Dynamic unbalanced force prediction modeling method and system for valve element of pneumatic control valve

Abstract

The invention discloses a dynamic unbalance force prediction modeling method and system for a pneumatic control valve element, and relates to the technical field of fluid control, and the method comprises the steps: constructing an initial physical information neural network model; constructing a loss function of the initial physical information neural network model according to the weighted sum of the data fitting item and the physical constraint item; and optimizing the initial physical information neural network model based on the loss function, and gradually increasing the weight of the equation residual loss in the physical constraint term, the dynamic balance data fitting term and the physical constraint term to obtain the physical information neural network model with real-time dynamic unbalance force prediction. According to the method, the loss function is constructed through the weighted sum of the data fitting item and the physical constraint item, model optimization and dynamic adjustment of the weight ratio of the data fitting item and the physical constraint item are carried out in sequence, the real-time dynamic prediction capability of the unbalance force of the valve element is further enhanced, the calculation efficiency is greatly improved while the prediction precision is guaranteed, and the method is suitable for large-scale popularization and application. And a new technical means is provided for dynamic performance prediction and design optimization of the pneumatic control valve.

Description

Technical Field

[0001] The present invention relates to the field of fluid control technology, and in particular to a dynamic unbalanced force prediction modeling method and system for a pneumatic control valve core. Background Art

[0002] Pneumatic control valves, a key component in industrial automation systems, are devices driven by compressed air to control the flow, pressure, or direction of fluids (liquids, gases, or steam). They are widely used in these systems. Their core operating principle is to convert air pressure signals into mechanical displacement via a pneumatic actuator, thereby adjusting the valve opening or switching flow paths. Their performance directly impacts the control accuracy and stability of the entire system. Valve core unbalance is a key factor affecting pneumatic control valve performance. Accurately predicting and effectively controlling valve core unbalance is crucial for improving valve performance.

[0003] In the prior art, there are the following research methods: (1) By analyzing the unbalanced torque at the bottom of the piston valve core, a transient numerical model of the piston valve core is established, and the dynamic flow characteristics and the pressure characteristics of the bottom of the piston valve core as the valve core moves dynamically are studied. The pressure characteristics and the sensitive range of the unbalanced torque under different valve core displacement conditions are obtained. In one method, the influence of different control valve valve core structures on the valve core performance under transient pressure is studied, and the distribution of fluid force and valve core pressure are obtained. Under the same flow characteristics, different valve core structures have certain unbalanced forces under different pressure differences and the fluid force distribution of the valve core is different, with the maximum difference being 140%. (2) To address technical problems such as unstable flow at small openings and uneven force on the valve core, a single-seat control valve with stable flow regulation is studied and designed, and its mechanical and flow characteristics are analyzed. The relationship between the maximum pressure, maximum velocity, maximum density, mass flow rate, volume flow rate and opening from small opening to large opening and small opening inlet is analyzed. (3) By studying the unbalanced force characteristics of the hydraulic spool, the mechanism behind this type of failure was determined. A numerical model of the hydraulic spool was established using sliding grid technology. The model had different control surface structures and spatial angles between the inlet and outlet, and was verified through experiments. The turbulence was modeled using the RNG k-ε model. The simulation results showed that a larger opening weakened the unbalanced force distribution, while a larger inlet flow rate aggravated the unbalanced force distribution. The control surface structure significantly affected the force distribution by changing the throttling effect of the throttling hole. (4) Based on the multi-stage pressure reduction principle and the compressible medium throttling equation, the R&D personnel preliminarily designed the structure of the pilot valve spool assembly and designed the sample space using the orthogonal method. The forces on different samples at different openings were numerically simulated and analyzed. The flow field inside the valve of different experimental samples of the control valve at the optimal working opening was calculated using ANSYS CFX software. The flow rate inside the valve and the force on the valve spool of different experimental samples were obtained. It was concluded that the structural parameters affecting the stability of the valve spool mainly include the equivalent diameter of the main valve spool opening, the effective diameter of the pilot valve seat throttling, and the pilot valve damping hole. (5) A high-fidelity computational fluid dynamics (CFD) numerical model was established to study the flow characteristics of a new type of balanced stop valve. The analysis showed that the orifice size is the key factor affecting the fluid force. At the same time, the influence of the orifice size on the fluid force was studied, and the coupling effect of different valve disc displacements and inlet pressures on the fluid force was quantitatively analyzed. (6) A three-dimensional model of the flow field and solid field was established to study the reliability of the GPV valve core in engineering applications. The flow field characteristics and the stability of the valve core were analyzed using a bidirectional fluid-solid coupling calculation method. As the valve core opens, vortex structures gradually form on both sides of the upper diaphragm, and the vibration intensity of the valve core gradually increases with the formation of the vortex. (7) Based on the Euler-Euler model, the solid-liquid flow characteristics of the hydraulic valve fitting clearance and the valve core stuck were analyzed.The effects of particle concentration and diameter on flow characteristics and valve core adhesion were analyzed. It was found that the highest volume fraction of particles was in the pressure equalization tank (PEG). The peak value increased with the increase of particle size, and the adhesion increased with the increase of particle concentration. (8) A fluid-solid coupling model was established using Ansys WorkBench. Static structural simulation analysis of the valve sleeve and valve core before and after structural improvement and parameter optimization was performed. Mathematical models of triangular buffer tank, U-shaped buffer tank and combined buffer tank were established. The bird flock optimization algorithm was used to optimize the structural parameters of the combined buffer tank. The analysis showed that the triangular buffer tank had a good pressure reduction effect, but the impact was large. The pressure of the U-shaped buffer tank was stable and mild, but the pressure reduction effect was not ideal. The combined buffer tank had a significant pressure reduction effect and good stability. (9) A new type of plug-in valve core structure was proposed to solve the problem of large lateral forces that easily affect the dynamic response speed of the valve core in the flow field of the eccentric conical gap. The flow characteristics of the gap flow field in terms of velocity distribution, pressure distribution, leakage, etc. were analyzed by numerical simulation method. A guide groove was set on the surface of the plug-in valve core to increase the connection length of the valve core, forming a uniform radial pressure distribution and velocity distribution, effectively reducing the lateral force. (10) A two-dimensional axisymmetric flow model of the hydraulic valve was established using ANSYS Fluent, and the flow field characteristics under different oil temperatures were studied. The analysis showed that as the oil temperature increased, the oil viscosity decreased, the flow rate and maximum flow rate in the flow channel also increased, while the turbulent kinetic energy at the valve port decreased. As the oil temperature increased, the axial force acting on the valve core decreased, and the steady-state fluid force also decreased. (11) A single-seat control valve with stable flow regulation was studied and designed, and the mechanical characteristics and flow characteristics were analyzed. The fluid flow simulation analyzed the fluid movement of the single-seat control valve under various conditions. The valve core movement simulation analyzed the changing trends of unbalanced force, linear displacement, etc. The simulation results were compared with the experimental results and the simulation results were consistent with the experimental results. (12) The effect of the eccentric undercut groove of the valve body on the radial unbalanced force of the sliding valve core was studied. By simplifying the fluid domain of the sliding valve and using numerical simulation methods to establish a transient flow analysis model of the sliding valve, the Bernoulli effect was used to analyze the uneven pressure distribution on the valve core surface under different eccentric grooves, and the calculation formula of the unbalanced force was obtained. The results showed that the eccentric groove reduced the turbulence level of the surface area under the outlet section, thereby effectively reducing the radial unbalanced force.

[0004] The above-mentioned traditional valve core unbalance force prediction method mainly relies on computational fluid dynamics (CFD) simulation. Although simulation research can provide relatively accurate results, the calculation cost is high and time-consuming, and the valve core unbalance force is easily affected by many factors. As a result, the traditional valve core unbalance force prediction method is difficult to meet the needs of real-time control and optimization of the start-up control valve. Summary of the Invention

[0005] The purpose of the present invention is to address the above-mentioned deficiencies in the prior art and to provide a method and system for predicting and modeling the dynamic unbalanced force of a pneumatic control valve core, so as to solve the problems in the prior art.

[0006] The present invention specifically provides the following technical solutions: The valve geometry parameters, fluid properties and working conditions of the pneumatic control valve are used as input variables, and the unbalanced force of the pneumatic control valve core is used as output variable. An initial physical information neural network model is constructed through a convolutional neural network. The loss function of the initial physical information neural network model is constructed as the weighted sum of the data fitting term and the physical constraint term. The data fitting term is the mean square error of the pressure field, valve opening, and fluid density. The physical constraint term includes the residual loss of the equations for mass conservation and directional momentum, the boundary condition loss for velocity and pressure, and the embedded unbalanced force calculation. The initial physical information neural network model is optimized based on the loss function, and the weight of the equation residual loss in the physical constraint term, the dynamic balance data fitting term and the physical constraint term are gradually increased to obtain a physical information neural network model with real-time dynamic unbalanced force prediction.

[0007] Preferably, in the physical constraint terms, the residual loss of the equations of mass conservation and directional momentum, the boundary condition loss of velocity and pressure, and the calculation of the unbalanced force are embedded in the calculation process as follows: Taking continuity residual and momentum residual as equation residual loss, the specific expression is: ; ; Among them, among them, is the continuity residual, is the momentum residual, is the total number of grid cells, The velocity field is x The direction component, The velocity field is y The direction component, The velocity field is z The direction component, is the spatial coordinate, For the i The volume of a grid cell, is the pressure field, is the dynamic viscosity, is the velocity component u The Laplace operator of is the convection operator of the velocity field; Boundary condition losses include Dirichlet boundary and Neumann boundary. The Dirichlet boundary is the valve core surface velocity loss, and the Neumann boundary is the pressure gradient loss. The specific expressions are: ; ; in, is the Dirichlet boundary loss, is the total number of discrete points on the Dirichlet boundary, is the three-dimensional coordinate of a point on the Dirichlet boundary, For the Dirichlet boundary point i The target speed value specified at For time, For the pressure field At the border point i Normal gradient at ; For the Neumann boundary point i The target normal flux value specified at ; For boundary points i The unit external normal vector at , is the total number of discrete points on the Neumann boundary; The unbalanced force formula is used as a post-processing layer or loss term, and the integral approximation is calculated at the sampling points on the valve core surface. The specific expression is: ; ; in, is the predicted total force or flux, is the total number of surface elements on the boundary or surface after discretization, For the i The pressure force vector on the surface element is is the pressure scalar at the surface element, is the unit external normal vector, is the dynamic viscosity coefficient, The velocity field u is i The gradient tensor at the bin, For the i The area of ​​a surface element, is the force prediction loss function, is the total force obtained from experiments or actual measurements.

[0008] Preferably, the loss function of the initial physical information neural network model is constructed by the weighted sum of the data fitting term and the physical constraint term, and the specific expression is: ; in, is the loss function of the initial physical information neural network model, is the data fitting loss, is the weight of the data fitting loss, is the weight of the continuity residual, is the weight of the momentum residual, is the weight of the Dirichlet boundary loss, is the weight of the force prediction loss function.

[0009] Preferably, when gradually increasing the weight of the residual loss of the equation in the physical constraint term, residual adaptive sampling is used to focus on the high error area, so as to dynamically balance the data fitting term and the physical constraint term by focusing on the high error area.

[0010] Preferably, when the valve geometric parameters, fluid properties and working conditions of the pneumatic control valve are used as input variables, the valve geometric parameters determine the spatial flow field structure, the fluid properties are affected by the working conditions, and the relationship between the valve geometric parameters, fluid properties and working conditions is dynamically adjusted by changing the valve opening over time; the spatial coordinates in the spatial flow field structure determine the distribution of flow field parameters including velocity, pressure and temperature.

[0011] The present invention provides a dynamic unbalanced force prediction modeling system for a pneumatic control valve core, comprising: An initial model building module is used to build an initial physical information neural network model through a convolutional neural network using the valve geometry parameters, fluid properties, and working conditions of the pneumatic control valve as input variables and the unbalanced force of the pneumatic control valve spool as output variables; A loss function definition module is used to construct the loss function of the initial physical information neural network model using the weighted sum of data fitting terms and physical constraint terms. The data fitting term is the mean square error of the pressure field, valve opening, and fluid density. The physical constraint terms include the residual loss of the equations for mass conservation and directional momentum, the boundary condition loss for velocity and pressure, and the embedded unbalanced force calculation. The model optimization module is used to optimize the initial physical information neural network model based on the loss function, and gradually increase the weight of the equation residual loss in the physical constraint term, the dynamic balance data fitting term and the physical constraint term to obtain a physical information neural network model with real-time dynamic unbalanced force prediction.

[0012] The present invention provides a computer device comprising a memory and a processor, wherein the memory stores a program, and when the program is executed by the processor, the processor executes the steps of the above-mentioned method for predicting and modeling the dynamic unbalanced force of a pneumatic control valve core.

[0013] The present invention provides a storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the above-mentioned method for predicting and modeling the dynamic unbalanced force of a pneumatic control valve core are realized.

[0014] Compared with the prior art, the present invention has the following significant advantages: The present invention embeds the basic equations of fluid mechanics into a neural network, constructs a loss function through the weighted sum of data fitting terms and physical constraint terms, optimizes the model in sequence, and dynamically adjusts the weight ratios of data fitting terms and physical constraint terms, thereby further enhancing the real-time dynamic prediction capability of valve core unbalanced force and the effect of real-time updating of valve core unbalanced force. While ensuring the prediction accuracy, the calculation efficiency is greatly improved, providing a new technical means for the dynamic performance prediction and design optimization of pneumatic control valves.

[0015] Figure Description Figure 1 This is a simplified structural diagram of the valve core in an embodiment of the present invention; Figure 2 This is a physical information neural network diagram in an embodiment of the present invention; Figure 3 This is a graph of model learning efficiency in an embodiment of the present invention; wherein, Figure 3 (a) is the PINN Solution diagram, Figure 3 (b) is the Real Solution diagram; Figure 4 This is a pressure diagram of the valve body structure in an embodiment of the present invention; wherein, Figure 4 (a) is the front view, Figure 4 (b) is the left view, Figure 4 (c) is the right side view; Figure 5 FIG. 1 is a force diagram of the valve core in an embodiment of the present invention; Figure 5 (a) Figure 5 (b) and Figure 5 (c) are the force diagrams of the valve core at different perspectives; Figure 6 This is a flow chart of a dynamic unbalanced force prediction modeling method for a pneumatic control valve core provided by the present invention. DETAILED DESCRIPTION

[0016] The following is a clear and complete description of the technical solutions of the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0017] Physical information neural network has been applied to fluid modeling to a certain extent and has shown great superiority. This invention aims to explore the use of PINN to establish a fluid prediction model for the unbalanced force of the valve core of a pneumatic control valve, and provide new technical means and reference methods for the structural design and optimization of pneumatic control valves.

[0018] 1. Analysis of unbalanced force of pneumatic control valve core: A pneumatic control valve is a device that regulates fluid flow to achieve process control. Its core component is the valve core. The valve core is subject to various forces under the action of the fluid, among which unbalanced force is a key factor affecting valve performance. This unbalanced force, primarily caused by the uneven distribution of fluid dynamic and static pressures, can make it difficult to control the valve core's position, affecting the valve's adjustment accuracy and stability.

[0019] 1.1 Fluid statics analysis: When the valve core is at a certain opening, the fluid flowing through the valve port generates a pressure drop (Δ P Assuming the fluid is incompressible steady-state flow and ignoring viscous friction, the axial force on the valve core can be derived according to the Bernoulli equation and the continuity equation:

[0020] (1); in, and are the pressures on the upper and lower ends of the valve core, A 1. A 2 is the corresponding effective area. Due to the geometric asymmetry of the valve core (such as conical valve core), A 1≠ A 2. Leading to static unbalanced force.

[0021] 1.2 Fluid dynamics analysis: Considering the dynamic effect of the fluid, the flow field change caused by the valve core displacement will generate additional inertia force and momentum force. According to the law of conservation of momentum, the dynamic unbalanced force of the valve core is:

[0022] (2); Where, p is the fluid density, Q is the volume flow rate, V1 and V2 are the flow rates at the valve inlet and outlet respectively.

[0023] 1.3 Total unbalanced force: Combining the static and dynamic components, the total unbalanced force is expressed as: (3); in, Is the friction resistance between the valve core and the valve seat. Figure 1 Shown is a simplified structural diagram of the valve core of a flow-to-open control valve.

[0024] The expression of the axial unbalanced force of the valve core is: (4); Where, is the axial unbalanced force of the valve core, 、 are the pressure before and after the valve respectively. d g 、 d s They are the valve core and valve stem diameters, It is the pressure difference before and after the valve.

[0025] By simplifying the valve core structure analysis, the calculation formula for the valve core unbalanced force is obtained. From the formula, it can be seen that the valve core unbalanced force is mainly caused by the pressure and pressure difference before and after the valve, which lays the foundation for the next step of establishing an unbalanced force prediction model.

[0026] 1.4 Analysis of influencing factors The unbalanced force of the control valve core is affected by many factors, mainly the following: (1) Valve structure: The structure of the valve has a significant impact on the unbalanced force. This includes the shape of the valve core, the throttling type, the valve core assembly method (normal or reverse installation), the relationship between the valve stem diameter and the valve seat diameter, etc. These factors will affect the flow state of the fluid in the valve and the force exerted on the valve core.

[0027] (2) Valve inlet pressure and valve front-to-back pressure differential: Valve inlet pressure and valve front-to-back pressure differential are two other important factors that affect unbalanced force. During the flow of the medium in the regulating valve body, part of the kinetic energy will be converted into thermal potential energy due to the resistance of the valve core and the friction between the medium and the regulating valve body, resulting in changes in the flow rate and pressure in the pipelines before and after the regulating valve body. This pressure differential acts on the valve core, generating an unbalanced force.

[0028] (3) Relative flow direction of the fluid and the valve core: The relative flow direction of the fluid and the valve core will also affect the unbalanced force. When the flow direction of the medium changes, it will change the pressure distribution before and after the valve, as well as the direction of the medium's flow around the valve core, thereby affecting the magnitude and direction of the unbalanced force.

[0029] (4) Medium type and state: The type and state of the medium (such as temperature, density, viscosity, etc.) will also affect the unbalanced force. Different media will generate different dynamic pressures and static pressures during the flow process, thereby affecting the unbalanced force on the valve core.

[0030] In summary, the unbalanced force in control valves is influenced by many factors, including valve structure, upstream pressure and differential pressure across the valve, the relative flow direction of the fluid and valve core, and the type and state of the medium. These factors interact to influence the performance and stability of the control valve. When designing and using a control valve, these factors must be comprehensively considered to ensure proper operation and long-term service.

[0031] The mechanism that generates valve core unbalance forces is complex, involving multiple disciplines such as fluid mechanics and thermodynamics. During valve operation, fluid flowing through the valve core generates uneven pressure distribution, resulting in radial and axial unbalanced forces acting on the valve core. The magnitude and direction of these forces vary with valve opening, fluid properties, and operating conditions, posing challenges to precise valve control. Accurately predicting valve core unbalance forces is crucial for optimizing valve design and improving control precision.

[0032] 8. Basic principles of physical information neural network (PINN): Physically Informed Neural Network (PINN) is a new machine learning method that embeds physical laws into neural networks. It combines deep learning with physical modeling to solve partial differential equations (PDEs) and other physical problems. Compared with traditional neural networks, PINN can not only learn patterns in data, but also obey known physical laws. The structure of physical information neural network is as follows: Figure 2 The key idea of ​​PINN is to embed physical constraints into a neural network, enabling it to learn the behavior of physical systems and satisfy physical equations. This approach is often used in situations where data is scarce or the problem is complex. This characteristic makes PINN outstanding in solving scientific computing problems involving partial differential equations. It is particularly effective in maintaining high prediction accuracy in data-scarce environments, providing a powerful numerical solution tool for science and engineering.

[0033] The basic principle of PINN is to add physical constraints to the loss function of the neural network so that the network can simultaneously satisfy data fitting and physical laws during training. Specifically, the loss function of PINN usually includes a data fitting term and a physical constraint term. The data fitting term measures the difference between the network's predicted value and the actual observed value, while the physical constraint term ensures that the network output satisfies the relevant physical equations. Through this dual constraint, PINN can learn solutions that conform to physical laws with a small amount of data, thereby improving the model's generalization ability and prediction accuracy. PINN achieves collaborative learning of physical laws and data features by incorporating the control equation as a soft constraint into the neural network loss function. The valve core area fluid control equation is used to establish a simplified NS equation for the flow field inside the axisymmetric valve.

[0034] 3. Construction of unbalanced force fluid model of pneumatic control valve core based on PINN: As Figure 6 shown in the figure, the dynamic unbalanced force prediction modeling method of the pneumatic control valve spool in the embodiment includes the following steps: Step S1: Taking the valve geometric parameters, fluid properties and working conditions of the pneumatic control valve as input variables, and taking the unbalanced force of the pneumatic control valve spool as output variable, an initial physical information neural network model is constructed by convolutional neural network.

[0035] 3.1 Physical problem modeling: 3.1.1 Control equation (Navier-Stokes equation), the fluid dynamic behavior is described by continuity equation (mass conservation) and momentum equation.

[0036] The specific expression of the continuity equation (mass conservation) is: (5); The specific expression of the momentum equation is: (6); Where, u is the velocity field, p is the pressure field, p is the fluid density, m is the dynamic viscosity, f is the body force.

[0037] 3.1.2 Unbalanced force calculation: The unbalanced force is obtained by integrating the fluid pressure and shear stress on the surface of the spool, and the specific expression is: (7); In the formula, S is the surface of the spool, n is the surface normal vector.

[0038] 3.1.3 Boundary conditions: Inlet / outlet: velocity or pressure boundary (uinlet=U0), select pressure study.

[0039] Spool surface: no-slip condition ( u =0).

[0040] Symmetry axis / far field: free slip or periodic condition.

[0041] 3.2 PINN model construction, where the input and output of the neural network are specifically: Input: spatial coordinates ( x , y , z ), time t , valve opening α (if dynamic change).

[0042] Output: velocity field u , pressure field p .

[0043] Spatial coordinates: determine the distribution of flow field parameters (velocity, pressure, temperature), and are affected by geometric parameters and flow state.

[0044] Time: Introduce dynamic effects (such as unsteady flow and heat conduction) and couple them with the opening changes.

[0045] Valve opening: As a core variable, it connects time and space and directly affects the flow field and working conditions by changing the flow area.

[0046] Coupling relationship: The valve geometric parameters determine the spatial flow field structure. The fluid properties are affected by the working conditions. By changing the valve opening over time, the relationship between the valve geometric parameters, fluid properties and working conditions is dynamically adjusted, forming a multi-physical field coupling problem; the spatial coordinates in the spatial flow field structure determine the distribution of flow field parameters including velocity, pressure and temperature.

[0047] The unbalanced force on the valve core is essentially a manifestation of the asymmetry of the pressure field, which is tightly coupled to the velocity field through the Bernoulli equation and the conservation of momentum. The magnitude and direction of the unbalanced force are determined by the valve core geometry, fluid properties, pressure differential, and flow conditions.

[0048] Step S2: Construct the loss function of the initial physical information neural network model with the weighted sum of the data fitting term and the physical constraint term; the data fitting term is the mean square error of the pressure field, valve opening and fluid density, and the physical constraint term includes the residual loss of the equations of mass conservation and directional momentum, the boundary condition loss of velocity and pressure, and the embedding of unbalanced force calculation.

[0049] 3.3 Physical Information Embedding: 3.3.1 Equation residual loss: The equation residual is calculated through automatic differentiation to force the network to satisfy the control equation.

[0050] Continuity residuals: (8); in is the grid cell volume.

[0051] Momentum residual: (9); in, is the continuity residual, is the momentum residual, is the total number of grid cells, The velocity field is x The direction component, The velocity field is y The direction component, The velocity field is z The direction component, is the spatial coordinate, For the i The volume of a grid cell, is the pressure field, is the dynamic viscosity, is the velocity component u The Laplace operator of is the convection operator of the velocity field.

[0052] 3.3.2 Boundary condition loss: Boundary condition losses include Dirichlet boundary and Neumann boundary. The Dirichlet boundary is the valve core surface velocity loss, and the Neumann boundary is the pressure gradient loss. The specific expressions are: Dirichlet boundary (spool surface velocity): (10); Neumann boundary (pressure gradient): (11); in, is the Dirichlet boundary loss, is the total number of discrete points on the Dirichlet boundary, is the three-dimensional coordinate of a point on the Dirichlet boundary, For the Dirichlet boundary point i The target speed value specified at For time, For the pressure field At the border point i Normal gradient at ; For the Neumann boundary point i The target normal flux value specified at ; is the unit normal vector at the boundary point i, is the total number of discrete points on the Neumann boundary.

[0053] 3.3.3 Unbalanced force calculation embedding: Formulate the unbalanced force as a post-processing layer or loss term: (12); At the sampling point on the valve core surface, calculate the integral approximation: (13); in, is the predicted total force (or flux), is the total number of surface elements on the boundary or surface after discretization, For the i The pressure force vector on the surface element is is the pressure scalar at the surface element, is the unit external normal vector, is the dynamic viscosity coefficient, Velocity field u In the i The gradient tensor at the bin, For the i The area of ​​a surface element, is the force prediction loss function, is the total force obtained from experiments or actual measurements.

[0054] 3.4 Loss Function Design: The total loss function is the weighted sum of each loss term: (14); Among them, the weight l Parameters need to be adjusted to balance convergence stability (such as =1.0, =10.0), is the loss function of the initial physical information neural network model, loss, is the weight of the data fitting loss, is the weight of the continuity residual, is the weight of the momentum residual, is the weight of the Dirichlet boundary loss, is the weight of the force prediction loss function. The physical constraints in PINN directly correspond to the residuals of the governing equations, which are equivalent to the unbalanced forces in the system. By optimizing the loss function, PINN brings the model predictions toward physical equilibrium (where the unbalanced forces approach zero) while also fitting the experimental data.

[0055] Step S3: Optimize the initial physical information neural network model based on the loss function, and gradually increase the weight of the equation residual loss in the physical constraint term, the dynamic balance data fitting term and the physical constraint term to obtain a physical information neural network model with real-time dynamic unbalanced force prediction.

[0056] 3.5 Training and Validation: 3.5.1 Data preparation: Simulation / experimental data: velocity and pressure data from CFD or sensors.

[0057] Data-free training: relying only on physical equation residuals and boundary conditions.

[0058] 3.5.2 Training strategy: Using the Adam optimizer, we gradually increase the residual weight of the equation and use residual adaptive sampling (RAR) to focus on high error areas. The training results are as follows: Figure 3 shown.

[0059] 3.5.3 Verification indicators: Residual convergence curve, compared with CFD results (unbalanced forces).

[0060] Building a PINN-based fluid model for the unbalanced spool force of a pneumatic control valve involves the following steps: First, determine the network input and output variables and network structure. Input variables typically include valve geometry, fluid properties, and operating conditions, while the output variable is the unbalanced spool force. The network structure can employ a multilayer perceptron or convolutional neural network, with the specific design tailored to the complexity of the problem.

[0061] Next, define the loss function. This should include both a data fitting term and a physical constraint term. Common loss functions such as mean squared error can be used for the data fitting term, while the physical constraint term should be constructed based on the fundamental equations of fluid dynamics (the Navier-Stokes equations). By incorporating the residuals of the physical equations into the loss function, we ensure that the network output complies with the laws of fluid dynamics.

[0062] Finally, a training strategy is developed. Because the PINN loss function typically contains multiple competing terms, an appropriate weight distribution strategy and optimization algorithm are required. An adaptive weight adjustment method is used to dynamically balance data fitting and physical constraints during training. Adam is selected as the optimization algorithm to improve training efficiency and model accuracy. The following experiments verify and analyze the results, as follows:

[0063] To verify the effectiveness of the proposed method, the present invention conducted numerical experiments. The experimental data comes from the CFD simulation results of a certain type of pneumatic control valve, covering the valve core unbalanced force under different valve openings, fluid velocities and pressure conditions. The present invention divides the data set into a training set and a test set for model training and performance evaluation. The CFD simulation results are shown in Figure 2. Figure 4 and Figure 5 The comparison of the results is shown in Table 1.

[0064] Table 1 Comparison of results In terms of trend consistency, the PINN and CFD curves closely match in the mid-range of valve opening (30%-70%), demonstrating that PINN accurately learns physical laws under key operating conditions. Deviations at the extreme ends (<10% or >90% opening) may be due to complex boundary conditions (such as turbulent separation), which can lead to a decrease in PINN's generalization ability.

[0065] Error analysis: The maximum relative error occurs at the critical closing point (5% opening), with an error of approximately 8%, which may be related to the nonlinear flow field strength at low openings. The mean absolute error (MAE) is 1.2N, verifying the engineering practicality of PINN.

[0066] Efficiency comparison: Table note: A single CFD simulation takes 16 hours (mesh size 1 million), while a PINN prediction takes only 0.1 seconds (trained model).

[0067] It shows that the flow field pressure distribution predicted by PINN is highly consistent with the CFD results, and the maximum relative error occurs in the valve seat edge area (<8%).

[0068] Experimental results demonstrate that the PINN-based model accurately predicts valve core unbalance forces, with an average relative error of less than 8% on a test set. Compared to traditional CFD methods, the PINN model achieves a computational speed increase of two orders of magnitude while maintaining high accuracy. Furthermore, the authors compared the performance of the PINN model with a purely data-driven neural network model, finding that the PINN model exhibits superior generalization capabilities in data-scarce conditions.

[0069] Through analysis, the present invention found that the PINN model can accurately capture the pressure distribution characteristics of the flow field around the valve core, which explains its superior performance in predicting unbalanced forces. However, the study also found that under extreme operating conditions (ultra-fast flow), the PINN model's prediction accuracy decreases. This may be due to insufficient training data coverage or imperfect physical constraints.

[0070] In another exemplary embodiment, based on the same inventive concept as the method embodiment, the present invention proposes a dynamic unbalanced force prediction modeling system for a pneumatic control valve core, including: an initial model building module, a loss function definition module and a model optimization module.

[0071] Among them, the initial model construction module is used to construct the initial physical information neural network model through a convolutional neural network with the valve geometric parameters, fluid properties and working conditions of the pneumatic control valve as input variables and the unbalanced force of the pneumatic control valve core as output variables; the loss function definition module is used to construct the loss function of the initial physical information neural network model with the weighted sum of data fitting terms and physical constraint terms; the data fitting term is the mean square error, and the physical constraint term includes equation residual loss, boundary condition loss and unbalanced force calculation embedding; the model optimization module is used to optimize the initial physical information neural network model based on the loss function, and gradually increase the weight of the equation residual loss in the physical constraint term, dynamically balance the data fitting term and the physical constraint term, and obtain a physical information neural network model with real-time dynamic unbalanced force prediction.

[0072] In another exemplary embodiment, based on the same inventive concept as the method embodiment, the present invention also provides a computer device, including a memory and a processor, wherein a program is stored in the memory, and when the program is executed by the processor, the processor executes the steps of a dynamic unbalanced force prediction modeling method for the valve core of a pneumatic control valve.

[0073] According to the disclosed embodiments, a computing device can communicate with one or more external devices (e.g., a keyboard, pointing device, Bluetooth communication, etc.), or with any device that enables a computing device to communicate with one or more other computing devices (e.g., a router, a modem, etc.). The processor can be a single-core or multi-core central processing unit or a specific integrated circuit, or one or more integrated circuits configured to implement the present invention. The subject matter and functional operations described in this specification can be implemented in tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or a combination of one or more of these. The subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-transitory program carrier for execution by or control of the operation of a data processing device.

[0074] In another exemplary embodiment, based on the same inventive concept as the method embodiment, ordinary technicians in this field can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. That is, the present invention also provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned dynamic unbalanced force prediction modeling method of the valve core of a pneumatic control valve are implemented.

[0075] Based on this understanding, the technical solution of this embodiment, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0076] The above content is a further detailed description of the present invention in combination with specific preferred embodiments. For those skilled in the art to which the present invention belongs, several simple deductions or replacements can be made without departing from the concept of the present invention, which should be regarded as falling within the scope of protection of the present invention.

Claims

1. A method for predicting the dynamic unbalanced force of a pneumatic control valve core, characterized in that: include: An initial physical information neural network model is constructed with the valve geometric parameters, fluid properties and working conditions of the pneumatic control valve as input variables and the unbalanced force of the pneumatic control valve core as output variable. The loss function of the initial physical information neural network model is constructed as the weighted sum of the data fitting term and the physical constraint term. The data fitting term is the mean square error of the pressure field, valve opening, and fluid density. The physical constraint term includes the residual loss of the equations for mass conservation and directional momentum, the boundary condition loss for velocity and pressure, and the embedded unbalanced force calculation. The initial physical information neural network model is optimized based on the loss function, and the weight of the equation residual loss in the physical constraint term, the dynamic balance data fitting term and the physical constraint term are gradually increased to obtain a physical information neural network model with real-time dynamic unbalanced force prediction.

2. A method for predicting and modeling the dynamic unbalanced force of a pneumatic control valve core according to claim 1, characterized in that: In the physical constraints, the residual loss of the equations for mass conservation and directional momentum, the boundary condition loss for velocity and pressure, and the embedded calculation of unbalanced forces are calculated as follows: Taking continuity residual and momentum residual as equation residual loss, the specific expression is: ； ； Among them, among them, is the continuity residual, is the momentum residual, is the total number of grid cells, The velocity field is x The direction component, The velocity field is y The direction component, The velocity field is z The direction component, is the spatial coordinate, For the i The volume of a grid cell, is the pressure field, is the dynamic viscosity, is the velocity component u The Laplace operator of is the convection operator of the velocity field; Boundary condition losses include Dirichlet boundary and Neumann boundary. The Dirichlet boundary is the valve core surface velocity loss, and the Neumann boundary is the pressure gradient loss. The specific expressions are: ； ； in, is the Dirichlet boundary loss, is the total number of discrete points on the Dirichlet boundary, is the three-dimensional coordinate of a point on the Dirichlet boundary, For the Dirichlet boundary point i The target speed value specified at For time, For the pressure field At the border point i Normal gradient at ; For the Neumann boundary point i The target normal flux value specified at ; For boundary points i The unit external normal vector at , is the total number of discrete points on the Neumann boundary; The unbalanced force formula is used as a post-processing layer or loss term, and the integral approximation is calculated at the sampling points on the valve core surface. The specific expression is: ； ； in, is the predicted total force or flux, is the total number of surface elements on the boundary or surface after discretization, For the i The pressure force vector on the surface element is is the pressure scalar at the surface element, is the unit external normal vector, is the dynamic viscosity coefficient, The velocity field u is i The gradient tensor at the bin, For the i The area of ​​a surface element, is the loss function for force prediction, is the total force obtained from experiments or actual measurements.

3. The dynamic unbalanced force prediction modeling method of a pneumatic control valve core according to claim 2, characterized in that: The loss function of the initial physical information neural network model is constructed by the weighted sum of the data fitting term and the physical constraint term. The specific expression is: ； in, is the loss function of the initial physical information neural network model, is the data fitting loss, is the weight of the data fitting loss, is the weight of the continuity residual, is the weight of the momentum residual, is the weight of the Dirichlet boundary loss, is the weight of the force prediction loss function.

4. The method for predicting and modeling the dynamic unbalanced force of a pneumatic control valve core according to claim 1, wherein: When gradually increasing the weight of the residual loss of the equation in the physical constraint term, residual adaptive sampling is used to focus on the high error area, so as to dynamically balance the data fitting term and the physical constraint term by focusing on the high error area.

5. The dynamic unbalanced force prediction modeling method of a pneumatic control valve core according to claim 1, characterized in that: When the valve geometric parameters, fluid properties and working conditions of the pneumatic control valve are used as input variables, the valve geometric parameters determine the spatial flow field structure, and the fluid properties are affected by the working conditions. By changing the valve opening over time, the relationship between the valve geometric parameters, fluid properties and working conditions is dynamically adjusted; The spatial coordinates in the spatial flow field structure determine the distribution of flow field parameters including velocity, pressure and temperature.

6. A dynamic unbalanced force prediction modeling system for a pneumatic control valve core, characterized in that: include: An initial model building module is used to build an initial physical information neural network model through a convolutional neural network using the valve geometry parameters, fluid properties, and working conditions of the pneumatic control valve as input variables and the unbalanced force of the pneumatic control valve spool as output variables; A loss function definition module is used to construct the loss function of the initial physical information neural network model using the weighted sum of data fitting terms and physical constraint terms. The data fitting term is the mean square error of the pressure field, valve opening, and fluid density. The physical constraint terms include the residual loss of the equations for mass conservation and directional momentum, the boundary condition loss for velocity and pressure, and the embedded unbalanced force calculation. The model optimization module is used to optimize the initial physical information neural network model based on the loss function, and gradually increase the weight of the equation residual loss in the physical constraint term, the dynamic balance data fitting term and the physical constraint term to obtain a physical information neural network model with real-time dynamic unbalanced force prediction.

7. A computer device, characterized in that: It includes a memory and a processor, wherein a program is stored in the memory, and when the program is executed by the processor, the processor executes the steps of a dynamic unbalanced force prediction modeling method for a pneumatic control valve core as described in any one of claims 1 to 5.

8. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a dynamic unbalanced force prediction modeling method for a pneumatic control valve core according to any one of claims 1 to 5 are implemented.

Images