Design method of non-eccentric structure self-generated torque type air flow measurement and control integrated valve

Abstract

The application discloses a design method of a non-eccentric self-generated torque type air volume measurement and control integrated valve, and comprises the following steps: on the basis of the structure of an existing non-eccentric air valve, the valve shaft of the air valve is kept unchanged at the geometric center of the valve plate, the geometric shape of the valve plate is changed, the torque is increased, and the fast prediction of the performance parameters of the valve is realized under any valve plate geometric shape by combining parameterized modeling, computational fluid dynamics (CFD) simulation, symbolic distance field (SDF) representation and a neural network prediction model. The non-dominated sorting genetic algorithm (NSGA-II) is used to obtain the optimal valve plate geometric design. Compared with the prior art, the calculation efficiency of the application can be improved by 300-500 times. The non-eccentric torque valve developed based on the application avoids the problem of excessive torque of the eccentric air valve under the large air volume working condition, the air volume prediction accuracy is improved to about 4%, the torque can be increased by 4-5 times, and the energy-saving effect is achieved under the small opening condition (30°-50°), so that the application is suitable for accurate air volume measurement and control of a variable air volume air conditioning system.

Description

Technical Field

[0001] This invention relates to the field of building equipment optimization, and more particularly to a design method for a non-eccentric structure self-generating torque-type integrated air volume measurement and control valve. Background Technology

[0002] Variable air volume (VAV) air conditioning systems have been widely used in various types of buildings due to their advantages of good air quality, excellent energy-saving performance, and flexible zone control. High-precision airflow sensors, suitable airflow regulating valves, and reasonable valve control algorithms are key to fully realizing the energy-saving and comfort advantages of VAV systems.

[0003] Traditional airflow regulation separates monitoring and control, which not only increases the energy consumption of ventilation system equipment but also makes intelligent control more complex. As integrated airflow measurement and control becomes a trend, valve airflow sensors have been developed. Among these, using valve torque to indirectly measure airflow is considered a highly accurate method.

[0004] Ordinary non-eccentric dampers have relatively low torque, placing high demands on the accuracy of torque sensors. Therefore, most existing torque dampers employ an eccentric design to increase the torque signal for airflow measurement. However, in high-volume ventilation systems, the torque can become excessive, even exceeding the maximum driving torque specified in the "Standard for Building Ventilation Airflow Regulating Valves JG / T 436-2014," making the valve difficult to open or close. Furthermore, the flow characteristics of existing torque dampers are linear, leaving room for further optimization in practical ventilation network control technology. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a design method for a non-eccentric structure self-generating torque integrated air volume measurement and control valve that ensures accurate air volume prediction and has equal percentage flow characteristics.

[0006] To achieve the objectives of this invention, the following technical solution is adopted:

[0007] The present invention discloses a design method for a non-eccentric structure self-generating torque type integrated air volume measurement and control valve, comprising the following steps:

[0008] Step 1: Define the Cartesian coordinate system as follows: The endpoint of one side of the central valve axis of the non-eccentric valve is defined as the origin O, the horizontal direction is the x-axis, the vertical direction is the y-axis, the direction parallel to the central valve axis is the z-axis, and the valve opening θ is defined as the acute angle between the valve plate and the negative y-axis direction.

[0009] The valve plate structure is as follows: the shape is fixed along the z-axis, and the valve plates on both sides of the central valve shaft are rotationally symmetrical around the central valve shaft. The valve plate width w and valve plate thickness d are both kept constant. A non-uniform rational B-spline curve generated by five coordinate points is used to characterize the shape of the valve plate located on one side of the central valve shaft in the xOy plane. The five coordinate points are all located on the outer edge of the valve plate: the point farthest from the central valve shaft is the first coordinate point P1(x1,y1), and the point closest to the central valve shaft is the fifth coordinate point P5(x5,y5); at the first coordinate point P... Between P1(x1,y1) and the fifth coordinate point P5(x5,y5), the second coordinate point P2(x2,y2), the third coordinate point P3(x3,y3), and the fourth coordinate point P4(x4,y4) are defined sequentially from the first coordinate point to the fifth coordinate point. Among them, the first coordinate point P1(x1,y1) and the fifth coordinate point P5(x5,y5) are fixed coordinate points, while the second coordinate point P2(x2,y2), the third coordinate point P3(x3,y3), and the fourth coordinate point P4(x4,y4) are variable coordinate points.

[0010] Step 2: The valve plate structure is designed as a smooth S-shaped curve. The range of variation for the variable coordinate points P2(x2,y2), P3(x3,y3), and P4(x4,y4) is set: the coordinates of each variable coordinate point along the x-axis... i,i＝2,3,4 >d / 2, where d is the valve plate thickness; various variable coordinate points P i,i＝2,3,4 In the x-direction, it is controlled within a circle with the point (d / 2, l / 4) as the center and l / 2 as the diameter, and there is no overlap in the y-direction, where l is the length of the valve plate;

[0011] Step 3: Based on the range of change of the variable coordinate points set in Step 2, use the Latin hypercube sampling method to generate N sets of valve plate geometric designs, and select M typical valve openings θ in actual engineering practice to form M×N combinations of "valve opening-valve plate geometric design". Use CFD numerical simulation on all M×N combinations of "valve opening-valve plate geometric design" and select a fixed inlet wind speed to calculate two performance indicators of non-eccentric wind valve: torque and relative energy loss.

[0012] Step 4: Calculate the 2D symbolic distance field images of M×N combinations of "valve opening degree-valve plate geometry design" and establish the corresponding SDF image dataset to accurately characterize the structural information of different valve plate geometry designs at different opening angles; establish a CFD dataset composed of torque and relative energy loss, and divide both the SDF image dataset and the CFD dataset according to the ratio of 8:2 for training set and validation set.

[0013] Step 5: Construct a learning and prediction model integrating two neural networks. The first neural network takes the coordinates of three variable coordinate points and the valve opening θ as input variables and a two-dimensional SDF image as the output variable, named SP-SDF. The SP-SDF model can predict the two-dimensional SDF image representation of any combination of "valve opening-valve plate geometry design". The second neural network takes the two-dimensional SDF image of any combination of "valve opening-valve plate geometry design" and the valve plate length l as input variables and two performance indicators of the valve: torque and relative energy loss as output variables, named SDF-CFD. The SDF-CFD model can predict the torque and relative energy loss of any combination of "valve opening-valve plate geometry design" corresponding to any SDF image.

[0014] Step 6: Train and validate the learning and prediction model of the two neural networks that are concatenated together, namely the SP-SDF model and the SDF-CFD model;

[0015] Step 7: Using the reliable learning prediction model obtained in Step 6, combined with the non-dominated sorting genetic algorithm, a set of Pareto optimal solutions is obtained with the objective of "maximum average torque" and "minimum average relative energy loss" for any valve plate geometry design under all selected valve openings. Then, based on CFD numerical simulation, the performance of the valve plate geometry design corresponding to all solutions in the Pareto optimal solution set is calculated. The optimal valve plate structure is selected by using the valve's equal percentage flow characteristic curve as a constraint.

[0016] The present invention has the following advantages and effects:

[0017] (1) The design method of the non-eccentric structure self-generated torque air volume measurement and control integrated valve proposed in this invention ensures the calculation accuracy comparable to CFD numerical simulation while improving the calculation efficiency by 300-500 times.

[0018] (2) Compared with the eccentric torque valve, the non-eccentric valve designed based on the present invention avoids the problem of the valve torque being too large, which makes it difficult to open or close, and the air volume prediction accuracy reaches 4% (the air volume measurement accuracy of the eccentric torque valve is about 10%).

[0019] (3) Compared with the non-eccentric torque valve, the torque of the non-eccentric valve designed based on the present invention is increased by 4-5 times, and it has energy-saving effect at small opening (≤50°).

[0020] (4) The non-eccentric valve designed based on the present invention has equal percentage flow characteristics, which is more in line with actual engineering needs. Attached Figure Description

[0021] Figure 1 This is a structural diagram of an existing non-eccentric damper;

[0022] Figure 2 This is a flowchart of the valve plate shape optimization process of the present invention;

[0023] Figure 3-1 yes Figure 1 The diagram shows a three-dimensional structure of the non-eccentric damper plate.

[0024] Figure 3-2 yes Figure 1 The diagram shows a two-dimensional cross-section of the valve plate of the non-eccentric damper.

[0025] Figure 3-3 yes Figure 1 A schematic diagram of parametric modeling of the upper half of the valve plate on the center valve shaft of a non-eccentric air valve.

[0026] Figure 4 This is a schematic diagram of a learning and prediction model that integrates two neural networks;

[0027] Figure 5 This is a schematic diagram of the valve plate shape of the non-eccentric structure self-generating torque type integrated air volume measurement and control valve of the present invention. Detailed Implementation

[0028] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0029] This invention addresses the practical issues of small torque in non-eccentric torque valves and unsuitability of eccentric torque valves for large air volume systems. It aims to increase torque by altering the valve plate geometry, without changing the valve shaft's location at the valve plate center. Based on the method of "valve plate structure parameterization—CFD numerical simulation—model prediction and optimization," a non-eccentric structure self-generating torque-type integrated air volume measurement and control valve with equal percentage flow characteristics has been developed.

[0030] The method of this invention is an improvement on the existing non-eccentric damper structure, such as... Figure 1 The existing non-eccentric valve structure shown is as follows: it consists of a valve body, a central valve shaft, and a valve plate. The valve body dimensions are L×W×H (unit: mm). The central valve shaft is a cylindrical rod with a length of W and a diameter of D (unit: mm), fixed at the geometric center of the valve, serving as a rotation shaft and torque transmission component. The valve plate is a flat plate with dimensions of l×w×d (l is the valve plate length, w is the valve plate width, and d is the valve plate thickness, all in mm), symmetrically distributed on both sides of the central valve shaft.

[0031] As shown in the attached figure, the design method of a non-eccentric structure self-generating torque type integrated air volume measurement and control valve of the present invention includes the following steps:

[0032] Step 1: Define the Cartesian coordinate system as follows: The endpoint on one side of the central valve axis of the non-eccentric valve is defined as the origin O, the horizontal direction is the x-axis, the vertical direction is the y-axis, and the direction parallel to the central valve axis is the z-axis. The valve opening θ is defined as the acute angle between the valve plate and the negative y-axis direction.

[0033] To ensure valve operational stability, the valve plate structure is as follows: its shape is fixed along the z-axis, and the valve plates on both sides of the central valve shaft are rotationally symmetrical around the central valve shaft. Therefore, its 3D structure (such as...) Figure 3-1 As shown, it can be effectively simplified to a 2D outline without losing its versatility (e.g., Figure 3-2 As shown), the valve plate width w and valve plate thickness d are both kept constant. Furthermore, only half of the two-dimensional profile can be considered to improve computational efficiency (e.g., Figure 3-3 (As shown). To mathematically represent the shape of the valve plate, a non-uniform rational B-spline (NURBS) curve generated by five coordinate points is used to characterize the shape of the valve plate located on one side of the central valve axis in the xOy plane. The five coordinate points are all located on the outer edge of the valve plate: the point farthest from the central valve axis is designated as the first coordinate point P1(x1,y1), and the point closest to the central valve axis is designated as the fifth coordinate point P5(x5,y5). Between the first coordinate point P1(x1,y1) and the fifth coordinate point P5(x5,y5), the second coordinate point P2(x2,y2), the third coordinate point P3(x3,y3), and the fourth coordinate point P4(x4,y4) are defined sequentially from the first coordinate point to the fifth coordinate point. The first coordinate point P1(x1,y1) and the fifth coordinate point P5(x5,y5) are fixed coordinate points, while the second coordinate point P2(x2,y2), the third coordinate point P3(x3,y3), and the fourth coordinate point P4(x4,y4) are variable coordinate points.

[0034] Step 2: Based on Bernoulli's principle and the valve torque generation mechanism, the valve plate structure is designed with a smooth S-shaped curve, which is considered a contour that can fully promote the transformation from aerodynamics to solid mechanics to enhance valve torque. Figure 5 The upper half of the central valve shaft shown is designed to be convex, and the lower half is designed to be concave (assuming the airflow direction in the diagram is from right to left). Based on this, the range of variation for the variable coordinate points P2(x2,y2), P3(x3,y3), and P4(x4,y4) is set: considering that the coordinate points are located at the outer edge of the valve plate, the coordinates of each variable coordinate point along the x-axis are required to be... i,i＝2,3,4 >d / 2, where d is the valve plate thickness; due to the limitations of the valve's physical structure, each variable coordinate point P i,i＝2,3,4 In the x-direction, the valve plate is controlled to be within a circle with the point (d / 2, l / 4) as the center and l / 2 as the diameter, and there is no overlap in the y-direction, where l is the length of the valve plate. Preferably, the distance between two adjacent variable coordinate points in the y-direction is kept at more than 10 mm, which can be selected according to the actual valve plate size to ensure that the valve plate geometry is controllable.

[0035] Step 3: Based on the range of variable coordinate points set in Step 2, use the Latin Hypercube Sampling (LHS) method to generate N sets of valve plate geometry designs. Select M typical valve openings θ from actual engineering practice to form M×N combinations of "valve opening-valve plate geometry design". For all M×N combinations of "valve opening-valve plate geometry design", use CFD numerical simulation (CFD is short for Computational Fluid Dynamics) to calculate two performance indicators of the non-eccentric valve using a fixed inlet velocity: torque and relative energy loss (relative energy loss refers to the ratio of the energy consumed by the fluid after passing through the valve (static pressure drop)) to the total fluid pressure (static pressure + dynamic pressure) before passing through the valve).

[0036] Step 4: Calculate the 2D Signed Distance Field (SDF) images of M×N combinations of "valve opening degree-valve plate geometry design" and establish the corresponding SDF image dataset to accurately characterize the structural information of different valve plate geometries at different opening angles. Establish a CFD dataset consisting of torque and relative energy loss, where the torque and relative energy loss are obtained based on CFD numerical simulations and correspond one-to-one with the M×N combinations of "valve opening degree-valve plate geometry design". Divide both the SDF image dataset and the CFD dataset into a training set and a validation set in an 8:2 ratio, i.e., 80% of the data is used for training and 20% for validation. The training and validation sets are used to train and validate the subsequent neural network learning prediction model.

[0037] The symbolic distance field (SDF) used in this step is a fundamental concept in computer graphics, capable of accurately representing implicit 3D structures with different configurations. It is easily integrated into neural networks and has been widely applied to tasks such as flow field reconstruction. The calculation formula is shown below:

[0038] Z = {(i,j,k)∈R} 3 :f(i,j,k)=0} (1)

[0039]

[0040] Where f represents the level set function, f(i,j,k)=0 represents that point (i,j,k) is located on the geometric boundary of the valve plate, f(i,j,k)<0 represents that point (i,j,k) is located inside the valve plate geometry, and f(i,j,k)>0 represents that point (i,j,k) is located outside the valve plate geometry. D(i,j,k) represents the directional distance function used to measure the distance between any point (i,j,k) and the boundary of the closed geometric shape Z, and signf(i,j,k) is determined by whether (i,j,k) is inside or outside the shape. R 3 Let f(i,j,k) represent a real three-dimensional space, where (i,j,k) represent continuous real coordinates (where i represents the x-axis coordinate, j represents the y-axis coordinate, and k represents the z-axis coordinate), and Z represents the set of points for which the function f(i,j,k) = 0. The two-dimensional SDF image representation is chosen primarily because the fluid flow within the duct is symmetrical along the z-axis; this simplification significantly reduces the complexity of subsequent neural network computations.

[0041] Step 5: Construct a learning and prediction model integrating two neural networks, such as... Figure 4 As shown: The first neural network takes the coordinates of three variable coordinate points and the valve opening θ as input variables, and a two-dimensional SDF image as the output variable, named SP-SDF. The SP-SDF model can predict the two-dimensional SDF image representation of any combination of "valve opening-valve plate geometry design". The second neural network takes the two-dimensional SDF image of any combination of "valve opening-valve plate geometry design" and the valve plate length l as input variables, and two valve performance indicators: torque and relative energy loss as output variables, named SDF-CFD. The SDF-CFD model can predict the torque and relative energy loss of any combination of "valve opening-valve plate geometry design" corresponding to any SDF image. The learning and prediction model integrating the two neural networks can achieve accurate prediction of the torque and relative energy loss of the valve corresponding to changes in the coordinates of the three variable coordinate points and the valve opening.

[0042] Step Six: Training and Validating the Learning and Prediction Model of the Two Neural Networks Connected Together (SP-SDF Model and SDF-CFD Model). This step consists of two parts: The SP-SDF model is trained and validated using the training and validation sets from Step Four. The input consists of the coordinates of three variable coordinate points corresponding to the M×N "valve opening-valve plate geometry design" combinations generated in Step Three, and the valve opening θ. The output is the two-dimensional SDF image from the SDF image dataset obtained in Step Four.

[0043] The SDF-CFD model is trained and validated using the training and validation sets obtained in step four. The input consists of the 2D SDF image and valve length l from the SDF image dataset obtained in step four, and the output consists of the valve torque and relative energy consumption from the CFD dataset obtained in step four.

[0044] During training and validation, monitor the changes in the model's loss function until the loss function converges, at which point the model is considered reliable.

[0045] Step 7: Using the reliable learning prediction model obtained in Step 6, combined with the Non-Dominated Sorting Genetic Algorithm (NSGA-II), a set of Pareto optimal solutions is obtained with the objectives of "maximum average torque" and "minimum average relative energy loss" for any valve plate geometry design under all selected valve openings. Then, based on CFD numerical simulation, the performance of the valve plate geometry designs corresponding to all solutions in the Pareto optimal solution set is calculated. Using the valve's equal percentage flow characteristic curve as a constraint, the optimal valve plate structure is selected.

[0046] Example 1

[0047] Step 1: Assume the valve plate length l is 140mm, width w is 140mm, and thickness d is 2mm. Generate a non-uniform rational B-spline (NURBS) curve from 5 coordinate points (P1-P5) to define the valve plate shape, such as... Figure 1 As shown in the figure. Among them, P1 and P5 are fixed coordinate points with coordinates P1(1,70) and P5(1,0) respectively; P2, P3, and P4 are variable coordinate points.

[0048] Step 2: Define the range of variation for the three variable coordinate points. The specific principle is: considering that the coordinate points are located on the outer edge of the valve plate, therefore, x... i,i＝2,3,4 >1; Due to the limitations of the valve's physical structure, P i,i＝2,3,4 The variable coordinates are controlled in the x-direction within a circle centered at point (1, 17.5) with a diameter of l / 2 = 35 mm, and there is no overlap in the y-direction; the distance between two adjacent variable coordinate points in the y-direction is not less than 10 mm. Therefore, the specific range of variation for the three variable coordinate points is defined as follows: x2∈(1, 34), y2∈(46.67, 70); x3∈(1, 36), y3∈(23.33, 46.67); x4∈(1, 34), y4∈(0, 23.33).

[0049] Step 3: Based on the variation range of the variable coordinate points set in Step 2, the Latin Hypercube Sampling Method (LHS) is used to generate N=100 sets of valve plate geometric designs. Five typical valve openings (θ=30°, 40°, 50°, 60°, 70°) from actual engineering practice are selected to form 5×100 combinations of "valve opening-valve plate geometric design". CFD numerical simulation is used for all combinations, with the inlet wind speed set to a typical wind speed of 7 m / s in a ventilation and air conditioning system, to calculate the valve torque and relative energy loss.

[0050] Step 4: Calculate 5×100 two-dimensional SDF images of the "valve opening-valve plate geometry design" combinations and establish a corresponding SDF image dataset to accurately characterize the structural information of different valve plate geometries at different opening angles; establish corresponding CFD datasets for the torque and relative energy loss of the 5×100 "valve opening-valve plate geometry design" combinations obtained based on CFD numerical simulation. Divide the SDF dataset and CFD dataset into training and validation sets, respectively, with 80% used for training and 20% for validation. The training and validation sets are used to train and validate the subsequent neural network deep learning prediction model.

[0051] Step 5: Construct a learning and prediction model integrating two neural networks: the SP-SDF neural network and the SDF-CFD neural network. The SP-SDF neural network includes an encoder, intermediate layers, and a decoder. The input data dimension is (7×64×64), representing the coordinates of three variable points (x2, y2, x3, y3, x4, y4) defining the two-dimensional contour of the valve plate and the valve opening θ. The encoder consists of Conv2D, MaxPool2D, and ReLU layers, encoding four times to map the structural parameters to a high-dimensional feature space (128×4×4); the intermediate layer uses a fully connected layer to extract information while maintaining the array dimension; the decoder consists of...

[0052] The process consists of ConvTranspose2D, Upsample, and ReLU layers, and is decoded four times to obtain an array of dimensions (1×64×64) representing a two-dimensional SDF image.

[0053] The SDF-CFD neural network employs an encoder-based global aggregation architecture. In addition to the 2D SDF image, the input data, after normalizing the valve plate length l = 140mm, is copied into a matrix matching the size of the SDF image and concatenated along the channel dimension to form a dual-channel input of size (2×64×64). The encoder is the same as that of SP-SDF, encoding four times, and outputting an array of size (128×4×4). Next, global average pooling (GAP) is used to compress the spatial dimension, and then two fully connected layers directly output 2D scalars, namely valve torque and relative energy loss. The two neural network models are cascaded to form a complete learning and prediction model. This achieves accurate prediction of torque and relative energy loss for any valve plate geometry design at any opening degree.

[0054] Step Six: Using the two datasets obtained in Step Four, each containing 500 data points (SDF image dataset and CFD dataset), train and validate the two neural network models from Step Five, respectively. During training and validation, the model loss function converged to below 5%, indicating model reliability.

[0055] Step 7: Using the reliable learning prediction model obtained in Step 6, combined with the Non-Dominated Sorting Genetic Algorithm (NSGA-II), and taking the "maximum average torque" and "minimum average relative energy loss" of any valve plate geometry design under all selected valve openings as objectives, a Pareto optimal solution set containing 22 solutions is obtained. Then, based on CFD numerical simulation, the performance of the valve plate geometry designs corresponding to the 22 solutions is calculated. Using the valve's equal percentage flow characteristic curve as a constraint, the optimal design scheme is selected, such as... Figure 5 As shown. Calculations show that the airflow prediction accuracy of the non-eccentric structure self-generated torque integrated airflow measurement and control valve obtained in this embodiment reaches 5%, and the torque is increased by 4-5 times, while also exhibiting energy-saving effects at small openings (≤50°).

Claims

1. A design method for a non-eccentric structure self-generating torque type integrated air volume measurement and control valve, characterized in that... Includes the following steps: Step 1: Define the Cartesian coordinate system as follows: The endpoint of one side of the central valve axis of the non-eccentric valve is defined as the origin O, the horizontal direction is the x-axis, the vertical direction is the y-axis, the direction parallel to the central valve axis is the z-axis, and the valve opening θ is defined as the acute angle between the valve plate and the negative y-axis direction. The valve plate structure is as follows: the shape is fixed along the z-axis, and the valve plates on both sides of the central valve shaft are rotationally symmetrical around the central valve shaft. The valve plate width w and valve plate thickness d are both kept constant. A non-uniform rational B-spline curve generated by five coordinate points is used to characterize the shape of the valve plate located on one side of the central valve shaft in the xOy plane. The five coordinate points are all located on the outer edge of the valve plate: the point farthest from the central valve shaft is the first coordinate point P1(x1,y1), and the point closest to the central valve shaft is the fifth coordinate point P5(x5,y5); at the first coordinate point P... Between P1(x1,y1) and the fifth coordinate point P5(x5,y5), the second coordinate point P2(x2,y2), the third coordinate point P3(x3,y3), and the fourth coordinate point P4(x4,y4) are defined sequentially from the first coordinate point to the fifth coordinate point. Among them, the first coordinate point P1(x1,y1) and the fifth coordinate point P5(x5,y5) are fixed coordinate points, while the second coordinate point P2(x2,y2), the third coordinate point P3(x3,y3), and the fourth coordinate point P4(x4,y4) are variable coordinate points. Step 2: The valve plate structure is designed as a smooth S-shaped curve. The range of variation for the variable coordinate points P2(x2,y2), P3(x3,y3), and P4(x4,y4) is set: the coordinates of each variable coordinate point along the x-axis... i , i＝2,3,4 >d / 2, where d is the valve plate thickness; various variable coordinate points P i,i＝2,3,4 In the x-direction, it is controlled within a circle with the point (d / 2, l / 4) as the center and l / 2 as the diameter, and there is no overlap in the y-direction, where l is the length of the valve plate; Step 3: Based on the range of change of the variable coordinate points set in Step 2, generate N sets of valve plate geometric designs using the Latin hypercube sampling method, and select M typical valve openings θ in actual engineering practice to form M×N combinations of "valve opening-valve plate geometric design". Use CFD numerical simulation on all M×N combinations of "valve opening-valve plate geometric design" and select a fixed inlet wind speed to calculate two performance indicators of the non-eccentric wind valve: torque and relative energy loss. Step 4: Calculate the 2D symbolic distance field images of M×N combinations of "valve opening degree-valve plate geometry design" and establish the corresponding SDF image dataset to accurately characterize the structural information of different valve plate geometry designs at different opening angles; establish a CFD dataset composed of torque and relative energy loss, and divide both the SDF image dataset and the CFD dataset according to the ratio of 8:2 between the training set and the validation set. Step 5: Construct a learning and prediction model integrating two neural networks. The first neural network takes the coordinates of three variable coordinate points and the valve opening θ as input variables and a two-dimensional SDF image as the output variable, named SP-SDF. The SP-SDF model can predict the two-dimensional SDF image representation of any combination of "valve opening-valve plate geometry design". The second neural network takes the two-dimensional SDF image of any combination of "valve opening-valve plate geometry design" and the valve plate length l as input variables and two performance indicators of the valve: torque and relative energy loss as output variables, named SDF-CFD. The SDF-CFD model can predict the torque and relative energy loss of any combination of "valve opening-valve plate geometry design" corresponding to any SDF image. Step 6: Train and validate the learning and prediction model of the two neural networks that are concatenated together, namely the SP-SDF model and the SDF-CFD model; Step 7: Using the reliable learning prediction model obtained in Step 6, combined with the non-dominated sorting genetic algorithm, a set of Pareto optimal solutions is obtained with the objective of "maximum average torque" and "minimum average relative energy loss" for any valve plate geometry design under all selected valve openings. Then, based on CFD numerical simulation, the performance of the valve plate geometry design corresponding to all solutions in the Pareto optimal solution set is calculated. The optimal valve plate structure is selected by using the valve's equal percentage flow characteristic curve as a constraint.

2. The design method of the non-eccentric structure self-generating torque type integrated air volume measurement and control valve according to claim 1, characterized in that... The distance between two adjacent variable coordinate points in the y direction should be kept above 10mm.

3. The design method of the non-eccentric structure self-generating torque type integrated air volume measurement and control valve according to claim 1, characterized in that... Step six is ​​divided into two steps: the SP-SDF model is trained and validated using the training set and validation set from step four. The coordinate values ​​of the three variable coordinate points corresponding to the M×N "valve opening degree-valve plate geometric design" combinations generated in step three and the valve opening degree θ are used as inputs, and the two-dimensional SDF image in the SDF image dataset obtained in step four is used as output. The SDF-CFD model is trained and validated using the training and validation sets from step four: the two-dimensional SDF image and valve plate length l from the SDF image dataset obtained in step four are used as inputs, and the valve torque and relative energy consumption from the CFD dataset obtained in step four are used as outputs.

Images