work data, including deleting outliers and filling missing values, and then divide the data set into training set, validation set and test set in a ratio of 8:1:1.

[0137] A gated recurrent unit network based on multi-scale feature fusion and attention is constructed. Its core goal is to improve the performance of Granger causality through a multi-scale feature fusion and attention mechanism, extracting causal relationships between production variables. A Granger causality enhancement module is then proposed, combining the Granger causality matrix with an input variable selection algorithm to effectively calculate process variables that have direct and indirect relationships with product quality. Finally, the selected set of process variables is used as the input variable set of the gated recurrent unit network based on multi-scale feature fusion and attention, thereby obtaining product quality prediction results. The steps are as follows:

[0138] First, N input variables and one output variable are aggregated into a variable set to learn the causal relationship between input variables and the causal relationship between input and output variables. The formula is:

[0139] ;

[0140] in, represents a variable set of N input variables, and y represents the output variable.

[0141] The multi-scale feature fusion attention mechanism, gated recurrent unit and nonlinear Granger causality function are introduced to construct a gated recurrent unit network based on multi-scale feature fusion attention, extract the features of the input variables, and learn the causal relationship between the input variables and the causal relationship between the input and output variables. Then, the output of the gated recurrent unit is passed through the multi-layer perceptron layer to calculate the i-th prediction variable set and the i-th original variable set The residual formula is:

[0142] ;

[0143] ;

[0144] Among them, t represents the time t, represents the attention weight vector, GRU function represents the gated recurrent unit, f represents multi-scale feature fusion, represents the variable set vector, Represents the inner product of vectors. represents the i-th residual term at time t, and T represents a configurable hyperparameter.

[0145] The mean square error function is introduced as the loss function of the gated recurrent unit network based on multi-scale feature fusion, and the weight parameters of the network are continuously updated until the loss value converges or the target training times are reached. The formula is:

[0146] ;

[0147] Among them, M represents the number of input variables and N represents the number of training samples.

[0148] Set the jth element in the attention weight vector to 0 and construct a set of predictor variables that excludes the jth variable set. The attention weight vector is then used to obtain the predicted variable set and calculate the residual between it and the original variable set using the trained multi-scale feature fusion gated recurrent unit network. The formula is:

[0149] ;

[0150] ;

[0151] in, represents the adjusted attention weight vector, represents the i-th predictor variable set excluding the j-th variable set at time t, It represents the residual between the i-th set of predicted variables and the i-th set of original variables at time t.

[0152] Use the residual of the predicted variable set excluding the jth element and the original variable set and the residuals of the predicted variable set and the original variable set , calculate the j-th variable set and the i-th variable set The causal relationship is:

[0153] ;

[0154] Where N represents the number of training samples.

[0155] Construct a Granger causality matrix based on the causal relationship between the variable sets , and set a threshold , retain the most proportional Larger values, and The smallest value is set to 0. The variables corresponding to the non-zero elements in the last column of the Granger causality matrix are determined to be directly related to product quality. For variables with a zero value in the last column of the Granger causality matrix, the remaining columns of the Granger causality matrix can be used to determine whether there are variables indirectly related to the quality variable. If there is a non-zero element, and the variable corresponding to the non-zero element is not a variable directly related to product quality, it is determined to be a variable indirectly related to product quality, thereby achieving the selection of variables directly and indirectly related to product quality.

[0156] The time-delayed attention mechanism is constructed using the selected variables X related to product quality and the output variable y. A time-delayed attention weight is assigned to each input variable X as the input variable set of the gated recurrent unit network based on multi-scale feature fusion. The formula is:

[0157] ;

[0158] in, represents the time-delayed attention weight vector.

[0159] Construct a gated recurrent unit network based on multi-scale feature fusion and input variables with time-delayed attention weights into the network The output of the gated recurrent unit is aggregated through the multi-layer perceptron layer to obtain the predicted value of the product quality at time t , the mean square error is used as the loss function to update the network parameters until the loss value converges or the target number of training times is reached. Finally, a quality prediction system for the disc material production process can be obtained. The formula is:

[0160] ;

[0161] in, is the true value of product quality at time t, and N is the number of training samples.

[0162] Product quality predictions are fed back to the digital management system in real time during the disc production process, allowing technicians to adjust production conditions based on the predictions, thereby reducing production costs. Furthermore, the digital management system allows technicians to view process variables related to product quality, enabling full traceability of the entire production process, further improving transparency and controllability of the production process.

[0163] Figure 5 This is a practical scenario diagram of the material quality prediction system provided in an embodiment of the present application.

[0164] like Figure 5 As shown, the material quality prediction system includes: material production equipment, a digital acquisition module, a digital management system, and a material production process quality prediction system. The material production equipment provides the digital acquisition module to collect material production data, and the production data is input into the digital management system. The production data is processed by the network model in the digital management system, and the processing results are input into the material production process quality prediction system. The material production process quality prediction system predicts the quality data of the produced materials, and the quality data is fed back to the digital management system to update the network model parameters in the digital management system.

[0165] The technical solution provided by the embodiment of this application uses a data acquisition module to read the working data of production equipment in real time; establishes a digital management system to preprocess the data; constructs a gated recurrent unit network based on multi-scale feature fusion attention, introduces a nonlinear Granger causality function, and extracts the causal relationship between long-time-delay production variables; constructs a Granger causality enhancement module, combines the Granger causality matrix and the proposed input variable selection algorithm, and effectively calculates process variables that have direct and indirect relationships with product quality; and uses the selected process variable set as the input variable set of the gated recurrent unit network based on multi-scale feature fusion attention to predict the quality of the disc material production process. The technical route constructed by this invention has the function of real-time monitoring of product quality, meeting the needs of product quality control and product management.

[0166] Example 4

[0167] Figure 6 This is a schematic diagram of the structure of the material quality prediction device provided in Example 4 of this application. Figure 6 As shown, specifically including the following:

[0168] The variable set acquisition module 601 is used to acquire a production data set of a material production equipment, input the production data set into a pre-built material quality prediction network, and obtain at least two variable sets related to the production data set and the causal relationship between the variable sets;

[0169] a quality-related variable determination module 602, configured to construct a causal relationship matrix based on the at least two variable sets and the causal relationships between the variable sets, and determine the quality-related variables of the produced material based on the values ​​in the causal relationship matrix;

[0170] The material quality prediction module 603 is used to assign time weights to the quality-related variables through a time attention mechanism to obtain a real-time prediction result of the quality of the produced material.

[0171] Furthermore, the material quality prediction network includes:

[0172] A model building module is used to build a material quality prediction network model; wherein the material quality prediction network is a neural network that combines a multi-scale feature fusion attention mechanism, a gated recurrent unit, and a nonlinear Granger causality function;

[0173] A model training module is used to train the material quality prediction network model using a pre-configured training data set, and obtain prediction results for the training data set through the material quality prediction network model; wherein the training data set includes: input variables and output variables;

[0174] a weight adjustment module, configured to calculate residual data between the prediction result and the output variable, and adjust the attention weight of the material quality prediction network according to the residual data;

[0175] The parameter updating module is used to introduce a mean square error function as the loss function of the material quality prediction network, and update the network parameters of the material quality prediction network by the mean square error function until the loss value converges or the target training number is reached.

[0176] Furthermore, the model training module is specifically used to:

[0177] The input variables and output variables in the training data set are aggregated into a variable set, and the causal relationship between the input variables and the causal relationship between the input variables and the output variables are extracted by combining a nonlinear Granger causality function with a gated recurrent unit in the material quality prediction network model;

[0178] The material quality prediction network model is trained according to the causal relationship between the input variables and the causal relationship between the input variables and the output variables.

[0179] Furthermore, the variable set acquisition module 601 is specifically configured to:

[0180] Inputting the production data set into a pre-built material quality prediction network, and obtaining a prediction result for the production data set through the material quality prediction network; the prediction result is a variable set related to the production data set;

[0181] A residual between the prediction result of the production data set and the production data set is calculated, and a causal relationship between the variable sets is calculated based on the residual.

[0182] Furthermore, the quality-related variable determination module 602 is specifically configured to:

[0183] Comparing the values ​​in the causal relationship matrix with a preset quality-related threshold, and updating the values ​​in the causal relationship matrix according to the comparison result to obtain an updated matrix;

[0184] Identify whether the value of the last column in the update matrix is ​​0;

[0185] If so, the variables corresponding to the non-zero elements in the update matrix are determined as indirect related variables of the quality of the produced materials;

[0186] If not, the variables corresponding to the elements with non-zero values ​​in the last column of the update matrix are determined as the variables directly related to the quality of the produced materials.

[0187] Furthermore, the quality-related variable determination module 602 is specifically configured to:

[0188] According to the comparison result, the values ​​in the causal relationship matrix that are greater than the preset quality-related threshold are retained, and the values ​​in the causal relationship matrix that are smaller than the preset quality-related threshold are set to 0.

[0189] The technical solution provided by the embodiment of the present application includes a variable set acquisition module for acquiring a production data set of a material production device, inputting the production data set into a pre-built material quality prediction network, and obtaining at least two variable sets related to the production data set and the causal relationship between each variable set; a quality-related variable determination module for constructing a causal relationship matrix based on the at least two variable sets and the causal relationship between each variable set, and determining the quality-related variables of the produced material based on the values ​​in the causal relationship matrix; and a material quality prediction module for assigning time weights to the quality-related variables through a time attention mechanism to obtain a real-time prediction result of the quality of the produced material. The above-mentioned material quality prediction device solves the problems of high human resource consumption, low material quality detection efficiency and inaccurate detection results in the prior art. By inputting the production data set of the material production device into the pre-built material quality prediction network, determining the quality-related variables of the produced material, and assigning time weights to the quality-related variables, a real-time prediction result of the quality of the produced material is obtained, which can achieve the purpose of automatically detecting material quality and improve the intelligence, efficiency and reliability of material quality detection.

[0190] The material quality prediction device in the embodiments of the present application can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, the mobile electronic device can be a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc. The non-mobile electronic device can be a server, network attached storage (NAS), personal computer (PC), television (TV), ATM, or self-service machine, etc., which are not specifically limited in the embodiments of the present application.

[0191] The material quality prediction device in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0192] The material quality prediction device provided in the embodiment of the present application can implement each process implemented in the above-mentioned method embodiments. To avoid repetition, it will not be described here.

[0193] Example 5

[0194] like Figure 7 As shown, an embodiment of the present application also provides an electronic device 700, including a processor 701, a memory 702, and a program or instruction stored in the memory 702 and executable on the processor 701. When the program or instruction is executed by the processor 701, each process of the above-mentioned material quality prediction device embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0195] It should be noted that the electronic devices in the embodiments of the present application include the mobile electronic devices and non-mobile electronic devices mentioned above.

[0196] Example 6

[0197] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of the above-mentioned material quality prediction device embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0198] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer-readable storage medium, such as a computer read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0199] Example 7

[0200] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned material quality prediction device embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0201] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0202] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be noted that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0203] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of this application, or the part that contributes to the existing technology, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of this application.

[0204] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

[0205] The above are only preferred embodiments of the present application and the technical principles employed. The present application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that are possible for those skilled in the art will not depart from the scope of protection of the present application. Therefore, although the present application has been described in more detail through the above embodiments, the present application is not limited to the above embodiments and may include more other equivalent embodiments without departing from the concept of the present application. The scope of the present application is determined by the scope of the claims.

Claims

1. A method for predicting material quality, characterized in that: The method comprises: Obtaining a production dataset of a material production device, inputting the production dataset into a pre-built material quality prediction network, and obtaining at least two variable sets related to the production dataset and causal relationships between the variable sets, wherein the production dataset includes an operating parameter set of the material production device, material attribute parameters, and input parameters and output parameters in the material production process, and the material quality prediction network is a neural network that combines a multi-scale feature fusion attention mechanism, a gated recurrent unit, and a nonlinear Granger causality function; Constructing a Granger causality matrix based on the at least two variable sets and the causal relationship between the variable sets, and determining quality-related variables of the produced material based on the values ​​in the Granger causality matrix, wherein the quality-related variables include indirect related variables or direct related variables of the quality of the produced material; A time weight is assigned to the quality-related variables through a time attention mechanism to obtain a real-time prediction result of the quality of the produced material.

2. The method for predicting material quality according to claim 1, characterized in that: The process of constructing the material quality prediction network includes: Construct a material quality prediction network model; The material quality prediction network model is trained using a pre-configured training data set, and a prediction result for the training data set is obtained using the material quality prediction network model; wherein the training data set includes: input variables and output variables; Calculating residual data between the prediction result and the output variable, and adjusting the attention weight of the material quality prediction network according to the residual data; A mean square error function is introduced as the loss function of the material quality prediction network, and the network parameters of the material quality prediction network are updated by the mean square error function until the loss value converges or the target training number is reached.

3. The method for predicting material quality according to claim 2, characterized in that: The method of training the material quality prediction network model using a pre-configured training data set includes: The input variables and output variables in the training data set are aggregated into a variable set, and the causal relationship between the input variables and the causal relationship between the input variables and the output variables are extracted by combining a nonlinear Granger causality function with a gated recurrent unit in the material quality prediction network model; The material quality prediction network model is trained according to the causal relationship between the input variables and the causal relationship between the input variables and the output variables.

4. The method for predicting material quality according to claim 1, characterized in that: Inputting the production data set into a pre-built material quality prediction network to obtain at least two variable sets related to the production data set and the causal relationship between the variable sets includes: Inputting the production data set into a pre-built material quality prediction network, and obtaining a prediction result for the production data set through the material quality prediction network; the prediction result is a variable set related to the production data set; A residual between the prediction result of the production data set and the production data set is calculated, and a causal relationship between the variable sets is calculated based on the residual.

5. The method for predicting material quality according to claim 2, characterized in that: Determining the quality-related variables of the produced materials according to the values ​​in the Granger causality matrix includes: Comparing the values ​​in the Granger causality matrix with a preset quality-related threshold, and updating the values ​​in the Granger causality matrix according to the comparison result to obtain an updated matrix; Identify whether the value of the last column in the update matrix is ​​0; If so, the variables corresponding to the non-zero elements in the update matrix are determined as indirect related variables of the quality of the produced materials; If not, the variables corresponding to the elements with non-zero values ​​in the last column of the update matrix are determined as the variables directly related to the quality of the produced materials.

6. The method for predicting material quality according to claim 5, characterized in that: The updating of the values ​​in the Granger causality matrix according to the comparison result to obtain an updated matrix includes: According to the comparison result, the values ​​in the Granger causality matrix that are greater than the preset quality-related threshold are retained, and the values ​​in the Granger causality matrix that are smaller than the preset quality-related threshold are set to 0.

7. A device for predicting material quality, characterized in that: The device comprises: a variable set acquisition module, configured to acquire a production dataset of a material production device, input the production dataset into a pre-built material quality prediction network, and obtain at least two variable sets related to the production dataset and the causal relationship between the variable sets. The production dataset includes an operating parameter set of the material production device, material attribute parameters, and input and output parameters of the material production process. The material quality prediction network is a neural network that combines a multi-scale feature fusion attention mechanism, a gated recurrent unit, and a nonlinear Granger causality function. a quality-related variable determination module, configured to construct a Granger causality matrix based on the at least two variable sets and the causal relationship between the variable sets, and determine quality-related variables of the produced material based on the values ​​in the Granger causality matrix, wherein the quality-related variables include indirect related variables or direct related variables of the quality of the produced material; The material quality prediction module is used to assign time weights to the quality-related variables through a time attention mechanism to obtain real-time prediction results of the quality of the produced materials.

8. The material quality prediction device according to claim 7, characterized in that: The material quality prediction network includes: Model building module, used to build material quality prediction network model; A model training module is used to train the material quality prediction network model using a pre-configured training data set, and obtain prediction results for the training data set through the material quality prediction network model; wherein the training data set includes: input variables and output variables; a weight adjustment module, configured to calculate residual data between the prediction result and the output variable, and adjust the attention weight of the material quality prediction network according to the residual data; The parameter updating module is used to introduce a mean square error function as the loss function of the material quality prediction network, and update the network parameters of the material quality prediction network by the mean square error function until the loss value converges or the target training number is reached.

9. An electronic device, characterized in that: The method comprises a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the material quality prediction method according to any one of claims 1 to 6.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the steps of the material quality prediction method according to any one of claims 1 to 6 are implemented.

Images