An intelligent calculation method for equipment physical coupling field driven by virtual-real sensor augmentation

Abstract

The present application discloses an intelligent calculation method of equipment physical coupling field driven by virtual-real sensor augmentation, which relates to the field of equipment physical performance simulation. The method encodes the real position and real sensor data of each real node to obtain the real sensor position feature, and based on the real sensor position feature of each real node, uses a real-domain spatiotemporal coding network to create the virtual sensor position feature of the virtual node, and then based on the virtual-real sensor features of the virtual-real node, uses a virtual-real coding network to create the virtual-real sensor data of the virtual-real node, so as to expand the spatial distribution of the sensor data, and then constructs a temperature field neural network model and a stress field neural network model based on the temperature and stress mechanism equations, establishes a multi-parameter nesting mechanism under thermal-mechanical coupling, and uses the temperature field calculation results and strain-stress conversion relationship to standardize the stress field training process, so as to realize the intelligent and efficient calculation of the equipment temperature-stress coupling field.

Description

Technical Field

[0001] The present application relates to the field of equipment physical performance simulation, and in particular to an intelligent calculation method for equipment physical coupling field driven by virtual-real sensor augmentation. Background Art

[0002] With the rapid development of the manufacturing industry, higher requirements are placed on the service performance of complex equipment such as CNC machine tools and gas turbines. Among them, the operating status of key components represented by machine tool spindles and gas turbine blades is directly related to the performance and reliability of the entire equipment. Since these components often operate in a changeable and harsh working environment and are simultaneously subjected to the coupling influence of multiple physical factors, it becomes difficult to monitor the operating status of key components of the equipment. Therefore, how to efficiently and accurately calculate the physical coupling field of the equipment is of great significance to ensuring the service performance of the equipment and extending its service life.

[0003] At present, the research methods in the calculation of equipment physical coupling fields can be roughly divided into two categories: mechanism-driven methods and data-driven methods. Among them, the mechanism-driven method usually determines the boundary conditions based on the physical mechanism and uses simulation methods such as finite element to solve the physical field offline. However, the calculation accuracy of this type of method usually depends on the fineness of the grid, the calculation time increases exponentially with the number of grids, and each working condition needs to be modeled separately, which makes it difficult to adapt to the equipment physical field monitoring tasks under complex working conditions. The data-driven method builds a physical field calculation model based on machine learning, deep learning, etc. This type of method can make full use of sensor data, but due to the limited number and sampling frequency of real sensors, the input of the physical field calculation model is insufficient and the calculation accuracy cannot be guaranteed. Summary of the invention

[0004] The purpose of this application is to provide an intelligent calculation method for the physical coupling field of equipment driven by virtual-real sensor augmentation, so as to expand the input of the physical field calculation model and improve the calculation accuracy of the physical coupling field.

[0005] To achieve the above objectives, this application provides the following solutions.

[0006] In a first aspect, the present application provides a method for intelligent calculation of equipment physical coupling field driven by virtual-real sensor augmentation, comprising:

[0007] Collect real sensor data of different real nodes during the service of the equipment; the real node is a first real node with a temperature sensor deployed on the equipment, or a second real node with a stress sensor deployed on the equipment, and the different real nodes include at least one first real node and one second real node.

[0008] The real position and real sensing data of each real node are encoded to obtain the real sensing position features of each real node.

[0009] According to the real sensor position features of different real nodes, a real-domain spatiotemporal coding network is used to generate virtual sensor position features of multiple virtual nodes.

[0010] The real sensing position features of different real nodes and the virtual sensing position features of different virtual nodes are spliced ​​through the jump connection layer to obtain the virtual and real sensing features of different virtual and real nodes.

[0011] Based on the virtual-real sensing characteristics of different virtual-real nodes, a virtual-real coding network is used to generate virtual-real sensing data of different virtual-real nodes; the virtual-real sensing data of different virtual-real nodes include: virtual-real temperature data of different first virtual-real nodes and virtual-real stress data of different second virtual-real nodes.

[0012] The temperature field neural network model and stress field neural network model are constructed respectively according to the temperature mechanism equation, stress mechanism equation and sensor function loss term.

[0013] The temperature field is calculated according to the virtual and real temperature data of each first virtual and real node using a temperature field neural network model.

[0014] The temperature data of the temperature field is converted into thermal strain data using the temperature-strain conversion equation.

[0015] The stress field is calculated according to the virtual and real stress data of each second virtual and real node and the thermal strain data using a stress field neural network model.

[0016] According to the specific embodiments provided in this application, this application has the following technical effects.

[0017] The present application provides an intelligent calculation method for the physical coupling field of equipment driven by virtual-real sensor augmentation. The present application encodes the real position and real sensor data of each real node to obtain the real sensor position feature. Based on the real sensor position feature of each real node, a real-domain spatiotemporal coding network is used to create the virtual sensor position feature of the virtual node. Then, based on the virtual-real sensor features of the virtual and real nodes, a virtual-real coding network is used to create the virtual-real sensor data of the virtual and real nodes, thereby expanding the spatial distribution of the sensor data. Furthermore, based on the temperature and stress mechanism equations, a temperature field neural network model and a stress field neural network model are constructed, a multi-parameter nesting mechanism under thermal-mechanical coupling is established, and the temperature field calculation results and the strain-stress conversion relationship are used to standardize the stress field training process, thereby realizing the intelligent and efficient calculation of the equipment temperature-stress coupling field. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flowchart of an intelligent calculation method for a physical coupling field of equipment driven by virtual-reality sensing augmentation is provided in one embodiment of the present application.

[0020] Figure 2 A schematic diagram of encoding real sensor data and its position provided in accordance with an embodiment of the present application.

[0021] Figure 3 A schematic diagram of the generation of virtual-real sensing data provided in one embodiment of the present application.

[0022] Figure 4 A schematic diagram of the temperature field and stress field simulation provided in one embodiment of the present application. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only 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 without creative work are within the scope of protection of this application.

[0024] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0025] In an exemplary embodiment, Figure 1 As shown, the present application provides a method for intelligent calculation of physical coupling field of equipment driven by virtual-real sensing augmentation, including the following steps 101 to 109.

[0026] Step 101, collecting real sensor data of different real nodes during the service of the equipment; the real node is a first real node on which a temperature sensor is arranged on the equipment, or a second real node on which a stress sensor is arranged on the equipment, and the different real nodes include at least one first real node and one second real node.

[0027] Step 102, encoding the real position and real sensing data of each real node to obtain the real sensing position feature of each real node.

[0028] Step 103 , based on the real sensor position features of different real nodes, a real-domain spatiotemporal coding network is used to generate virtual sensor position features of multiple virtual nodes.

[0029] Step 104, the real sensing position features of different real nodes and the virtual sensing position features of different virtual nodes are spliced ​​through a jump connection layer to obtain virtual-real sensing features of different virtual-real nodes.

[0030] Step 105, based on the virtual-real sensing features of different virtual-real nodes, a virtual-real coding network is used to generate virtual-real sensing data of different virtual-real nodes; the virtual-real sensing data of different virtual-real nodes include: virtual-real temperature data of different first virtual-real nodes and virtual-real stress data of different second virtual-real nodes.

[0031] Step 106, constructing a temperature field neural network model and a stress field neural network model respectively according to the temperature mechanism equation, the stress mechanism equation and the sensor function loss term.

[0032] Step 107, calculating the temperature field using the temperature field neural network model according to the virtual and real temperature data of each first virtual and real node.

[0033] Step 108, using the temperature-strain conversion equation, convert the temperature data of the temperature field into thermal strain data.

[0034] Step 109, calculating the stress field using a stress field neural network model according to the virtual and real stress data of each second virtual and real node and the thermal strain data.

[0035] In another exemplary embodiment of the present application, in the above step 101, when the real node is the first real node, the real sensing data of the real node is the real temperature data of the first real node, and when the real node is the second real node, the real sensing data of the real node is the real stress data of the second real node. Further, the above real sensing data is obtained by arranging temperature and stress sensors at the limited nodes of key components of the equipment to collect the real sensing data of the equipment during service, for example, arranging temperature and stress sensors at both ends of the ball screw feed system of the CNC machine tool and on the slider. Temperature and M stress sensors are used to collect temperature and stress sensing data of the ball screw of CNC machine tools during service.

[0036] In another exemplary embodiment of the present application, the above step 102 can be implemented using the following steps 201 to 203.

[0037] Step 201, encode the real position of each real node to obtain the real position feature of each real node.

[0038] Step 202: Encode the real sensing data of each real node to obtain the real sensing features of each real node.

[0039] Step 203 , feature fusion is performed on the real position features and the real sensing features of the same real node to obtain the real sensing position features of each real node.

[0040] In another exemplary embodiment of the present application, Figure 2 As shown, in the above steps 201 to 203, the position of each sensor (i.e., the real position of the real node) is encoded and input into the real-domain spatiotemporal coding network as the real sensor position feature together with the real sensor feature obtained by encoding the real sensor data to obtain the virtual sensor position features at more nodes (i.e., virtual nodes).

[0041] For example, a fixed sensor (a temperature sensor or a stress sensor) at the end of the ball screw feed system of a CNC machine tool is used as a reference sensor to calculate the The coordinates of the real nodes relative to the reference sensor , and encode the position of each fixed sensor (i.e. the real position of the real node) to obtain the real position feature , the specific formula is as follows.

[0042] ;

[0043] ;

[0044] in, For the The real location features of the real nodes, For the The position vector of the real node, , , and Respectively The real location of the real node Axis coordinates, Axis coordinates and Axis coordinates, For the The real position encoding vector of the real node, the superscript T represents the transposition; , Respectively The real location of the real node The 0th order sum of the axis coordinates Taylor expansion value of order; , Respectively The real location of the real node The 0th order sum of the axis coordinates Taylor expansion value of order; , Respectively The real location of the real node The 0th order sum of the axis coordinates Taylor expansion value of order; is the variance of the Taylor expansion, is the Taylor expansion series.

[0045] In another exemplary embodiment, in the above step 201, the real sensor data collected by the fixed sensor is Transformed into real sensor features through convolutional coding layers ,in, is the length of the time window, is the number of channels of the multi-source sensor, The number of channels of the convolutional layer is defined. This step converts the original The number of channels can be expanded to any Dimensions, Figure 2 In , , are the first channel, the second channel, and the The real sensor data of the channel. At this time, due to and are selected by the user, so , the final true sensor position feature can be written in the following format.

[0046]

[0047] in, For the The real sensor location features of real nodes, For the The real sensing features of real nodes, Indicates The real position features of the real nodes, superscript Indicates transpose.

[0048] In another exemplary embodiment, in the above step 103, the virtual node is a first virtual node or a second virtual node. When the virtual node is a first virtual node, the virtual sensing position characteristic of the virtual node is a virtual temperature characteristic of the first virtual node, and when the virtual node is a second virtual node, the virtual sensing position characteristic of the virtual node is a virtual stress characteristic of the second virtual node; the first virtual node is a node on which a temperature sensor cannot be deployed on the device, and the second virtual node is a node on which a stress sensor cannot be deployed on the device.

[0049] In the above step 103, the real sensor position features of different real nodes are input into the real-domain spatiotemporal coding network to obtain virtual sensor position features at more nodes (virtual nodes).

[0050] The structure of the real-domain space-time coding network is as follows Figure 3 As shown. The real-domain spatiotemporal coding network includes: a real-domain spatiotemporal coding layer and a multi-layer perceptron; the real-domain spatiotemporal coding layer includes: a first residual connection module (i.e. Figure 3 " ”), a first layer normalization module and a first time attention layer, a first channel attention layer and a first spatial attention layer connected in sequence; a first input end of the first residual connection module is connected to an output end of the first spatial attention layer, a second input end of the first residual connection module is used to input real sensor position features, an output end of the first residual connection module is connected to an input end of the first layer normalization module, and an output end of the first layer normalization module is connected to an input end of the first multi-layer perceptron.

[0051] The first time attention layer uses the self-attention mechanism to extract the correlation weights before and after the time sequence. The first channel attention is used to remove the negative impact of redundant channels on the results. The first spatial attention layer uses the graph attention network to mine the correlation features of the sensor spatial layout. Through this real-domain spatiotemporal encoding layer, the real sensor position feature is mapped to a virtual sensor position feature embryo. , and through the history of the real node The time step data imposes a loss constraint on it The purpose of this step is to construct an intermediate variable related to virtual sensing in the real-domain spatiotemporal coding network to characterize the virtual sensor that does not exist in reality, and to preliminarily construct a mapping model from real sensor position features to virtual sensor position features, and to continuously improve the accuracy of the virtual sensor position feature embryo in subsequent iterative optimization. At the same time, In the data part, neural network is introduced to assist training, which effectively reduces the difficulty of prediction by converting the complex problem of virtual sensor location feature differences into a node-to-node prediction problem.

[0052] The above step 103 calculates the attention weight values ​​of the real sensor position features in the time dimension and the space dimension in series, and uses the channel attention to select the channel with the best correlation to avoid the interference of redundant sensor features. Then, the virtual sensor position features of more nodes, that is, the virtual sensor position features of multiple virtual nodes, are obtained.

[0053] In another exemplary embodiment, in the above step 104, the virtual-real node is the first virtual-real node or the second virtual-real node; when the virtual-real node is the first virtual-real node, the virtual-real sensing characteristics of the virtual-real node are the virtual-real temperature characteristics of the first virtual node; when the virtual-real node is the second virtual-real node, the virtual-real sensing characteristics of the virtual-real node are the virtual-real stress characteristics of the second virtual node; the first virtual-real node is the first real node or the first virtual node, and the virtual-real temperature characteristics are the real temperature characteristics or the virtual temperature characteristics; the second virtual-real node is the second real node or the second virtual node, and the virtual-real stress characteristics are the real stress characteristics or the virtual stress characteristics.

[0054] In another exemplary embodiment, in the above step 104, a skip connection mechanism is used to splice the real sensor position feature and the virtual sensor position feature (i.e. Figure 3 The virtual sensor position feature embryo in the image is obtained by obtaining the real-virtual sensor position feature, namely the virtual-real sensor feature. On the one hand, a deeper virtual-real sensor association model (namely the virtual-real coding network) is constructed. On the other hand, a residual-like structure is used to effectively prevent the performance degradation of the neural network.

[0055] In another exemplary embodiment, in the above step 105, the virtual-real sensing features are input into the virtual-real coding network, and the temporal and spatial attention serial analysis and calculation in the virtual-real coding network are used, and the most relevant channel features are selected in combination with the channel attention, so as to achieve accurate prediction of the virtual-real sensing data in time. Figure 3 As shown, the virtual-real coding network includes: a virtual-real coding layer and a second multi-layer perceptron; the virtual-real coding layer includes a second residual connection module (i.e. Figure 3 " ”), a second layer normalization module and a second time attention layer, a second channel attention layer and a second space attention layer connected in sequence; the first input end of the second residual connection module is connected to the output end of the second space attention layer, the second input end of the second residual connection module is connected to the output end of the jump connection layer, the output end of the second residual connection module is connected to the input end of the second layer normalization module, and the output end of the second layer normalization module is connected to the input end of the second multi-layer perceptron. Compared with the previous real-domain spatiotemporal coding layer, the virtual-real coding layer in the virtual-real coding network acts on the entire real-virtual sensor network. Because the virtual sensors that did not originally exist are characterized, they can be merged into the sensor cluster to further extract the complex correlation between real and virtual sensors. With the iterative optimization of the network model, the virtual sensor position feature embryo also evolves, and eventually approaches the virtual temperature data and virtual stress data at the required sampling time.

[0056] In another exemplary embodiment, in the above steps 106 to 109, thermal strain is used as an intermediate variable to establish an efficient and intelligent calculation model of the temperature-thermal strain-thermal stress multi-parameter nested physical coupling field, and the real-time calculation results of the temperature field neural network model are converted into thermal strain loads as additional inputs of the stress field neural network model. The strain-stress mechanism equation is used as an additional loss function term of the stress field neural network model to solve the stress field, thereby realizing efficient and intelligent calculation of the equipment's thermal and stress physical coupling fields.

[0057] In another exemplary embodiment, the above step 106 specifically includes the following steps 301 to 306 .

[0058] Step 301, assign separate physical information neural network models to the temperature field and stress field, respectively, and efficiently and intelligently calculate the temperature field of the ball screw feed system under the condition that the temperature mechanism equation and the virtual and real temperature data are constrained. In step 301, the temperature mechanism equation and the root mean square error between the real value and the predicted value of the virtual and real temperature data are used as the first loss function to standardize the training convergence direction of the temperature field neural network and obtain the temperature field neural network model.

[0059] Among them, the temperature mechanism equation includes steady-state equation and unsteady-state equation.

[0060] The steady-state equation is: .

[0061] in, is the thermal conductivity, is the temperature, The Laplace operator is used to describe the local changes of the field. Its function is to calculate the second-order spatial derivative of the temperature around a point. In the steady-state equation, the Laplace operator can represent the degree of diffusion of temperature in space. It helps describe how heat is distributed and propagated in an object.

[0062] The unsteady state equation is: .

[0063] in, is the material density of the equipment, is the specific heat capacity, is the heat flow, For the moment.

[0064] The first loss function is:

[0065] ;

[0066] ;

[0067] ;

[0068] in, represents the first loss function, is the temperature mechanism equation loss of the equipment in the first loss function, is the virtual and real sensor data loss of the equipment in the first loss function; is the temperature output by the temperature field neural network model, For the moment, For location, is the thermal conductivity, is the heat flow; The first virtual and real node output by the temperature field neural network model At the first sampling moment The temperature, The first virtual node At the first sampling moment The real and virtual temperature data, is the number of the first virtual and real nodes, is the number of first sampling moments; the first sampling moment is the moment when the temperature sensor samples.

[0069] Step 302, establish a mathematical conversion equation between temperature and strain, namely, a temperature-strain conversion equation, and convert the real-time calculation result of the temperature field into a thermal strain load form as an additional input of the stress field neural network model.

[0070] Among them, the temperature-strain conversion equation is:

[0071] ;

[0072] In the formula, is the thermal strain, is the change in temperature, is the coefficient of thermal expansion.

[0073] Step 303, construct the differential equation between strain and stress, i.e., the strain-stress conversion equation, and use it as the additional loss function term of the stress field neural network of the ball screw feed system to drive the efficient intelligent calculation of the stress field, thereby realizing the efficient intelligent calculation of the temperature-stress coupling field of the ball screw feed system of the CNC machine tool. Among them, the physical coupling field calculation model structure based on multi-parameter nesting is as follows: Figure 4 The model structure includes a temperature field neural network model and a stress field neural network model.

[0074] The strain-stress conversion equation is:

[0075] ;

[0076] In the formula, For thermal stress, is Young's modulus.

[0077] In the temperature-strain conversion equation of step 302, since the strain therein is caused by heat, it is called thermal strain. Correspondingly, in step 303, based on the thermal strain, the stress obtained by converting the equation using the strain-stress conversion equation is called thermal stress.

[0078] Step 304, deriving the temperature mechanism equation of the ball screw feed system under the action of factors such as heat conduction and heat convection, and the stress mechanism equation of the ball screw feed system under the action of factors such as assembly force and component gravity.

[0079] Among them, the stress mechanism equation includes the elastic mechanics equation and the contact mechanics equation.

[0080] The elastic equation is: .

[0081] In the formula, is the divergence of thermal stress, describing the distribution change of stress inside the material. is the body force density, which represents the external forces acting on the material (such as gravity), is the second-order time derivative of displacement, representing the acceleration of the material point, is the material density of the equipment.

[0082] The contact mechanics equation is: .

[0083] In the formula, P It represents the pressure per unit area on the contact surface and is used to describe the strength of the force on the contact surface material. It represents the deformation of the contact surface after the force is applied, which is used to calculate the contact stiffness and understand the behavior of the contact surface. R Represents the effective contact radius.

[0084] Step 305, continuously optimizing the stress mechanism equations based on simulation / experiments, etc., simplifying the description of the heat and stress mechanism equations by using unified symbolization, symmetry, linearization and other strategies, and uniformly representing them in the form of differential equations.

[0085] Step 306, embed the temperature mechanism equation, stress mechanism equation, and virtual and real sensor data into the loss function of the stress field neural network model, and use the strain-stress mechanism equation as an additional loss term of the stress field neural network model to constrain the training convergence direction of the stress field of the ball screw feed system and construct a second loss function.

[0086] Among them, the second loss function is:

[0087] ;

[0088] ;

[0089] ;

[0090] in, represents the second loss function, is the stress mechanism equation loss of the equipment in the second loss function, is the virtual and real sensor data loss of the equipment in the second loss function; is the stress output by the stress field neural network model, For the moment, For location, is the material density of the equipment, For thermal stress, is the pressure per unit area on the contact surface, is Young's modulus, is the effective contact radius, is the deformation of the contact surface after the force is applied; The second virtual and real node output by the stress field neural network model At the second sampling time The stress, The second virtual node At the second sampling time The real and imaginary stress data, is the number of the second virtual and real nodes, is the number of the second sampling moments, is the thermal strain; and the second sampling moment is the moment when the stress sensor samples.

[0091] Compared with the prior art, the method provided in the embodiment of the present application has the following beneficial effects:

[0092] 1) This application expands the virtual sensor data of more node positions on the basis of limited node sensor data by constructing real-domain spatiotemporal coding and virtual-real collaborative mapping network, providing sufficient data support for efficient intelligent computing of equipment physical coupling fields.

[0093] 2) This application uses the temperature field calculation results and the strain-stress conversion relationship to standardize the stress field training process, establishes a multi-parameter nesting mechanism under thermal-mechanical coupling, and realizes efficient and explainable calculation of the equipment temperature-stress coupling field.

[0094] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0095] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for intelligent calculation of equipment physical coupling field driven by virtual-real sensor augmentation, characterized in that: include: Collect real sensor data from different real nodes during the service life of the equipment; The real node is a first real node on which a temperature sensor is arranged on the device, or a second real node on which a stress sensor is arranged on the device, and the different real nodes include at least one first real node and one second real node; Encode the real position and real sensing data of each real node to obtain the real sensing position feature of each real node; According to the real sensor position features of different real nodes, a real-domain spatiotemporal coding network is used to generate virtual sensor position features of multiple virtual nodes; The real sensing position features of different real nodes and the virtual sensing position features of different virtual nodes are spliced ​​through the jump connection layer to obtain the virtual and real sensing features of different virtual and real nodes; Based on the virtual-real sensing features of different virtual-real nodes, a virtual-real coding network is used to generate virtual-real sensing data of different virtual-real nodes; The virtual and real sensing data of different virtual and real nodes include: virtual and real temperature data of different first virtual and real nodes and virtual and real stress data of different second virtual and real nodes; According to the temperature mechanism equation, stress mechanism equation and sensor function loss term, a temperature field neural network model and a stress field neural network model are respectively constructed; Calculate the temperature field using a temperature field neural network model based on the virtual and real temperature data of each first virtual and real node; The temperature data of the temperature field is converted into thermal strain data using the temperature-strain conversion equation; Calculating the stress field using a stress field neural network model according to the virtual and real stress data of each second virtual and real node and the thermal strain data; The real-domain spatiotemporal coding network comprises: a real-domain spatiotemporal coding layer and a first multi-layer perceptron; The real-domain spatiotemporal coding layer includes: a first residual connection module, a first layer normalization module, and a first time attention layer, a first channel attention layer, and a first space attention layer connected in sequence; The first input end of the first residual connection module is connected to the output end of the first spatial attention layer, the second input end of the first residual connection module is used to input the real sensor position feature, the output end of the first residual connection module is connected to the input end of the first layer normalization module, and the output end of the first layer normalization module is connected to the input end of the first multi-layer perceptron; The virtual-real coding network includes: a virtual-real coding layer and a second multi-layer perceptron; The virtual-real coding layer includes a second residual connection module, a second normalization module, and a second temporal attention layer, a second channel attention layer, and a second spatial attention layer connected in sequence; The first input end of the second residual connection module is connected to the output end of the second spatial attention layer, the second input end of the second residual connection module is connected to the output end of the jump connection layer, the output end of the second residual connection module is connected to the input end of the second layer normalization module, and the output end of the second layer normalization module is connected to the input end of the second multi-layer perceptron.

2. The method for intelligent calculation of physical coupling field of equipment driven by virtual-real sensing augmentation according to claim 1 is characterized in that: The real position and real sensor data of each real node are encoded to obtain the real sensor position features, including: Encode the real position of each real node to obtain the real position feature of each real node; Encode the real sensing data of each real node to obtain the real sensing features of each real node; The real position features and real sensor features of the same real node are fused to obtain the real sensor position features of each real node.

3. The method for intelligent calculation of physical coupling field of equipment driven by virtual-real sensing augmentation according to claim 2 is characterized in that: The real position characteristics are: ； ； in, For the The real location features of real nodes, For the The position vector of the real node, , , and Respectively The real location of the real node Axis coordinates, Axis coordinates and Axis coordinates, For the The real position encoding vector of the real node, superscript T represents transpose; , Respectively The real location of the real node The 0th order sum of the axis coordinates Taylor expansion value of order; , Respectively The real location of the real node The 0th order sum of the axis coordinates Taylor expansion value of order; , Respectively The real location of the real node The 0th order sum of the axis coordinates Taylor expansion value of order; is the variance of the Taylor expansion, is the Taylor expansion series.

4. The method for intelligent calculation of physical coupling field of equipment driven by virtual-real sensing augmentation according to claim 2 is characterized in that: Encode the real sensor data of each real node to obtain the real sensor features of each real node, including: The real sensing data of each real node is feature expanded and encoded through the convolutional coding layer to obtain the real sensing features of each real node.

5. The method for intelligent calculation of physical coupling field of equipment driven by virtual-real sensing augmentation according to claim 2 is characterized in that: The real position features and real sensor features of the same real node are fused to obtain the real sensor position features of each real node, including: Using the following formula, the real position features and real sensing features of the same real node are fused to obtain the real sensing position features of each real node; ； in, For the The real sensor location features of real nodes, For the The position vector of the real node, For the The real sensing features of real nodes, Indicates The real position features of the real nodes, superscript Indicates transpose.

6. The method for intelligent calculation of physical coupling field of equipment driven by virtual-real sensing augmentation according to claim 1 is characterized in that: The temperature-strain conversion equation is: ； in, is the thermal strain, is the change in temperature, is the coefficient of thermal expansion.

7. The method for intelligent calculation of physical coupling field of equipment driven by virtual-real sensing augmentation according to claim 1 is characterized in that: The temperature field neural network model is obtained through the training of the first loss function; The first loss function is: ； ； ； in ， represents the first loss function, is the temperature mechanism equation loss of the equipment in the first loss function, is the virtual and real sensor data loss of the equipment in the first loss function; is the temperature output by the temperature field neural network model, For the moment, For location, is the thermal conductivity, is the heat flow; The first virtual and real node output by the temperature field neural network model At the first sampling moment The temperature, The first virtual node At the first sampling moment The real and virtual temperature data, is the number of the first virtual and real nodes, is the number of first sampling moments; the first sampling moment is the moment when the temperature sensor samples.

8. The method for intelligent calculation of physical coupling field of equipment driven by virtual-real sensing augmentation according to claim 1 is characterized in that: The stress field neural network model is obtained by training based on the second loss function; The second loss function is: ； ； ； ； in ， represents the second loss function, is the stress mechanism equation loss of the equipment in the second loss function, is the virtual and real sensor data loss of the equipment in the second loss function; is the stress output by the stress field neural network model, For the moment, For location, is the material density of the equipment, For thermal stress, is the pressure per unit area on the contact surface, is Young's modulus, is the effective contact radius, is the deformation of the contact surface after the force is applied; The second virtual and real node output by the stress field neural network model At the second sampling time The stress, The second virtual node At the second sampling time The real and imaginary stress data, is the number of the second virtual and real nodes, is the number of the second sampling moments, is the thermal strain; and the second sampling moment is the moment when the stress sensor samples.

Images