Explosive melting casting process temperature prediction model construction method and explosive melting casting process temperature control system

Abstract

The application discloses a kind of explosive melting casting process temperature prediction model construction methods, comprising the following steps: step one: respectively constructs two PINN neural network models;Wherein, the first PINN neural network model is used to learn the thermal boundary condition in explosive melting casting process, to obtain the best approximation in thermal boundary condition parameter space;Second PINN neural network model is used to construct proxy model to realize the fast prediction of melting casting process temperature;Step two: temperature field data in explosive melting casting process is used to train and verify proxy model;Step three: the best approximation in thermal boundary condition parameter space obtained by the first PINN neural network model is input into the proxy model obtained by training in step two, and explosive melting casting process temperature prediction model capable of obtaining temperature information of explosive melting casting process is constructed.The application further discloses an explosive melting casting process temperature control system.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of explosive melt casting, and particularly relates to a method for constructing a temperature prediction model for an explosive melt casting process and a temperature control system for the explosive melt casting process. BACKGROUND

[0002] Melt-cast explosive charges are widely filled in various weapons, ammunition and warheads. In the forming process of melt-cast charges, phenomena such as state change, volume shrinkage and heat release occur, which can easily cause quality defects such as shrinkage cavity and crack. The root cause is the lack of internal temperature regulation during the forming process. If the internal temperature field is uneven, thermal stress will be generated, and if the thermal stress is large and not fully released, cracks will occur in the explosive castings. In addition, a small temperature gradient in the inner layer during the melt casting process will also lead to the formation of loose columnar or dendritic crystal regions, which ultimately affects the quality performance and detonation performance. Therefore, whether the temperature of the explosive melt casting process can be accurately regulated is a key factor in improving the quality of material solidification forming.

[0003] Currently, the research on the temperature and temperature field of the explosive melt casting process mainly focuses on two routes. The first route is to measure the temperature distribution in the process by arranging thermocouples or thermal resistors or fiber Bragg grating sensors, and then monitor the temperature field by combining Labview software, and dynamically study the internal temperature field of the melt casting forming process and its change rule. For example, a Chinese patent with the patent number CN201310432117.2 adopts a fiber Bragg grating sensor array method, which is suitable for continuous and multi-point distributed measurement of internal temperature field changes. However, using the above experimental measurement method, only a limited number of sensors can obtain temperature data at a limited number of positions in the mold internal temperature field. The measurement results are not only sparse but also invasive, and it is difficult to master the change and distribution rule of the internal temperature field of the entire melt casting process. Moreover, there is a lack of exact mathematical model to predict the temperature of the melt casting process, and a large number of melt casting process tests need to be conducted for testing, which results in high cost, long cycle and low precision of process design.

[0004] In the second route, scholars take the explosive melting process as the object, and realize temperature field prediction and analysis by calculating the internal temperature field distribution through numerical simulation technology. For example, the Chinese patent with the patent number CN201910506622.4 discloses a modeling method for the pressurized casting process of melt-cast explosive. By using the ProCAST software, selecting the calculation model of heat conduction and setting the parameters, the modeling and calculation research of the temperature field of the pressurized casting process of melt-cast explosive in a wide pressure range can be achieved. The Chinese patent with the patent number CN201910506022.8 summarizes the core process links such as melt-cast explosive casting and solidification as heat transfer model problems. Similarly, the simulation software is used to obtain the temperature field results of the casting process of the semi-spherical melt-cast explosive. The simulation and simulation technology based on the mechanism model rely on complex physical and related theories, and are often composed of a large number of algebraic equations. The simulation process still has the shortcomings of large calculation amount and low efficiency, and cannot get rid of the problem of real-time temperature field.

[0005] In particular, the temperature control method of the traditional melt-cast process still mainly depends on manual debugging, and the process control mostly relies on the experience judgment of the "black box" type, so that the solidification process temperature is uncontrollable, which is not conducive to guiding the on-site production, and the research on temperature regulation in the melt-cast explosive forming process is basically blank. The existing research can only obtain the internal temperature field of the melt-cast process, but cannot realize the internal temperature regulation, which will inevitably cause quality or safety problems when the temperature fluctuates or the temperature of a certain point position is high, greatly increase the design and manufacturing cost, and cause the problem of waste of raw materials. Therefore, it is of great significance to seek a method to balance the modeling accuracy and calculation time of the temperature field, establish a temperature field model and apply it to the scene. With the rapid development of artificial intelligence technologies such as deep learning, new ideas and solutions are provided to solve this problem. SUMMARY

[0006] Therefore, the purpose of the present application is to provide an explosive melt-cast process temperature prediction model construction method and an explosive melt-cast process temperature control system. The explosive melt-cast process temperature prediction model can balance the modeling accuracy and calculation time of the temperature field to predict the temperature information of the explosive melt-cast process in real time. The explosive melt-cast process temperature control system can realize real-time temperature regulation of the explosive melt-cast process.

[0007] To achieve the above purpose, the present application provides the following technical solutions:

[0008] The present application first proposes an explosive melt-cast process temperature prediction model construction method, which comprises the following steps:

[0009] Step one: two PINN neural network models are respectively constructed; wherein the first PINN neural network model is used to learn the thermal boundary condition in the explosive melt casting process to obtain the best approximation in the thermal boundary condition parameter space; the second PINN neural network model is used to construct a proxy model to realize the rapid prediction of the temperature in the melt casting process;

[0010] Step two: the proxy model is trained and verified by using the temperature field data in the explosive melt casting process;

[0011] Step three: the best approximation in the thermal boundary condition parameter space obtained by the first PINN neural network model is input into the proxy model trained in step two, and an explosive melt casting process temperature prediction model capable of obtaining the temperature information of the explosive melt casting process is constructed.

[0012] Further, in the step one, the thermal boundary condition parameter space in the explosive melt casting process is represented as:

[0013] q(T,t)=h(T,t)(T-T ∞ )

[0014] Wherein, q represents the heat flux; h represents the heat transfer coefficient; T represents the material temperature in the explosive melt casting process; T ∞ represents the external environment temperature; t represents time;

[0015] The heat transfer coefficient is a function of temperature and time, and is represented as:

[0016]

[0017] Wherein, H (i) represents the heat transfer coefficient of the mold surface i; O (i) represents the direction factor of the mold surface i; i represents the decomposed mold surface, and i∈(1,2,3), respectively representing the top surface, side surface and bottom surface of the mold;

[0018] The heat transfer coefficient of each decomposed mold surface is represented by a linear combination of the basis functions of the heat transfer coefficient curve:

[0019]

[0020] Wherein, represents the interpolation coefficient of the heat transfer coefficient defined on the temperature interval k to the temperature interval N k of the mold surface i; represents the piecewise linear basis function defined on the temperature interval k to the temperature interval N k of the mold surface i; k=1,2,…,N k ;

[0021] Each mold surface orientation will affect the heat transfer process, so the direction factors of the top, side and bottom surfaces of the mold are represented as:

[0022]

[0023]

[0024]

[0025] where O (1) , O (2) and O (3) represent the direction factors of the top, side and bottom surfaces of the mold; n z represents the component of the unit outward normal vector n on the corresponding surface in the direction of gravity.

[0026] Further, the proxy model regards the temperature field in the explosive melting process as a Gaussian process, represented as:

[0027] T ~ N(μ,σ)

[0028] μ = μ(t,x,α)

[0029] σ = σ(t,x,α)

[0030] μ M ,σ M = M(t,x,α;W * ,b * )

[0031] where μ and σ represent the mean and standard deviation of the temperature field, respectively; x and t represent space and time, respectively, and constitute the space-time coordinates (x,t); α represents the parameter space of the thermal boundary condition; μ M and σ M represent the mean temperature and standard deviation, respectively; M represents the proxy model; W * and b * represent the weights and biases of the neural network when the loss function of the neural network is optimal, and are obtained by the following optimization problem:

[0032]

[0033] L(W,b) = λ data L data (W,b) + λ pde L pde (W,b) + λ bc L bc (W,b)

[0034] where L(W,b) represents the loss function, W represents the weights of the neural network, and b represents the bias of the neural network; Ldata , L pde , and L bc represent real data condition loss, partial differential structure loss and boundary condition loss respectively; λ data , λ pde , and λ bc are corresponding weights respectively.

[0035] Further, in the step two, a high-fidelity heat transfer model is created based on the finite element method to obtain the temperature field data in the explosive melt casting process, and the semi-discrete formula of the high-fidelity heat transfer model is:

[0036]

[0037] wherein, Ω and Γ represent the spatial domain and the boundary domain of the mold respectively; w and t represent the test function and the time respectively; ρ represents the density; c p represents the specific heat capacity; k represents the thermal conductivity; represents the Laplace operator; d represents the differential; q(T, t) represents the heat flux.

[0038] Further, in the step three, the temperature prediction model of the explosive melt casting process is represented as a minimization problem about the thermal boundary condition parameter:

[0039]

[0040]

[0041] wherein, α * represents the optimal value of the thermal boundary condition parameter; L p represents the loss function; α represents the parameter space of the thermal boundary condition; E represents the sensor position set; f represents the Gaussian distribution function; represents the measured temperature; represents the calculated temperature value obtained by the finite element method; represents the corresponding standard deviation of the calculation;

[0042] After obtaining α * , the surrogate model M is replaced by, and the full-field temperature prediction and its standard deviation are obtained as follows:

[0043] μ * (t, x) = μ(t, x, α * ; W * , b * )

[0044] σ * (t, x) = σ(t, x, α * ; W * , b * )

[0045] wherein, μ * and σ * respectively represent the full-field temperature prediction value and the standard deviation; x and t respectively represent the space and time, and constitute the space-time coordinate (x, t); α represents the parameter space of the thermal boundary condition; W * and b * respectively represent the weight and the bias when the loss function of the neural network is optimal.

[0046] The application further provides an explosive melting and casting process temperature control system, which comprises a cloud service layer, an edge service layer and a terminal device layer.

[0047] The terminal device layer comprises a detection module and an execution module; the detection module comprises a data acquisition unit and a quality detection unit, the data acquisition unit is used for acquiring environmental information and running information in the explosive melting and casting process, and the acquired environmental information and running information are uploaded to the edge service layer; the quality detection unit is used for quality detection on the solidified explosive; the execution module controls the running parameters in the explosive melting and casting process according to the control parameters derived by the edge service layer, so as to control the temperature of the explosive melting and casting process.

[0048] The edge service layer comprises a data processing module, an edge storage module, an explosive melting and casting process temperature prediction module and an optimization decision module; the data processing module pre-processes the environmental information and running information collected by the detection module and obtains pre-processed data; the edge storage module is used for storing the pre-processed data and uploading the pre-processed data to the cloud service layer; the explosive melting and casting process temperature prediction module is provided with an explosive melting and casting process temperature prediction model trained by the cloud service layer, the explosive melting and casting process temperature prediction model predicts the temperature of the explosive melting and casting process according to the real-time pre-processed data provided by the data processing module; the optimization decision module comprises a parameter optimization unit and a temperature control unit, the parameter optimization unit obtains the process parameter set value with the explosive melting and casting process temperature as the optimization target according to the temperature information of the explosive melting and casting process predicted by the explosive melting and casting process temperature prediction module, and the temperature control unit obtains the optimization control parameters of the execution module according to the process parameter set value and sends the optimization control parameters to the execution module.

[0049] The cloud service layer comprises a cloud storage module and a cloud temperature model training and verification module; the cloud storage module is internally provided with a historical database for storing preprocessed data uploaded by the edge storage module; the cloud temperature model training and verification module trains and verifies the explosive melting process temperature prediction model by using historical data stored in the historical database, and transmits the trained explosive melting process temperature prediction model to the explosive melting process temperature prediction module to update the explosive melting process temperature prediction model in the edge service layer;

[0050] The explosive melting process temperature prediction model is constructed by using the explosive melting process temperature prediction model construction method described above.

[0051] Further, the cloud service layer further comprises an external resource device, which dynamically allocates cloud computing resources of the cloud service layer according to the amount and uploading speed of data uploaded by the edge, in combination with the real-time processing capacity of the edge service layer and the cloud service layer, so as to realize the fusion of cross-level interaction of edge computing data and cloud computing data, guarantee the real-time operation of the equipment control optimization decision module of the edge service layer, and realize efficient processing of edge-cloud cooperation.

[0052] Further, the terminal device layer is further provided with an equipment communication security module, the edge service layer is further provided with an edge communication security module, and the cloud service layer is further provided with a cloud communication security module; the edge communication security module is in communication connection with the equipment communication security module and the cloud communication security module, so as to guarantee the safe transmission and communication of data between the edge service layer and the terminal device layer and the cloud service layer.

[0053] Further, the edge storage module is provided with a data storage threshold judgment unit, which is used to judge whether the cloud service layer is normally running and whether the preprocessed data stored in the edge storage module reaches the capacity threshold.

[0054] Further, the user layer comprises a state detection module and a man-machine interaction module, and the state detection module and the man-machine interaction module are in communication connection with the cloud storage module; the state detection module is used to monitor the explosive melting process in real time, and the man-machine interaction module is used to visually display the explosive melting process.

[0055] The present application has the following advantages:

[0056] The explosive melting casting process temperature prediction model construction method of the application can realize the technical purpose of effectively and accurately obtaining the temperature distribution of the melting casting process only by measuring the temperature at sparse positions, and can balance the temperature field modeling precision and the calculation time to meet the real-time requirement of predicting the temperature information of the explosive melting casting process without calling complex multi-physical simulation.

[0057] The explosive melting casting process temperature control system of the application is constructed through a cloud service layer, an edge service layer and a terminal device layer; the cloud storage module in the cloud service layer can store a large amount of pretreated data pretreated by the data processing module of the edge service layer, and the explosive melting casting process temperature prediction model is trained and verified by using these data as historical data in combination with the powerful computing resources of the cloud service layer, so as to avoid the reaction delay or collapse of the terminal device layer caused by training the explosive melting casting process temperature prediction model in the edge service layer, ensure the real-time performance of the control system and enhance the reliability of the system; meanwhile, the explosive melting casting process temperature prediction model in the explosive melting casting process temperature prediction module is updated by using the trained explosive melting casting process temperature prediction model, so as to ensure the prediction accuracy of the explosive melting casting process temperature prediction model in the explosive melting casting process temperature prediction module; the edge service layer is arranged close to the terminal device layer, the detection module of the terminal device layer is used to collect the environmental information and the running information of the explosive melting casting process in real time, the data processing module of the edge service layer is used to pretreat the environmental information and the running information, and then the temperature information of the explosive melting casting process is predicted in the explosive melting casting process temperature prediction model of the explosive melting casting process temperature prediction module, and the process parameter setting value with the explosive melting casting process temperature as the optimization target and the optimization control parameter of the execution module are obtained by using the optimization decision module, so that the explosive melting casting process can be real-time regulated and controlled by the execution module of the terminal device layer, and the real-time performance problem of the explosive melting casting process control is solved. BRIEF DESCRIPTION OF DRAWINGS

[0058] In order to make the purpose, technical scheme and beneficial effects of the application more clear, the application provides the following drawings for illustration:

[0059] Figure 1 It is a frame principle diagram of the explosive melting casting process temperature control system of the application;

[0060] Figure 2 It is a principle diagram of the terminal device layer;

[0061] Figure 3 It is a principle diagram of the edge service layer;

[0062] Figure 4This is a schematic diagram of the cloud service layer.

[0063] Figure 5 This is a flowchart illustrating temperature control using the temperature control system for the explosive melting and casting process in this embodiment. Detailed Implementation

[0064] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0065] like Figure 1 As shown, the temperature control system for the explosive melting and casting process in this embodiment includes a user layer, a cloud service layer, an edge service layer, and a terminal device layer.

[0066] Specifically, such as Figure 2 As shown, the terminal device layer in this embodiment includes a detection module, an execution module, and a backup communication security module, which is connected to the edge communication security module. The detection module includes a data acquisition unit and a quality inspection unit. The data acquisition unit collects environmental and operational information during the explosive casting process and uploads this information to the edge service layer. Specifically, the data acquisition unit consists of sensors installed on the mold and execution module, used to collect environmental and operational information in real time and upload it to the edge data processing module for preprocessing. Components used for data acquisition include pressure gauges from the pressurization assembly, ambient temperature sensors, pressure sensors, acceleration sensors, vibration sensors, displacement sensors, infrared cameras, NI voltage input modules, and NI current input modules. For example, temperature sensors are installed at the inlet and outlet of the water bath for temperature monitoring; liquid flow controllers are installed at the inlet and outlet of the heat pipe assembly to control the flow rate of the medium. The quality inspection unit is used to inspect the quality of the solidified explosive. Specifically, the quality inspection unit is used to detect defects such as shrinkage and porosity after the explosive solidifies, and to obtain the temperature control effect during the explosive casting process. The technologies used in the quality inspection unit include, but are not limited to, ultrasonic and X-ray non-destructive testing. The execution module controls the operating parameters during the explosive casting process based on the control parameters obtained from the edge service layer, so as to regulate the temperature of the explosive casting process. The execution module includes, but is not limited to, temperature control devices such as gas valves of the gas pressurization component, explosion-proof motors of the mold lifting system, water bath insulation tank, steam heating chamber of riser funnel, heating pipe assembly of mold outer wall, vibration table, vacuum unit, and probe care component. According to the optimized control parameter instructions issued by the temperature control module in the edge service layer, the execution module performs the above-mentioned operations to achieve precise care and control of the temperature during the explosive casting process.

[0067] like Figure 3As shown, the edge service layer of the embodiment includes a data processing module, an edge storage module, an explosive melting process temperature prediction module, an optimization decision module, and an edge communication security module. The edge communication security module is in communication connection with the device communication security module and the cloud communication security module, thereby ensuring the safe transmission and communication of data between the edge service layer and the terminal device layer and the cloud service layer. Specifically, the data processing module pre-processes the environmental information and operation information collected by the detection module and obtains pre-processed data. The data processing module is connected with the detection module in the terminal service layer, acquires environmental information and operation information, and pre-processes the data to obtain pre-processed data. The pre-processed data includes production environment temperature, pouring frequency, pouring time, pouring temperature, mold preheating temperature, water bath water level rising speed, solidification pressure, hot core rod inclination angle, vibration table vibration frequency and time, vacuum unit vacuumizing time, mold outer wall heating pipe assembly internal temperature and steam time, riser funnel steam heating assembly steam flow and temperature, mold lifting system explosion-proof motor speed and power, and inflation pressurization assembly air valve opening and closing degree. The data preprocessing includes data continuity and validity detection, data mining, data cleaning, stream data processing, and data persistence, thereby ensuring data validity and correctness. The edge storage module is used to store the pre-processed data and upload the pre-processed data to the cloud service layer. In the preferred scheme of the embodiment, the edge storage module is provided with a data storage threshold judgment unit. The data storage threshold judgment unit is used to judge whether the cloud service layer is normally running and whether the pre-processed data stored in the edge storage module reaches the capacity threshold. Specifically, when it is judged that the cloud service layer is normally running, the data in the edge storage module is uploaded to the cloud service layer; otherwise, the pre-processed data is continuously stored in the edge storage module until the capacity threshold is reached. The explosive melting process temperature prediction module is provided with an explosive melting process temperature prediction model trained by the cloud service layer. The explosive melting process temperature prediction model predicts the explosive melting process temperature according to the real-time pre-processed data provided by the data processing module. The optimization decision module is connected with the explosive melting process temperature prediction module. Specifically, the optimization decision module includes a parameter optimization unit and a temperature control unit. The parameter optimization unit obtains process parameter set values with the explosive melting process temperature as the optimization target according to the temperature information of the explosive melting process predicted by the explosive melting process temperature prediction module, and sends the optimized set values to the temperature control unit. The temperature control unit obtains the optimization control parameters of the execution module according to the process parameter set values, and sends the optimization control parameters to the execution module.

[0068] As Figure 4As shown, the cloud service layer of the present embodiment includes a cloud storage module, a cloud temperature model training and verification module, an external resource device, and a cloud communication security module. Specifically, the cloud communication security module is connected to the edge communication security module to ensure the secure transmission and communication of data between the cloud service layer and the edge service layer. Data encryption, data decryption, and data verification technologies are used. Data encryption ensures the security of data uploaded to the cloud service layer, data decryption decrypts the obtained data, and data verification checks the security and real-time performance to ensure data security and reliability.

[0069] The cloud storage module is connected to the cloud communication security module, and the cloud storage module is provided with a historical database for storing pre-processed data uploaded by the edge storage module. Specifically, the cloud storage module is responsible for storing and accessing data uploaded by the edge, and the cloud storage module includes a historical database. The historical data stored in the historical database includes, but is not limited to, explosive melting process state, melting explosive running process historical data, edge service layer fault record, and user feedback data. These historical data can be used for statistical analysis of data, and through big data analysis, knowledge graph, machine learning, and other related technologies, cloud computing services such as historical data storage analysis are provided for users. In addition, by combining the edge storage module and the cloud storage module, network bandwidth can be more reasonably and effectively utilized. For example, the edge service layer records high-definition melting process temperature control video, and the staff checks the standard quality video; when the network bandwidth is not limited, high-definition video data is uploaded and stored in the cloud storage module, and the staff can observe more detailed scene information through the high-definition video called by the user layer.

[0070] The cloud temperature model training and verification module is connected with the cloud storage module. The cloud temperature model training and verification module trains and verifies the explosive melting process temperature prediction model by using the historical data stored in the historical database, and transmits the trained explosive melting process temperature prediction model to the explosive melting process temperature prediction module to update the explosive melting process temperature prediction model in the edge service layer. Specifically, the cloud temperature model training and verification module is mainly responsible for training and verifying the melting process temperature prediction model in combination with the data stored in the cloud. In order to solve the problem that it is difficult to obtain accurate overall temperature field information by experimental method, and to solve the problem that numerical simulation technology has large amount of calculation and low efficiency, the embodiment adopts artificial intelligence learning algorithm such as Physics-Informed Neural Network (PINN) to create explosive melting process temperature prediction model. In addition, in order to cope with the temperature control of the melting process in different application scenarios, when a new explosive formula appears, a new model training is needed. Therefore, the explosive temperature prediction model adopts artificial intelligence learning algorithm including but not limited to transfer learning, which can combine the model of the old scene and the data of the new scene to cope with the situation that the model performance fluctuates greatly with the change of the scene. The cloud temperature model training and verification module uses the stored historical data to model, train and verify the explosive melting process temperature prediction model, and grasps the latest control state through the sensor. When idle, the trained and verified explosive temperature prediction model is regularly delivered to the temperature prediction module in the edge service layer to keep the model up to date.

[0071] The external resource device dynamically allocates cloud computing resources of the cloud service layer according to the amount and speed of data uploaded by the edge, in combination with the real-time processing capacity of the edge service layer and the cloud service layer, to realize the fusion of cross-level interaction of edge computing data and cloud computing data, guarantee the real-time operation of the device control of the edge service layer optimization decision module, and realize efficient processing of edge-cloud cooperation.

[0072] The user layer of the embodiment includes a state detection module and a human-computer interaction module, both of which are in communication connection with the cloud storage module. The state detection module is used for real-time monitoring of the explosive melting process, and the human-computer interaction module is used for visual display of the explosive melting process.

[0073] Specifically, in the embodiment, the explosive melting process temperature prediction model is constructed by using the explosive melting process temperature prediction model construction method. Specifically, the explosive melting process temperature prediction model construction method of the embodiment includes the following steps:

[0074] Step one: two PINN neural network models are constructed respectively; wherein the first PINN neural network model is used to learn the thermal boundary condition in the explosive melting process to obtain the best approximation in the thermal boundary condition parameter space; the second PINN neural network model is used to construct a proxy model to realize the rapid prediction of the temperature in the melting process.

[0075] Specifically, in the embodiment, the thermal boundary condition parameter space in the explosive melting process is represented as:

[0076] q(T,t)=h(T,t)(T-T ∞ )

[0077] Wherein, q represents the heat flux; h represents the heat transfer coefficient; T represents the material temperature in the explosive melting process; T ∞ represents the external environment temperature; t represents time;

[0078] The heat transfer coefficient is a function of temperature and time, and is represented as:

[0079]

[0080] Wherein, H (i) represents the heat transfer coefficient of the mold surface i; O (i) represents the directional factor of the mold surface i; i represents the decomposed mold surface, and i∈(1,2,3), respectively representing the top surface, side surface and bottom surface of the mold;

[0081] The heat transfer coefficient of each decomposed mold surface is represented by the linear combination of the basis function of the heat transfer coefficient curve:

[0082]

[0083] Wherein, represents the interpolation coefficient of the heat transfer coefficient defined on the temperature interval k to the temperature interval N k of the mold surface i; represents the piecewise linear basis function defined on the temperature interval k to the temperature interval N k of the mold surface i; k=1,2,…,N k ;

[0084] Each mold surface orientation will affect the heat transfer process, so the directional factors of the top surface, side surface and bottom surface of the mold are respectively represented as:

[0085]

[0086]

[0087]

[0088] Wherein, O(1) , O (2) , and O (3) represent the direction factors of the top, side and bottom surfaces of the mold, respectively; n z represents the component of the unit outward normal vector n on the corresponding surface in the direction of gravity.

[0089] The proxy model of the present embodiment takes the temperature field in the explosive melting and casting process as a Gaussian process, and is expressed as:

[0090] T ~ N(μ, σ)

[0091] μ = μ(t, x, α)

[0092] σ = σ(t, x, α)

[0093] μ M , σ M = M(t, x, α; W * , b * )

[0094] wherein μ and σ represent the mean value and standard deviation of the temperature field, respectively; x and t represent space and time, respectively, and constitute the space-time coordinates (x, t); α represents the parameter space of the thermal boundary condition; μ M and σ M represent the mean temperature and standard deviation, respectively; M represents the proxy model; W * and b * represent the weights and biases of the neural network, and are obtained through the following optimization problem:

[0095]

[0096] L(W, b) = λ data L data (W, b) + λ pde L pde (W, b) + λ bc L bc (W, b)

[0097] wherein L(W, b) represents the loss function, W represents the weights of the neural network, and b represents the biases of the neural network; L data , L pde , and L bc represent the real data condition loss, the partial differential structure loss, and the boundary condition loss, respectively; λ data , λ pde , and λ bc are the corresponding weights.

[0098] 11、Step two: training and validating the surrogate model using the temperature field data in the explosive casting process. In this embodiment, a high-fidelity heat transfer model is created based on the finite element method to obtain the temperature field data in the explosive casting process, and the semi-discrete formula of the high-fidelity heat transfer model is:

[0099]

[0100] where Ω and Γ represent the spatial domain and the boundary domain of the mold, respectively; w and t represent the test function and time, respectively; ρ represents the density; c p represents the specific heat capacity; k represents the thermal conductivity; represents the Laplace operator; d represents differentiation; q(T, t) represents the heat flux.

[0101] Of course, in the explosive casting process, the historical data stored in the historical database of the cloud storage module can also be used to train and validate the surrogate model.

[0102] To verify the accuracy of the model, the absolute error in the data set is defined as:

[0103]

[0104] where e(A) represents the absolute error in the data set; N(A) represents the number of elements in the data set; A represents the data set; T i s represents the measured temperature data on the data set.

[0105] Step three: input the best approximation in the thermal boundary condition parameter space obtained by the first PINN neural network model into the surrogate model trained in step two, and build an explosive casting process temperature prediction model that can obtain the temperature information of the explosive casting process. The explosive casting process temperature prediction model of this embodiment is represented as a minimization problem with respect to the thermal boundary condition parameters:

[0106]

[0107]

[0108] where α * represents the optimal value of the thermal boundary condition parameters; L p represents the loss function; α represents the parameter space of the thermal boundary condition; E represents the set of sensor positions; f represents the Gaussian distribution function; represents the measured temperature; represents the calculated temperature value obtained by the finite element method; represents the corresponding standard deviation of the calculation;

[0109] After obtaining α *Afterwards, the proxy model M is replaced to obtain the temperature prediction and its standard deviation of the whole field, as follows:

[0110] μ * (t,x) = μ(t,x,α * ; W * ,b * )

[0111] σ * (t,x) = σ(t,x,α * ; W * ,b * )

[0112] where μ * and σ * represent the temperature prediction and its standard deviation of the whole field, respectively; x and t represent the space and time, respectively, and constitute the space-time coordinate (x,t); α represents the parameter space of the thermal boundary condition; W * and b * represent the weight and bias, respectively, when the loss function of the neural network is optimal.

[0113] The specific implementation of the explosive melting and casting process temperature control system of the present application will be described below in conjunction with specific examples.

[0114] 1. Test environment

[0115] 1.1 Main hardware and configuration

[0116] Taking the temperature control experiment of an explosive melting and casting process as an example, the hardware and configuration required for developing the explosive melting and casting process temperature control system based on cloud edge-end collaboration are shown in the following table.

[0117] Table 1 Hardware equipment and configuration

[0118]

[0119]

[0120] The cloud service layer needs to handle high-concurrency resource access and training and verification of prediction models, etc., so the cloud service layer is deployed using high-performance servers, and virtualization technology is used to make efficient use of system resources. Specifically, in the resource pool established by the server, a virtual machine is constructed for storage, calculation domain analysis of the cloud service layer, and the virtual servers of the cloud service layer are as shown in the table.

[0121] Table 2 Virtual server list of cloud service layer

[0122]

[0123] The edge service layer uses a hyper-converged infrastructure to form a resource pool and build virtualized servers, as shown in the table.

[0124] Table 3 List of virtual servers of the edge service layer

[0125]

[0126] The terminal device layer collects information through external sensors of the detection module, and performs experimental testing and analysis to verify the effectiveness of the proposed explosive melting and casting process temperature control system based on cloud edge-end collaboration. Some data is programmed by Labview-based data acquisition program, and NI high-speed acquisition card and other devices are used to connect the data acquisition card and the edge processing module through the case to ensure the real-time data transmission and the scalability of the acquisition card. For example, NI-9203 is used for temperature, NI9234 is used for vibration, NI 9203 is used for current, and NI 9220 is used for power. The models of some data acquisition units are as follows.

[0127] Table 4 Models of some sensors and data acquisition units

[0128]

[0129] The production environment temperature, pouring temperature, mold preheating temperature, water bath water level rising speed, solidification pressure, vibration table vibration frequency and time, vacuum unit vacuum time, mold outer wall heating pipe assembly internal temperature and steam time, riser funnel steam heating assembly steam flow and temperature, etc. during the explosive melting and casting process are monitored, and different data flux is realized by changing the sampling frequency. Among them, the vibration sensor can be fixed on the monitoring part by adhesion.

[0130] 1.2 Main software and configuration

[0131] The operating system mainly uses Windows10x64bit, and uses B / S network architecture based on local area network environment. Each service layer is accessed through optical fiber, and the gateway between stations and devices uses super five category shielded twisted pair or 2.4G WiFi network access system platform. The system development platform can include Visual Studio 2008, and the development language of the system uses Java, Python, HTML, Javescript, C#; the integrated development environment includes IDEA, JDK. The main software and configuration are shown in the table.

[0132] Table 5 Software and configuration

[0133]

[0134] 2. Explosive melt-casting process temperature control system

[0135] The explosive melt-casting process temperature control system of the embodiment is a control system based on cloud edge-end collaborative computing system. Specifically, the explosive melt-casting process temperature control system based on cloud edge-end collaboration includes a user layer, a cloud service layer, an edge service layer and a terminal device layer.

[0136] The cloud service layer includes a cloud communication security module, a cloud storage module, a cloud temperature model training and verification module, and an external resource device. The cloud storage module can be placed in random access memory (RAM), memory, hard disk, removable hard disk or any other form of storage medium, which also includes a historical database. The historical database is used to accept data uploaded by the edge service layer and divide the data into training data and verification data, which are 70% and 30% respectively. The temperature model training and verification module uses 70% of the training data stored in the historical database to learn and train the melt-casting process temperature model, with production environment temperature, pouring frequency, pouring time, pouring temperature, mold preheating temperature, water bath water level rising speed, solidification pressure, hot core rod inclination angle as model input, and mold internal temperature as output to obtain melt-casting process temperature data. In addition, 30% of the verification data from the historical database is used to evaluate and verify the prediction model, and the related parameters are adjusted accordingly during the verification process to obtain the best explosive melt-casting process temperature prediction model; further, after obtaining the model, the temperature prediction module in the edge service layer is periodically issued with T1 period.

[0137] The edge service layer is used for preprocessing the environmental information and the running information, and storing and uploading the preprocessed data to the cloud storage module in the cloud service layer. In addition, according to the temperature prediction model issued by the temperature model training and verification module in the cloud server, the temperature prediction value of the melting and casting process is obtained through the edge temperature prediction module, and the optimization control parameter of the terminal device layer is further obtained through the optimization decision module. The edge service layer of the embodiment comprises: an edge data processing module and an edge security communication module in data communication with the detection module in the terminal device layer, an edge storage module connected with the data processing module, a temperature prediction module and an optimization control module. The edge data processing module can accept the collected data through ModBus, RS232 and other field bus protocols, can analyze and process most of the data on site, and reduce unnecessary real-time data upload; can be used for preprocessing the environmental information and the running information, including but not limited to data continuity and validity detection, data mining, data cleaning, stream data processing and data persistence; at the same time, it supports various typical transmission protocols such as AMQP, MQTT, and then uploads the preprocessed data to the edge storage module. The edge storage module is used for short-term storage of the above data, and uploads the data to the cloud storage module in the cloud service layer through Ethernet every set period. In addition, it is necessary to judge whether the cloud service layer is running normally before uploading to the cloud service layer for storage. If yes, the edge storage module output data is uploaded to the cloud service layer; if not, the edge storage module output data is returned to the edge storage module, and continues to store until the capacity threshold (i.e. the maximum storage capacity of the edge service layer) is reached, and after the cloud service layer resumes normal operation, the data is synchronized and uploaded to the cloud service layer. The temperature prediction module obtains the temperature prediction data of the explosive melting and casting process in combination with the temperature prediction model provided by the cloud service layer and the data provided by the edge data processing module. And send it to the parameter optimization module connected with the temperature prediction module. The parameter optimization module is used for calculating the parameter optimization setting value of the terminal device layer according to the temperature prediction data, and sending the optimization setting value to the temperature control algorithm; wherein the optimization algorithm can include but is not limited to genetic algorithm, tabu search algorithm, simulated annealing algorithm and other artificial intelligence optimization algorithms to calculate the optimization setting value of the terminal device layer. The temperature control module has more advanced algorithms and control strategies than PLC, which is used to calculate the optimization control parameter of the terminal device layer according to the optimization setting value, and send the optimization control parameter to the execution module in the terminal device layer.

[0138] The detection module of the terminal device layer is used to obtain environmental information and running information of the melt-casting process, and upload the environmental information and the running information to the edge storage module. The detection module includes a data acquisition unit and a quality detection unit. The data acquisition unit can include but is not limited to a pressure gauge of the air inflation pressurization assembly, an environmental temperature sensor, a pressure sensor, an acceleration sensor, a vibration sensor, a displacement sensor, an infrared camera, an NI voltage input module, an NI current input module, etc. The quality detection unit is used for defect detection such as shrinkage and porosity of the solidified explosive, and obtains the temperature control effect of the explosive melt-casting process. The technology used by the quality detection unit includes but is not limited to ultrasonic wave, X-ray nondestructive testing, etc. The data information can include production environment temperature, pouring temperature, mold preheating temperature, water bath water level rising speed, solidification pressure, vibration table vibration frequency and time, vacuum unit vacuumizing time, mold outer wall heating pipe assembly internal temperature and steam time, riser funnel steam heating assembly steam flow and temperature, etc. The execution module includes but is not limited to a temperature control device such as an air valve of the air inflation pressurization assembly, a mold lifting system explosion-proof motor, a water bath heat preservation tank, a riser funnel steam heating cavity, a mold outer wall heating pipe assembly, etc. According to the optimized control parameter instruction issued by the temperature control module in the edge service layer, the operation of the above execution module is performed to realize accurate care and control of the temperature of the explosive melt-casting process.

[0139] 3. An industrial control method for an explosive melt-casting process

[0140] In order to reduce the pressure of the cloud service layer and ensure the stable and efficient operation of the control system, the complex and computationally intensive model is deployed in the cloud service layer and is optimized accordingly. The optimized model is periodically issued to the edge service layer close to the terminal device layer. The edge service layer and the terminal device layer interact, reason, calculate and control in real time, which can efficiently distribute the computational pressure of the cloud service layer and the edge, and improve the efficiency and stability of the global optimization and temperature control of the control system.

[0141] Specifically, as shown in Figure 5 the temperature control steps of the explosive melt-casting process temperature control system are as follows:

[0142] Step 1: Obtain the environmental information and running information of the explosive melt-casting process through the detection module of the terminal device layer, and upload the environmental information and running information to the edge service layer.

[0143] Step 2: The edge data processing module in the edge service layer pre-processes the information, and the pre-processed data is short-term stored in the edge storage module. The pre-processing includes operations such as data continuity and validity detection, data mining, data cleaning, stream data processing and data persistence, so as to ensure the validity and correctness of the data. In addition, the following judgments need to be made before uploading to the cloud service layer.

[0144] Step2.1: judging whether the cloud service layer is running normally, if yes, uploading the data in the edge storage module of the edge service layer to the cloud service layer;

[0145] Step2.2: if not, continuing to store the pre-processed data until reaching the threshold;

[0146] Step3: the cloud explosive melting process temperature model training and verification module in the cloud service layer trains and verifies the explosive melting process temperature prediction model according to the stored data in the cloud storage module, and regularly issues the model to the temperature prediction module;

[0147] Step4: the temperature prediction module obtains the temperature prediction data of the melting process according to the obtained temperature prediction model, combining the real-time collected melting process environmental information and running information and the pre-processed data;

[0148] Step5: obtaining the optimized control parameters of the terminal device layer execution module through the parameter optimization unit and the temperature control unit in the optimization decision module;

[0149] Step6: the terminal device layer execution module adjusts its current running parameters according to the above-mentioned optimized control parameters, so as to realize the closed-loop feedback control of the melting process, and uploads its control information to the cloud storage module.

[0150] Specifically, the environmental information and the running information of the explosive melting process are obtained through the detection device in the terminal device layer, and are uploaded to the edge storage module in the edge service layer. The detection module can include a pressure gauge of the air inflation pressurization assembly, an environmental temperature sensor, a pressure sensor, an acceleration sensor, a vibration sensor, a displacement sensor, an infrared camera, etc.; the quality detection unit is used for detecting defects such as shrinkage and porosity of the solidified explosive, obtaining the temperature control effect of the explosive melting process, and the technologies used by the quality detection unit include but are not limited to ultrasonic, X-ray non-destructive testing, etc. The data information can include production environmental temperature, pouring temperature, mold preheating temperature, water bath water level rising speed, solidification pressure, vibration table vibration frequency and time, vacuum unit vacuum time, riser funnel steam heating assembly steam flow and temperature, mold lifting system explosion-proof motor speed and power, air inflation pressurization assembly air valve opening and closing degree, etc.

[0151] The edge data processing module in the edge service layer pre-processes the environmental information and the running information to obtain pre-processed data. The pre-processing process includes but is not limited to data continuity and validity detection, data mining, data cleaning, stream data processing, and data persistence, etc. The pre-processed data is uploaded to the cloud storage module in the cloud service layer through the edge security communication module and the cloud security communication module every T2 period and through the Ethernet. In addition, before uploading to the cloud service layer for storage, it is necessary to judge whether the cloud service layer is normally running. If yes, the edge storage module output data is uploaded to the cloud service layer; if not, the edge storage module output data is returned to the edge storage module and continues to be stored until the threshold (i.e. the maximum storage capacity of the edge service layer) is reached, and after the cloud service layer resumes normal operation, the data is synchronized to the cloud service layer.

[0152] The temperature model training and verification module in the cloud service layer trains and verifies the cloud storage data to obtain a temperature prediction model of the explosive melting and casting process, and sends the model to the temperature prediction module in the edge service layer. The cloud storage module divides 70% of the data as training data and 30% as verification data, and adjusts the relevant parameters during the verification process to obtain the best explosive melting and casting process temperature prediction model; further, after obtaining the model, it is periodically issued to the temperature prediction module in the edge service layer at a set period interval, or it is issued to the temperature prediction module in the edge service layer when the prediction accuracy exceeds the set threshold.

[0153] The optimization decision module in the edge service layer obtains the optimization control parameters of the execution module in the terminal device layer according to the temperature prediction value. Specifically, the optimization control parameters of the execution module are calculated using the corresponding method. The execution module includes but is not limited to the air valve of the air inflation and pressurization assembly, the explosion-proof motor of the mold lifting system, the water bath heat preservation tank, the steam heating cavity of the riser funnel, the mold outer wall heating pipe assembly, the vibration table, the vacuum unit, the probe, etc. Artificial intelligence optimization algorithms including but not limited to genetic algorithm, tabu search algorithm, etc. can be used to calculate the optimization set values such as pouring time, pouring temperature, pouring speed, water bath temperature, etc. After calculating the optimization set values, the corresponding optimization control parameters are calculated in real time, and the optimization control parameters are sent to the execution module in real time. The execution module performs the above operations according to the optimization control parameter instructions issued by the temperature control module in the edge service layer.

[0154] The above-described embodiments are only preferred embodiments to fully illustrate the present application, and the protection scope of the present application is not limited thereto. Equivalent substitutions or changes made by those skilled in the art based on the present application are within the protection scope of the present application. The protection scope of the present application is subject to the claims.

Claims

1. A method for constructing a temperature prediction model for the explosive melting and casting process, characterized in that: The method comprises the following steps: Step one: two PINN neural network models are respectively constructed; wherein the first PINN neural network model is used to learn the thermal boundary condition in the explosive melting process to obtain the best approximation in the thermal boundary condition parameter space; the second PINN neural network model is used to construct a proxy model to realize fast prediction of the temperature in the melting process; Step two: the proxy model is trained and verified by using the temperature field data in the explosive melting process; Step three: the best approximation in the thermal boundary condition parameter space obtained by the first PINN neural network model is input into the proxy model trained in step two, and an explosive melting process temperature prediction model capable of obtaining the temperature information of the explosive melting process is constructed; In step one, the thermal boundary condition parameter space in the explosive melting process is represented as: wherein, represents heat flux; represents heat transfer coefficient; represents material temperature during the explosive melt-casting process; represents external ambient temperature; represents time; The heat transfer coefficient is a function of temperature and time, represented as: wherein represents the heat transfer coefficient of the mold surface ; represents the directional factor of the mold surface ; represents the decomposed mold surface, and , respectively, represent the top surface, the side surface, and the bottom surface of the mold. The heat transfer coefficient of each mold surface to be decomposed is represented by a linear combination of the basis functions of the heat transfer coefficient curve: wherein denotes the mold surface in the temperature interval to the temperature interval an interpolation factor of the heat transfer coefficient defined above; denotes the mold surface in the temperature interval to the temperature interval a piecewise linear basis function defined above; ; The orientation of each mold surface will affect the heat transfer process, so the direction factors of the top surface, side surface and bottom surface of the mold are represented as: wherein, , and respectively denote the top, side and bottom face direction factors of the mold; denotes the component of the unit outward normal vector on the corresponding surface in the direction of gravity.

2. The explosive melt-cast process temperature prediction model construction method of claim 1, wherein: The proxy model regards the temperature field in the explosive melting process as a Gaussian process, represented as: where, and denote the mean and standard deviation of the temperature field, respectively; and denote space and time, respectively, and constitute the spatio-temporal coordinate ; denotes the parameter space of the thermal boundary condition; and denote the mean temperature and the standard deviation, respectively; denotes the surrogate model; and denote the weights and biases, respectively, that make the loss function of the neural network optimal, and are obtained by solving the following optimization problem: wherein, represents a loss function, represents a weight of the neural network, represents a bias of the neural network; , and respectively represent a real data condition loss, a partial differential structure loss, and a boundary condition loss; , and are corresponding weights, respectively.

3. The explosive melt-cast process temperature prediction model construction method of claim 1, wherein: In step two, a high-fidelity heat transfer model is created based on the finite element method to obtain the temperature field data in the explosive melting process, and the semi-discrete formula of the high-fidelity heat transfer model is: wherein, and denote the spatial domain and the boundary domain of the mold, respectively; and denote the test function and the time, respectively; denotes the density; denotes the specific heat capacity; denotes the thermal conductivity; denotes the Laplace operator; denotes the differential; denotes the heat flux.

4. The explosive melt-cast process temperature prediction model construction method of claim 1, wherein: In step three, the explosive melting process temperature prediction model is represented as a minimization problem with respect to the thermal boundary condition parameters: wherein, represents an optimal value of a thermal boundary condition parameter; represents a loss function; represents a parameter space of a thermal boundary condition; represents a set of sensor positions; represents a Gaussian distribution function; represents a measured temperature; represents a calculated value of temperature obtained by a finite element method; represents a standard deviation corresponding to the calculation; After obtaining Subsequently, the proxy model is replaced by the surrogate model The temperature prediction and its standard deviation for the full field are obtained as follows: where, and represent the full-field temperature prediction and the standard deviation, respectively; and represent space and time, respectively, and constitute a spatio-temporal coordinate ; denotes the parameter space of the thermal boundary condition; and represent the weights and biases, respectively, at which the loss function of the neural network is optimal.

5. An explosive melt-cast process temperature control system characterized by: It comprises a cloud service layer, an edge service layer and a terminal device layer; The terminal device layer comprises a detection module and an execution module; the detection module comprises a data acquisition unit and a quality detection unit, the data acquisition unit is used to acquire environmental information and running information in the explosive melting process, and upload the acquired environmental information and running information to the edge service layer; the quality detection unit is used to detect the quality of the solidified explosive; The execution module controls the running parameters in the explosive melting process according to the control parameters obtained by the edge service layer to control the temperature of the explosive melting process; The edge service layer comprises a data processing module, an edge storage module, an explosive melting process temperature prediction module and an optimization decision module; the data processing module pre-processes the environmental information and running information collected by the detection module and obtains pre-processed data; The edge storage module is configured to store the preprocessed data and upload the preprocessed data to a cloud service layer; the explosive melt-casting process temperature prediction module is internally provided with an explosive melt-casting process temperature prediction model trained by the cloud service layer, and the explosive melt-casting process temperature prediction model is configured to predict the temperature of the explosive melt-casting process according to the real-time preprocessed data provided by the data processing module; the optimization decision module includes a parameter optimization unit and a temperature control unit, the parameter optimization unit is configured to obtain process parameter set values with the temperature of the explosive melt-casting process as an optimization target according to the temperature information of the explosive melt-casting process predicted by the explosive melt-casting process temperature prediction module, and the temperature control unit is configured to obtain optimization control parameters of the execution module according to the process parameter set values and send the optimization control parameters to the execution module. The cloud service layer includes a cloud storage module and a cloud temperature model training and verification module; the cloud storage module is internally provided with a historical database configured to store the preprocessed data uploaded by the edge storage module; the cloud temperature model training and verification module is configured to train and verify the explosive melt-casting process temperature prediction model by using the historical data stored in the historical database, and transmit the trained explosive melt-casting process temperature prediction model to the explosive melt-casting process temperature prediction module to update the explosive melt-casting process temperature prediction model in the edge service layer. The explosive melt-casting process temperature prediction model is constructed by the explosive melt-casting process temperature prediction model construction method according to any one of claims 1-4.

6. The explosive melt-cast process temperature control system of claim 5, wherein: The cloud service layer further includes an external resource device, which is configured to dynamically allocate cloud computing resources of the cloud service layer according to the amount and uploading speed of the data uploaded by the edge, in combination with the real-time processing capacity of the edge service layer and the cloud service layer, so as to realize the fusion of cross-level interaction of edge computing data and cloud computing data, guarantee the real-time operation of the equipment control of the optimization decision module of the edge service layer, and realize efficient processing of edge-cloud collaboration.

7. The explosive melt-cast process temperature control system of claim 5, wherein: The terminal device layer is further provided with an equipment communication security module, the edge service layer is further provided with an edge communication security module, and the cloud service layer is further provided with a cloud communication security module; the edge communication security module is in communication connection with the equipment communication security module and the cloud communication security module, so as to guarantee the safe transmission and communication of data between the edge service layer and the terminal device layer and the cloud service layer.

8. The explosive melt-cast process temperature control system of claim 5, wherein: The edge storage module is internally provided with a data storage threshold judgment unit, which is configured to judge whether the cloud service layer is normally running and whether the preprocessed data stored in the edge storage module reaches a capacity threshold.

9. The explosive melt-cast process temperature control system of claim 5, wherein: Further including a user layer, the user layer includes a state detection module and a human-computer interaction module, and the state detection module and the human-computer interaction module are in communication connection with the cloud storage module; the state detection module is configured to monitor the explosive melt-casting process in real time, and the human-computer interaction module is configured to visually display the explosive melt-casting process.

Images