Design method of spinning drafting air duct structure

Abstract

The invention relates to a design method of a spinning drafting air duct structure, which comprises the following steps of: 1) inputting structure parameters and fluid parameters, constructing a parameterized spinning drafting air duct fluid dynamic simulation model, and generating flow field state original data; 2) preprocessing the original data, and constructing a training data set of simulation model input parameters and output results; 3) establishing a mapping relation between the input parameters and the output results by adopting a support vector regression model; 4) establishing a multi-objective optimization model based on a genetic algorithm, and generating a filial generation population; 5) selecting a Pareto frontier solution of a new generation of population to perform fluid dynamic verification, adding a high-error sample into the training data set, and retraining the support vector regression model; according to the multi-objective optimization method for the air duct structure, multi-objective optimization of the air duct structure is achieved, and airflow uniformity is improved.

Description

Technical Field

[0001] This invention relates to a method for designing air ducts, specifically a method for designing a spinning and drawing air duct structure, belonging to the field of digital design technology. Background Technology

[0002] The uniformity of airflow in spunbond drafting ducts directly affects the quality of fiber forming. Traditional duct design relies on trial and error, resulting in uneven airflow distribution and significant pressure loss. Existing technologies lack systematic optimization methods, making it difficult to balance fluid performance and structural constraints. Furthermore, the nonlinear coupling effect between duct geometry and fluid response is significant, necessitating the use of intelligent algorithms for efficient optimization.

[0003] Therefore, in order to solve the above problems, it is indeed necessary to provide an innovative design method for the spinning drawing air duct structure to overcome the defects in the prior art. Summary of the Invention

[0004] The purpose of this invention is to provide a design method for spinning and drawing air duct structures, which solves the problems of uneven airflow distribution and high pressure loss in traditional air duct design through the synergistic optimization of fluid dynamics simulation model (CFD) and genetic algorithm.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a design method for a spinning drawing air duct structure, which includes the following process steps: 1) Input structural and fluid parameters to construct a parameterized spinning and drawing duct fluid dynamics simulation model and generate raw flow field state data; 2) Preprocess the raw data from step 1) to construct a training dataset for the input parameters and output results of the simulation model; 3) Use a support vector regression model to establish a mapping relationship between the input parameters and output results obtained in step 2); 4) Establish a multi-objective optimization model based on genetic algorithms to generate offspring populations; 5) Select the Pareto front solution of the new generation population for hydrodynamic verification, and add high error samples to the training dataset to retrain the support vector regression model in step 3); 6) Repeat steps 4-5) until the multi-objective optimization model reaches the convergence condition; 7) Output the optimal duct structure design scheme, which includes optimized geometric parameters, fluid parameters, performance indicators and results that meet engineering constraints.

[0006] The design method of the spinning drawing air duct structure of the present invention is further described as follows: In step 1), the input structural parameters include the cross-sectional shape of the air duct, the number of guide plates, the planar position of the guide plates, the inlet / outlet diameter ratio, and the radius of curvature of the bending section. Fluid parameters include inlet wind speed, turbulence intensity, fluid density, and viscosity; The simulation model is constructed as a three-dimensional steady-state turbulence model. The turbulence model selected is the Realizable k-ε model. The mesh generation is achieved by ICEM CFD, with a mesh size of >500,000. The boundary conditions are the minimum spacing of the guide vanes, the total volume of the air duct, and the inlet / outlet ratio. The raw flow field data includes the standard deviation of airflow uniformity σ, the pressure loss coefficient ΔP, and the energy consumption E; the formula for the standard deviation of airflow uniformity is: In the formula: N is the total number of monitoring points on the duct cross-section, and Ti is the wind speed at the i-th monitoring point. Average wind speed; E = ΔP × Q, where Q is the volumetric flow rate.

[0007] The design method of the spinning and drawing air duct structure of the present invention is further as follows: In step 2), the preprocessing includes data cleaning, normalization and feature selection; firstly, the 3σ criterion is used to identify and remove abnormal samples caused by backflow or numerical instability; then, the Min-Max normalization method is used to map input parameters of different dimensions to the [0,1] interval; finally, based on the Pearson correlation coefficient |r|>0.7, feature variables strongly related to airflow uniformity, pressure loss and energy consumption are screened, and redundant parameters are removed.

[0008] The design method of the spinning and drawing air duct structure of the present invention is further described as follows: In step 2), the input of the simulation model includes geometric and fluid parameters, and the output includes wind speed distribution, pressure loss and airflow uniformity, which together constitute the training dataset.

[0009] The design method of the spinning and drawing air duct structure of the present invention further comprises: in step 3), the kernel function of support vector regression is selected as the radial basis function, the expression of which is: Where, input vector a m a n It consists of duct geometry and fluid parameters, specifically including the guide vane position coordinates, inlet / outlet diameter ratio, radius of curvature of the bend section, inlet wind speed, and turbulence intensity; core bandwidth. Used to adjust the model's sensitivity to local airflow disturbances caused by the deflector.

[0010] The design method of the spinning drawing air duct structure of the present invention is further as follows: Step 4) specifically involves: by setting the design variable range of the number, position, radius of curvature and inlet / outlet diameter ratio of the guide plates, and combining the initialization population, fitness function and genetic operation, the construction of the optimization model is realized; Among them, Latin hypercube sampling was used to generate the initial population to ensure uniform coverage of the parameter space; The fitness function is defined as a weighted composite objective of the standard deviation of airflow uniformity σ, pressure loss ΔP, and energy consumption E, and its specific calculation formula is as follows: In the formula: α, β, γ are weighting coefficients; Genetic operations include tournament selection, uniform crossover, and Gaussian mutation, which generate offspring populations through continuous iteration.

[0011] The design method of the spinning drawing air duct structure of the present invention is further as follows: Step 5) specifically involves: performing hydrodynamic verification on the Pareto front solution of each generation; if the relative error between the predicted value and the actual value of the support vector regression model is >10%, then the sample is marked as a high error sample and added to the training set; after every 5 iterations, the support vector regression model is updated through an incremental learning strategy.

[0012] The design method of the spinning drawing air duct structure of the present invention is further as follows: the incremental learning strategy is as follows: without retraining all samples, the newly added high error samples are incorporated into the training set; the hyperparameters C and γ of the support vector regression model are optimized only for the newly added samples, the Bayesian optimization method is adopted, and the adjustment range is limited to ±10%.

[0013] The design method of the spinning drawing air duct structure of the present invention is further as follows: the convergence condition of step 6) is specifically: the fitness function change rate is <1% for 3 consecutive generations, or the optimization is terminated when the total number of iterations reaches 50 generations.

[0014] The design method of the spinning and drawing air duct structure of the present invention further includes: the engineering constraints of step 7) include: the minimum spacing between the guide plates ≥ 50 mm; Total duct volume ≤ 0.5 m³ 3 ; The import / export ratio is ≤3:1.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The design method of the spinning drawing air duct structure of the present invention achieves multi-objective optimization of the air duct structure through the collaborative optimization of fluid dynamics simulation model and genetic algorithm, solves the problems of uneven airflow distribution and high pressure loss in traditional air duct design, and can be widely applied to the design of spinning drawing air ducts, significantly improving the flow field energy efficiency.

[0016] 2. The design method of the spinning drawing air duct structure of the present invention adopts a support vector regression model, which can map the input parameters to a high-dimensional feature space to enhance the model's ability to fit complex nonlinear relationships and improve prediction accuracy.

[0017] 3. The design method of the spinning drafting air duct structure of the present invention is not only applicable to the design of drafting air ducts in nonwoven equipment, but can also be extended to multi-objective optimization scenarios of other fluid transport systems. Attached Figure Description

[0018] Figure 1 This is a flowchart of the design method for the spinning and drawing air duct structure of the present invention.

[0019] Figure 2 This is a schematic diagram of the spinning and drawing air duct structure of the present invention.

[0020] Figure 3 This is a flowchart of steps 3) to 6) of the present invention for optimizing the model. Detailed Implementation

[0021] Please refer to the instruction manual appendix. Figure 1 As shown, this invention is a design method for a spinning drawing air duct structure, which includes the following process steps: 1) Input structural and fluid parameters to construct a parametric spinning drawing duct fluid dynamics (CFD) simulation model and generate raw data of flow field state for different drawing duct structural parameters.

[0022] The input structural parameters include the cross-sectional shape of the duct, the number of guide vanes, the planar position of the guide vanes, the inlet / outlet diameter ratio, and the radius of curvature of the curved section. Based on the ANSYS Fluent simulation platform, these geometric parameters are bound as variables to the modeling script, directly driving the automatic generation of the duct shape and guide vane arrangement.

[0023] The fluid parameters include inlet wind speed, turbulence intensity, fluid density, and viscosity, which are used to set boundary conditions to ensure that the simulation results are consistent with actual engineering conditions.

[0024] The constructed simulation model is a three-dimensional steady-state turbulence model. The turbulence model selected is the Realizable k-ε model, which has higher accuracy in rotating and strong shear flow fields. Mesh generation was achieved using ICEM CFD, with a mesh size greater than 500,000 to ensure computational convergence and accuracy. Boundary conditions include minimum guide vane spacing, total duct volume, and inlet / outlet ratio.

[0025] The original flow field data, i.e., the simulation output, includes the standard deviation of airflow uniformity σ, the pressure loss coefficient ΔP, and the energy consumption E; wherein, the formula for the standard deviation of airflow uniformity is: In the formula: N is the total number of monitoring points on the duct cross-section, and Ti is the wind speed at the i-th monitoring point. Average wind speed; E = ΔP × Q, where Q is the volumetric flow rate.

[0026] 2) Preprocess the raw data from step 1) to construct a training dataset of input parameters and output results for the simulation model.

[0027] The preprocessing includes data cleaning, normalization, and feature selection. First, the 3σ criterion is used to identify and remove abnormal samples caused by backflow or numerical instability (such as pressure mutation points caused by backflow). Then, the Min-Max normalization method is used to map input parameters of different dimensions to the [0,1] interval. Finally, based on the Pearson correlation coefficient |r|>0.7, feature variables that are strongly correlated with airflow uniformity, pressure loss, and energy consumption are screened, and redundant parameters (such as the influence of wall roughness on low Reynolds number conditions) are removed.

[0028] The inputs to the above simulation model include geometric and fluid parameters, and the outputs are wind speed distribution, pressure loss, and airflow uniformity. Together, they constitute the training dataset, which is used for the establishment and optimization of subsequent surrogate models.

[0029] Please continue to refer to the instruction manual appendix. Figure 3 As shown in Figure 3), a support vector regression (SVR) model is used to establish a mapping relationship between the input parameters and output results obtained in step 2).

[0030] Specifically, the SVR model takes geometric and fluid parameters as inputs and outputs performance indicators such as airflow uniformity, pressure loss, and energy consumption, thereby achieving a mapping between the two.

[0031] The kernel function used in the support vector regression model is the radial basis function (RBF). Its purpose is to map the input parameters to a high-dimensional feature space, thereby enhancing the model's ability to fit complex nonlinear relationships and improving prediction accuracy. Its expression is: The input vectors am and an are composed of duct geometric parameters and fluid parameters, specifically including the guide vane position coordinates, inlet / outlet diameter ratio, radius of curvature of the bend, inlet wind speed, and turbulence intensity; kernel bandwidth. Used to adjust the model's sensitivity to local airflow disturbances caused by the deflector.

[0032] The hyperparameters of the SVR model (penalty coefficient C, kernel bandwidth) The optimization is achieved through grid search combined with 5-fold cross-validation, with a search range of C∈[1,100]. ∈[0.01,1], the final optimal value is C=50. =0.1; The validation accuracy of the model was evaluated by mean squared error (MSE<0.05) and coefficient of determination (R2>0.92).

[0033] 4) Establish a multi-objective optimization model based on a genetic algorithm and generate a progeny population. The specific process is as follows: By setting the design variable ranges for the number, position, radius of curvature, and inlet / outlet diameter ratio of the baffles, and combining the initialization population, fitness function, and genetic operations, the optimization model is constructed. This step uses the surrogate model and training data obtained in steps 1) to 3) as support to explore the design variable space and achieve multi-objective optimization of the duct structure.

[0034] Among them, Latin hypercube sampling is used to generate the initial population to ensure uniform coverage of the parameter space.

[0035] The fitness function is defined as a weighted composite objective of the standard deviation of airflow uniformity σ, pressure loss ΔP, and energy consumption E, and its specific calculation formula is as follows: In the formula: α, β, and γ are weighting coefficients used to balance the importance of different optimization objectives.

[0036] The genetic operations include tournament selection, uniform crossover, and Gaussian mutation, which iteratively generate the offspring population. A tournament strategy is used to select the top 20% of individuals by fitness as parents. Uniform crossover (probability 0.8) randomly swaps the deflector position codes of parent individuals. Gaussian mutation (probability 0.05) applies a ±10% perturbation to the radius of curvature.

[0037] 5) Select the Pareto front solution of the new generation population for hydrodynamic verification, add high error samples to the training dataset, and retrain the support vector regression model of step 3).

[0038] Specifically, hydrodynamic verification is performed on the Pareto front solution of each generation. If the relative error between the predicted value and the actual value of the support vector regression model is greater than 10%, the sample is marked as a high error sample and added to the training set. After every 5 iterations, the support vector regression model is updated through an incremental learning strategy.

[0039] The incremental learning strategy is as follows: without retraining all samples, newly added high-error samples are incorporated into the training set; the hyperparameters C and γ of the support vector regression model are optimized only for the newly added samples, using the Bayesian optimization method, and the adjustment range is limited to ±10% to improve local prediction accuracy while maintaining the overall stability of the model.

[0040] 6) Repeat steps 4-5) until the multi-objective optimization model reaches the convergence condition. The convergence condition is specifically: the fitness function change rate is <1% for three consecutive generations, or the optimization terminates when the total number of iterations reaches 50 generations.

[0041] 7) Output the optimal duct structure design scheme, as shown in the attached diagram. Figure 2 As shown, the proposed solution includes optimized geometric parameters, fluid parameters, performance indicators, and results that meet engineering constraints. The engineering constraints include: minimum spacing between guide vanes ≥ 50 mm; Total duct volume ≤ 0.5 m³ 3 ; The import / export ratio is ≤3:1.

[0042] The specific implementation process of the present invention will be described in detail below with reference to specific embodiments.

[0043] The geometric parameters of the duct model are as follows: duct cross-sectional shape (converging inlet diameter 200 mm, outlet diameter 100 mm), number of guide vanes (3), their planar coordinates are (x1, y1) = (0.2L, 0.5H), (x2, y2) = (0.5L, 0.3H), (x3, y3) = (0.8L, 0.7H), where L = 2 m is the duct length, H = 200 mm is the cross-sectional height; the radius of curvature of the curved section is R = 1.5D (D = 200 mm). Fluid parameters are set as follows: inlet wind speed 20 m / s, turbulence intensity 10%, fluid medium is air (density ρ = 1.225 kg / m³). 3 The viscosity is μ = 1.8 × 10⁻⁵ Pa⋅s. A k-ε turbulence model is used. A high-precision mesh is generated. Boundary conditions are set as mass flow rate inlet (velocity 20 m / s) and static pressure outlet (0 Pa gauge pressure). The simulation outputs wind speed distribution data at 100 monitoring points in the duct cross-section, and the standard deviation of airflow uniformity σ is calculated. In the formula, N=100 is the total number of monitoring points, and T i Let i be the wind speed at point i. The average wind speed is 18.5 m / s. The total pressure drop is ΔP = 420 Pa, and the energy consumption is E = ΔP × Q, where the volumetric flow rate is Q = 0.0314 m³ / s. 3 / s, corresponding to a wind turbine power of 4.8 kW.

[0044] Data cleaning, normalization, and feature selection were performed on the simulation data. Data cleaning employed the 3σ criterion to remove outlier samples (such as monitoring points in the recirculation zone) with a wind speed standard deviation > 5 m / s. Normalization was performed using the Min-Max method to map parameters such as the guide vane coordinates (x, y) and radius of curvature R to the 0, 1 interval. In the formula, X min and X maxThese represent the minimum and maximum values ​​of the parameters, respectively. Feature selection is based on the Pearson correlation coefficient r, filtering and optimizing parameters that are strongly correlated with the objectives (σ, ΔP, E) (such as inlet wind speed r=0.85, number of guide vanes r=0.78), and eliminating parameters that are weakly correlated (such as wall roughness |r|<0.3).

[0045] The SVR model is constructed using the RBF kernel function, and the kernel function expression is as follows: Hyperparameters (penalty coefficient C, kernel bandwidth) Through grid search optimization, with a search range C∈{1,10,50,100} and γ∈{0.01,0.1,1}, a set of hyperparameters was obtained. The model training set and test set were split in an 8:2 ratio. The 5-fold cross-validation results showed a mean squared error (MSE) of 0.042, a coefficient of determination (R²) of 0.93, and a prediction relative error of <8%.

[0046] The initial population was generated using Latin hypercube sampling, resulting in 100 individuals. Parameters included the number of baffles n ∈ [1, 5], radius of curvature R ∈ [100 mm, 400 mm], and inlet / outlet diameter ratio ∈ [1:1, 1:3]. The fitness function was defined as: .

[0047] Genetic manipulation includes: Selection: The tournament strategy selects the top 20% of individuals in terms of fitness; Crossover: Uniform crossover (probability 0.8), offspring parameters are generated by a linear combination of parent parameters. Mutation: Gaussian mutation (probability 0.05), which applies a perturbation to the parameters. .

[0048] Pareto front solutions are extracted through a multi-objective optimization strategy, and non-dominated sorting and crowding calculation ensure the diversity and convergence of the solution set.

[0049] CFD validation is performed on the Pareto front solution for each generation. If the SVR prediction error is >10%, the sample is added to the training set. The SVR model is updated every 5 generations through incremental learning, with a hyperparameter adjustment step size of ±5%. The convergence condition is that the rate of change of the fitness function is <1% for 3 consecutive generations, and the iteration continues until the convergence condition is met (F decreases from 1.82 to 1.80).

[0050] The optimal solution selects 4 baffles with position coordinates (0.3L, 0.6H), (0.6L, 0.4H), (0.7L, 0.8H), and (0.9L, 0.2H), with an inlet / outlet diameter ratio of 1:2 and a radius of curvature R = 300 mm.

[0051] Engineering constraint verification: Minimum spacing of guide vanes 60 mm > 50 mm, duct volume 0.48 m³ 3 <0.5 m 3 The import / export ratio is 1:2 < 3:1.

[0052] The above-described specific embodiments are merely preferred embodiments of this invention and are not intended to limit this invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the protection scope of this invention.

Claims

1. A design method for a spinning drawing air duct structure, characterized in that: The process includes the following steps: 1) Input structural and fluid parameters to construct a parameterized spinning and drawing duct fluid dynamics simulation model and generate raw flow field state data; 2) Preprocess the raw data from step 1) to construct a training dataset for the input parameters and output results of the simulation model; 3) Use a support vector regression model to establish a mapping relationship between the input parameters and output results obtained in step 2); 4) Establish a multi-objective optimization model based on genetic algorithms to generate offspring populations; 5) Select the Pareto front solution of the new generation population for hydrodynamic verification, and add high error samples to the training dataset to retrain the support vector regression model in step 3); 6) Repeat steps 4-5) until the multi-objective optimization model reaches the convergence condition; 7) Output the optimal duct structure design scheme, which includes optimized geometric parameters, fluid parameters, performance indicators and results that meet engineering constraints.

2. The design method of the spinning drawing air duct structure as described in claim 1, characterized in that: In step 1), the input structural parameters include the cross-sectional shape of the air duct, the number of guide vanes, the planar position of the guide vanes, the inlet / outlet diameter ratio, and the radius of curvature of the curved section. Fluid parameters include inlet wind speed, turbulence intensity, fluid density, and viscosity; The simulation model is constructed as a three-dimensional steady-state turbulence model. The turbulence model selected is the Realizable k-ε model. The mesh generation is achieved by ICEM CFD, with a mesh size of >500,000. The boundary conditions are the minimum spacing of the guide vanes, the total volume of the air duct, and the inlet / outlet ratio. The raw flow field data includes the standard deviation of airflow uniformity σ, the pressure loss coefficient ΔP, and the energy consumption E; the formula for the standard deviation of airflow uniformity is: In the formula: N is the total number of monitoring points on the duct cross-section, and Ti is the wind speed at the i-th monitoring point. Average wind speed; E = ΔP × Q, where Q is the volumetric flow rate.

3. The design method of the spinning drawing air duct structure as described in claim 1, characterized in that: In step 2), the preprocessing includes data cleaning, normalization, and feature selection. First, the 3σ criterion is used to identify and remove abnormal samples caused by backflow or numerical instability. Then, the Min-Max normalization method is used to map input parameters of different dimensions to the [0,1] interval. Finally, based on the Pearson correlation coefficient |r|>0.7, feature variables that are strongly correlated with airflow uniformity, pressure loss, and energy consumption are screened, and redundant parameters are removed.

4. The design method of the spinning drawing air duct structure as described in claim 1, characterized in that: In step 2), the inputs to the simulation model include geometric and fluid parameters, and the outputs are wind speed distribution, pressure loss, and airflow uniformity, which together constitute the training dataset.

5. The design method of the spinning drawing air duct structure as described in claim 1, characterized in that: In step 3), the kernel function for support vector regression is the radial basis function, whose expression is: Where, input vector a m a n It consists of duct geometry and fluid parameters, specifically including the guide vane position coordinates, inlet / outlet diameter ratio, radius of curvature of the bend section, inlet wind speed, and turbulence intensity; core bandwidth. Used to adjust the model's sensitivity to local airflow disturbances caused by the deflector.

6. The design method of the spinning drawing air duct structure as described in claim 1, characterized in that: Step 4) specifically involves: by setting the design variable ranges for the number, position, radius of curvature, and inlet / outlet diameter ratio of the guide vanes, and combining the initialization population, fitness function, and genetic operations, the optimization model is constructed. Among them, Latin hypercube sampling was used to generate the initial population to ensure uniform coverage of the parameter space; The fitness function is defined as a weighted composite objective of the standard deviation of airflow uniformity σ, pressure loss ΔP, and energy consumption E, and its specific calculation formula is as follows: In the formula: α, β, γ are weighting coefficients; Genetic operations include tournament selection, uniform crossover, and Gaussian mutation, which generate offspring populations through continuous iteration.

7. The design method of the spinning drawing air duct structure as described in claim 1, characterized in that: Step 5) specifically involves: performing hydrodynamic verification on the Pareto front solution for each generation; if the relative error between the predicted value and the actual value of the support vector regression model is >10%, then the sample is marked as a high-error sample and added to the training set; after every 5 iterations, the support vector regression model is updated through an incremental learning strategy.

8. The design method of the spinning drawing air duct structure as described in claim 7, characterized in that: The incremental learning strategy is as follows: without retraining all samples, newly added high-error samples are incorporated into the training set; the hyperparameters C and γ of the support vector regression model are optimized only for the newly added samples, using the Bayesian optimization method, and the adjustment range is limited to ±10%.

9. The design method of the spinning drawing air duct structure as described in claim 1, characterized in that: The convergence condition for step 6) is specifically: the rate of change of the fitness function is less than 1% for three consecutive generations, or the optimization is terminated when the total number of iterations reaches 50 generations.

10. The design method of the spinning drawing air duct structure as described in claim 1, characterized in that: The engineering constraints in step 7) include: minimum spacing between guide vanes ≥ 50 mm; Total duct volume ≤ 0.5 m³ 3 ; The import / export ratio is ≤3:1.

Images