Optimization design method and system for turbomachinery, storage medium and computer program

Abstract

The invention relates to the technical field of fluid machine design and artificial intelligence crossing, in particular to an optimization design method and system oriented to turbomachinery, a storage medium and a computer program. According to the method, a standardized data bus of a parameter layer, a configuration layer, a geometric layer and a result layer penetrates through an average line, parameterized geometry and multi-objective optimization are carried out, thermophysical property calibers are unified, and sample-level archiving and playback are achieved. Through LHS / Sobol sampling, Kriging proxy, NSGA-II and other algorithms, isentropic efficiency, pressure ratio and structure / noise indexes are collaboratively optimized under the condition that constraints of mass flow, minimum thickness, tip clearance and the like are met; the system is coupled with CFD / FEM through an adapter, and supports GUI and batch processing. Compared with the prior art, the design efficiency, the cross-module consistency and the result traceability are improved, and a manufacturable Pareto optimal blade scheme is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fluid machinery design and artificial intelligence, in particular to an optimization design method and system for turbomachinery, a storage medium and a computer program. BACKGROUND

[0002] Turbomachinery (including centrifugal impellers, axial flow fans, compressor impellers, etc.) undertakes the core energy conversion task in aerospace, energy power and process industry, and its performance directly determines the efficiency, reliability and cost of the whole machine. The traditional design process of turbomachinery usually goes through: preliminary design, one-dimensional mean line calculation, two-dimensional airfoil / section design, three-dimensional blade / level geometric construction, numerical simulation evaluation, test or field data checking, design iteration. With the continuous improvement of the demand for high pressure ratio, high load, low noise and multi-working condition adaptability, the design problem presents the characteristics of "multi-objective, multi-constraint and multi-disciplinary coupling", prompting the engineering field to generally adopt the design mode based on parameterized geometry and intelligent optimization, that is, to introduce the idea of multidisciplinary design optimization (MDO) at the system level to efficiently search the large-scale design space with limited computing budget.

[0003] At the preliminary design / mean line analysis level, the public literature still generally uses the mean line method as the cornerstone of early schemes and performance estimation. In the traditional liquid pump field, Chinese invention patent CN102146932A combines similarity conversion method and velocity coefficient method for centrifugal pump impeller preliminary design based on one-dimensional flow theory, thereby improving efficiency while maintaining design simplicity, indicating that preliminary design based on simplified theory is still an important grasp of the industry.

[0004] Around the parameterized construction of two-dimensional section and three-dimensional blade, domestic patents not only retain the engineering realizability of classical airfoil families, but also introduce more strict geometric quality control to reduce losses. The programmable airfoil represented by double circular arc (DCA) is systematically explained in Chinese invention patent CN106250644B: the document gives the steps of aerodynamic quantity calculation, geometric parameter optimization of each radius section, and three-dimensional blade formation through radial stacking; DCA adjusts diffusion and shock through arc segment position and curvature adjustment, and has applicability in the range of subsonic to Ma≈1.2, and is easy to implement and transplant across products. At the same time, aiming at the problem that "traditional parameterization is difficult to ensure curvature continuity and easy to introduce additional losses at leading edge / trailing edge and connection", Chinese invention patent CN110727995A proposes an airfoil construction method taking curvature distribution as the main control quantity, emphasizing the role of curvature continuity and geometric fairing in suppressing separation and reducing losses, providing a new modeling approach for high-precision aerodynamic design.

[0005] In the optimization methodology and process, the "combination of DoE (design of experiment), surrogate model and multi-objective evolutionary algorithm" has become the mainstream practice in public patents. Chinese invention patent CN106650125A adopts the framework of "design of experiment, establishment of surrogate model, global optimization based on surrogate model", which reduces the engineering cost and time overhead caused by expensive simulation, and combines global search with reliability evaluation; this scheme reflects the general idea of covering the high-dimensional parameter space with statistical uniform sampling first, and then using the surrogate model to carry out evaluation and optimization. In the selection of specific optimizers, non-dominated sorting genetic algorithm is widely used: Chinese invention patent CN102251983B uses NSGA-II to optimize the geometric parameters of pump impeller for multiple objectives, and carries out compromise design for efficiency and anti-cavitation, which shows that the group evolution method under multi-objective and multi-constraint conditions has good engineering adaptability.

[0006] From the perspective of system platform and process integration, domestic patents have appeared the technical route of integrating CAD, CFD and optimizer into a unified "design optimization platform". CN106650105B integrates three-dimensional parameterized modeling, meshing and solving, and numerical optimization into an automatic pipeline through platformization, and gives the results of overall lifting of efficiency curve and expansion of generalized high-efficiency area, which shows that cross-software automatic integration can reduce manual back-and-forth, improve engineering throughput and consistency.

[0007] Overall, the above Chinese patent texts reflect the current technical baseline and development context of impeller mechanical design optimization: first, the average line / simplified theory is still an important starting point for engineering initial setting and performance estimation (the same for cross-medium and variable working condition scenarios); second, geometric parameterization evolves from traditional engineering parameters (chord length, thickness, curvature, etc.) to high-quality description of curvature control, fairness and shock / diffusion suppression; third, DoE+surrogate model+NSGA-II / III (or other global algorithms) constitute the general paradigm of "coverage-fitting-optimization-refinement"; fourth, platformization and automation become the direction of implementation, which connects parameterization-meshing-solving-optimization in series to support large-scale design space exploration.

[0008] However, combined with the above public texts, some engineering-level issues / pain points can also be observed: 1. Cross-module data consistency and reproducible experiments: public patents usually focus more on how to design / optimization, and less on the unified data schema, sample-level artifact archiving and deterministic replay mechanism among average line, geometry, simulation and optimization; this will increase the cost when team collaboration, audit and secondary re-optimization.

[0009] 2. External physical properties and multi-disciplinary caliber unification: when involving multi-medium, cross-language library or external solver coupling, how to maintain the consistency of thermal physical properties and model caliber between average line, geometric evaluation and numerical simulation, the disclosure is given by the respective embodiments, but there is a lack of industry-wide through specification.

[0010] 3. Manufacturing / constraint explicit and interpretable optimization: although there have been attempts to target curvature, Q criteria and entropy production, under more complex manufacturing and multi-physical constraints, how to integrate interpretable physical quantities and engineering implementability system into the parameterization and optimizer interface is still a direction worth continuing to deepen. SUMMARY

[0011] To solve the above technical problems, the purpose of the present application is to provide an optimization design method for turbomachinery, which is an integrated turbomachinery design optimization platform deeply integrating average line design, parameterized geometry and multi-objective optimization. The method improves design efficiency, cross-module consistency and result traceability through standardized data interface and replayable process.

[0012] To achieve the above purpose, the present application adopts the following technical solutions: An optimization design method for turbomachinery, comprising the following steps: S1) Obtain model, design point and medium information, and write to parameter layer; S2) In the average line modeling module, perform inter-stage aerodynamic distribution and one-dimensional performance estimation based on the parameter layer workpiece, and write the obtained results to the configuration layer; S3) According to the configuration layer workpiece, convert the average line results into controllable blade parameterized model through defined geometric mapping relationship, generate two-dimensional section and three-dimensional blade shape, and write point set / grid to geometry layer; S4) In the optimization scheduling module, take the design variables of the parameterized model as the decision vector, use the combination strategy of hierarchical sampling initialization and multi-objective population evolution optimization to solve, and establish a sample-level independent directory for each design evaluation; archive the parameter layer, configuration layer, geometry layer workpieces and evaluation logs and target / constraint values in the sample-level independent directory to form the result layer; S5) In the result analysis module, visualize the objectives and variables of multiple samples, including Pareto front and geometric superposition comparison, and support local re-optimization of the neighborhood of the labeled sample under the premise of maintaining the consistency of the four-layer workpieces.

[0013] As preferred, in step 2), the medium thermophysical parameters of density, specific heat, and viscosity are called by a cross-language external interface from the thermophysical property library to participate in the average line equation set, the cross-language external interface is a function call interface of an external language, through which the state quantities of air, steam, or refrigerant are accessed from the thermophysical property library, and the calculation strategy of the state quantities is the same in the average line modeling and subsequent evaluation, to ensure that the physical quantity caliber of the multi-module is consistent, and the calculation strategy includes table interpolation or a state equation.

[0014] As preferred, in step 3), the geometric mapping relationship includes chord length, camber line, maximum thickness position, and interpolation distribution along the spanwise, the geometric mapping relationship adopts a mixed parameterization of piecewise Bezier control and spanwise interpolation: the basic parameters include chord length, leading edge thickness, trailing edge thickness, and maximum thickness position; the engineering parameters include twist, sweep, lean, and camber; and the advanced parameters include tip clearance, blade number, and fillet radius, and the above parameter changes are all written back to the geometric layer workpiece in real time to maintain the correlation with the result layer.

[0015] As preferred, in step 4), the layered sampling initialization adopts Latin hypercube sampling (LHS) or Sobol sequence to cover the design space; the multi-objective population evolution optimization adopts a combination of NSGA-II and genetic algorithm, and is executed in parallel on a multi-core processor, and the optimization process has the ability of interruption and recovery, and the hyperparameters of different stages are recorded in the result layer workpiece; the geometric layer workpiece includes two-dimensional cross-sectional curves and three-dimensional blade grids, supports output in a standard exchange format and is used for third-party grid generation; in the multi-stage unit scene, the geometric generation traverses the rotor and stator and interpolates on at least three spanwise sections to control the cross-sectional smoothing and thickness distribution.

[0016] As preferred, in step 4), the optimization scheduling module is coupled with external physical solvers including computational fluid dynamics (CFD) and finite element analysis (FEM) through a black box adapter, the adapter converts the geometric layer workpiece into a grid or a solving input, and structures the key indicators of the solving result back to the result layer workpiece, while retaining the grid and log to support auditing and recalculation, and the key indicators include efficiency, pressure ratio, stress, deformation, and noise indicators.

[0017] As preferred, steps 1) to 5) are organized by a unified data bus mechanism, the intermediate workpieces in the four layers of the parameter layer, the configuration layer, the geometric layer, and the result layer are all serializable standardized data structures, realizing lossless transmission across modules, deterministic reproduction of experiments, and playback of the flow.

[0018] As preferred, the user interface layer shares the uniform schema of the four-layer artifacts with the scripting engine layer, and each parameter interaction on the interface side triggers the incremental serialization and version number update of the corresponding artifact, so as to realize the intercommunication of graphical interaction / command line batch processing and the consistency verification across the schema.

[0019] Further, the present application also provides a design optimization system for turbomachinery, which is used to realize the method, and comprises: an input and average line module, which is used to perform S1 and S2 and generate the parameter layer and configuration layer artifacts; a geometric parameterization module, which is used to perform S3 and generate the geometric layer artifact; an optimization scheduling and evaluation module, which is used to perform S4 and establish the sample level directory and the result layer artifact; a result analysis module, which is used to perform S5 and visualize and locally re-optimize the target, variable and geometry; a data bus and artifact management module, which is used to maintain the four-layer serializable intermediate artifacts and their versions, dependency relationships and playback control; an external interface module, which is used to access the thermophysical property library and external physical solver through a cross-language interface.

[0020] Further, the present application also provides a computer readable storage medium, which stores a computer program, and the program makes the computer realize the method when executed by a processor.

[0021] Further, the present application also provides a computer program product, which comprises a computer program or instructions, and the computer program or instructions realize the method when executed by a processor.

[0022] The present application makes the average line, geometric construction, CFD / FEM evaluation and optimization solution run cooperatively under the same semantics through the unified data artifacts and external caliber, realizes the consistency of cross-module results and the deterministic reproduction of experiments; the sample level independent directory and playback strategy significantly reduce the errors caused by artificial intervention and repeated modeling, shorten the iteration convergence period; the combination of parameterized geometry and multi-objective evolutionary optimization improves the design space coverage and compromise solution discovery ability, facilitates the acquisition of manufacturable optimal scheme under multiple indexes such as efficiency, pressure ratio, stress / deformation and noise; at the same time, the standard format of geometry and result output can be directly connected to the grid and post-processing, which is convenient for auditing, reviewing and secondary local re-optimization, and improves the design efficiency, traceability and engineering landing as a whole. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The flowchart of the method of the present application.

[0024] Figure 2The comparison chart of the original design and the optimized design of the axial flow fan designed for the examples of the present application, Figure 2 The original design is in medium blue; the optimized design is in red. DETAILED DESCRIPTION

[0025] The technical solutions in the examples will be clearly and completely described below. Obviously, the described examples are only some of the examples of the present application, but not all the examples. Based on the examples in the present application, all the other examples obtained by those skilled in the art without creative work belong to the protection scope of the present application.

[0026] I. Overall architecture and data organization 1. Overall structure of the system As shown in the figure, the software of the present application takes the four-layer serializable data artifacts of “parameter layer-configuration layer-geometry layer-result layer” as the data bus, penetrates through the six functional modules of basic parameters, mean line design, blade geometry generation, blade parameterized modeling, optimization setting, and optimization result, and forms the closed-loop process from preliminary design to optimization analysis. The above-mentioned modules can be interactively run in the graphical interface, or can be run in batch mode in the script mode, and both share the same data schema. Figure 1 2. Four-layer data artifacts and storage

[0027] Parameter layer: store the model, design points (mass flow, total pressure ratio, rotating speed, inlet state, etc.), working medium, and rotating direction, etc.; preferably saved in JSON or MAT structure. Configuration layer: store the stage parameters, loss models, span interpolation strategies, and one-dimensional calculation results (velocity triangle, reactivity, diffusion factor, efficiency estimation, etc.) required for mean line modeling.

[0028] Geometry layer: store the two-dimensional section point list and three-dimensional blade surface data, and support export of standard formats (.dat / .stl / .igs, etc.).

[0029] Result layer: archive the input variables, geometry, grid, simulation log, and target / constraint values (CSV / MAT) of each evaluation in a sample-level independent directory, for example, Results / Design_001 / …; and in Results / Summary / , summarize the Pareto front, sensitivity, and visualization results.

[0030] 3. Cross-language consistent interface

[0031] ​To ensure the consistency of average line, geometric evaluation and numerical simulation in medium thermal properties, the same thermal property library is uniformly called through a cross-language function interface. The interface name, version and calling parameters are written in the parameter layer and configuration layer workpieces; when the library is missing or the state point is out of bounds, the system gives clear error information in the interface and log and safely aborts the current sample, avoiding the propagation of error results.

[0032] II. Implementation of functional modules (I) Basic parameter module The user inputs in the interface: model (such as axial flow fan), design point (mass flow, total pressure ratio, speed, inlet total pressure / total temperature, etc.), working medium (air / steam / refrigerant, etc.), rotation direction (clockwise / counterclockwise). The system checks the units and ranges (such as total pressure ratio ≥1, temperature >0K, flow >0, etc.), and generates parameter layer workpieces after passing the check, which are used as the unified entrance for subsequent calculations.

[0033] (II) Average line design module Based on reading the parameter layer data, the user sets the stage parameters (number of stages, reactivity, inlet / outlet angle, thickness ratio, diffusion factor, axial velocity ratio, hub / rim radius ratio, etc.), and selects the airfoil library (such as NACA, DCA or custom). Calculation process: 1. Uniformly call the thermal property library to obtain medium density, specific heat, viscosity, etc. according to the inlet state; 2. Build a velocity triangle and perform stage load distribution, calculate reactivity, diffusion factor, etc. 3. Perform one-dimensional performance estimation using empirical loss / efficiency models to obtain efficiency, pressure drop distribution, and axial length, etc.

[0034] Output and check: output configuration layer workpieces (including spanwise stations and geometric / aerodynamic boundaries at each station), and perform preliminary legality checks, including monotonicity of passage area, reasonableness of local Mach number and diffusion, etc. The qualified result is transferred to the geometric module with one key.

[0035] (III) Blade geometry generation module The system generates multiple cross-section two-dimensional airfoils according to the span (such as 0%, 20%, 50%, 80%, 100% blade height), and performs spanwise extrapolation of chord length, thickness, camber, installation angle according to the selected interpolation strategy (linear / spline / Bezier, etc.), to construct a three-dimensional blade envelope. The interface allows rotation, scaling and layered viewing, and automatically checks constraints such as minimum thickness, leading / trailing edge radius, tip clearance, etc.; after confirmation, export the geometry layer workpiece (two-dimensional cross-section file and three-dimensional surface file) for use by the parameterization module.

[0036] (IV) Blade parameterization modeling module This module converts baseline geometry into controllable parametric model, variables are divided into three levels: 1. Basic parameters: Chord, ThickMax, ThickLE, ThickTE; 2. Engineering parameters: Twist, Sweep, Lean, Camber / CamberPosition; 3. Advanced parameters: TipClearance, BladeNumber, FilletRadius, etc.

[0037] • Recommend to control the number of variables between 20-40, and set upper and lower bounds and spanwise discrete points for each variable, typical range (relative to baseline) can be taken: Chord ±5%-10%, ThickMax ±8%-12%, Twist peak change ±3°-5°, Sweep / Lean in symmetric interval at millimeter level, FilletRadius floating within manufacturing limit.

[0038] Constraints and fairing: Real-time check of thickness lower limit, curvature continuity, avoidance of self-intersection and minimum fillet when changing parameters; if constraints are violated, roll back and prompt.

[0039] Output: Save variable list and spanwise distribution, record "baseline-modified" two / three-dimensional comparison chart as a benchmark for the optimization phase.

[0040] (Five) Optimization Settings Module 1. Experimental design: Choose Latin hypercube or Sobol uniform sampling method, initial sample size generally takes 5-10 times the variable dimension.

[0041] 2. Surrogate model (optional): Prefer Kriging / Gaussian process, use cross-validation to evaluate model error; trigger real simulation supplement when prediction error exceeds threshold, gradually improve surrogate accuracy.

[0042] 3. Optimization algorithm: According to the nature of the target, choose multi-objective genetic algorithm (MOGA / NSGA-II / NSGA-III), particle swarm or simulated annealing, etc. Give reasonable initial value to population size, generation number, crossover and mutation probability, and set stop criteria (upper limit of generation number, convergence of front, target stability, etc.).

[0043] 4. Objectives and constraints: Objectives: Maximum isentropic efficiency, maximum total pressure ratio, can add noise / entropy production / stress / mass, etc. as needed; Constraints: mass flow deviation (e.g. ±1.5%), total pressure or pressure ratio lower limit, minimum thickness and tip clearance, outlet angle deviation, material and rotational speed stress constraints, etc.

[0044] 5. Scheduling and parallelism: automatically create independent directories for each candidate sample at runtime, perform batch processing of "geometry update - mesh - solve - post-processing - archive"; automatically set parallelism according to hardware; support mid-way pausing and resuming, read in population, generation number and random seed to continue computation upon resuming.

[0045] (VI) Optimization result module 1. Pareto front: show target scatter points and non-dominated front, support filtering of trade-off solution set meeting engineering bottom line.

[0046] 2. Sensitivity analysis: based on proxy model partial derivative, regression coefficient or Sobol index, calculate global / local influence degree of variables on target, form column-line mixed chart, facilitate identification of key spanwise sections and key geometric quantities.

[0047] 3. Geometry comparison: compare differences of multiple candidate solutions on two-dimensional cross-section and three-dimensional surface, support hierarchical viewing by blade height, and output difference colorized chart.

[0048] 4. Export: one-click export final geometry, variable table and performance report to summary directory; provide basis for subsequent review, review or secondary optimization.

[0049] III. Complete flow from preliminary design to multi-objective optimization (as shown in Figure 1 ) 1. Set the model to single-stage axial flow fan, input mass flow, total pressure ratio, rotational speed and inlet total pressure / total temperature; medium is air, rotational direction is clockwise.

[0050] 2. In the mean line module, select DCA as the airfoil family, set reactivity to 0.45-0.55, diffusion factor to meet the upper limit of engineering, and spanwise station number to not less than 5; complete one-dimensional calculation and output spanwise distribution and efficiency estimation.

[0051] 3. Enter the geometry module to generate two-dimensional airfoils of five spanwise sections, construct three-dimensional blades using spline interpolation, check minimum thickness and leading / trailing edge fillets, and export three-dimensional geometry.

[0052] 4. In the parameterization module, select 20-30 variables such as chord length, maximum thickness, twist, forward / sweep, radial / axial offset and root fillet, set the recommended range, complete constraint check and export variable configuration.

[0053] 5、 In the optimization setting, the double objectives are "maximum isentropic efficiency and maximum total pressure ratio"; the constraints are mass flow deviation, minimum thickness, tip clearance and outlet angle deviation; the initial samples are obtained by Latin hypercube sampling, Kriging is used as the surrogate model, NSGA-II is used as the optimizer, and reasonable population size and generation number are given.

[0054] 6、 Start batch processing, the system automatically creates a directory for each sample and completes the grid and solution; mark and exclude failed samples (grid failure, convergence failure or property out of bounds) from the feasible solution.

[0055] 7、 After the calculation is completed, view the Pareto frontier and sensitivity ranking in the results module, select "high efficiency solution", "trade-off solution" and "high pressure ratio solution" as three representative schemes, and export the corresponding geometry and report.

[0056] 8、 If further improvement of engineering usability is required, reduce the upper and lower bounds of the variables around the trade-off solution, and perform 20-30 generations of local re-optimization to achieve a better balance between performance and manufacturing changes.

[0057] When there is an existing external blade profile curve file, you can skip the average line step, directly import the curve and bind the parametric variables, and enter the optimization setting and solution process. To maintain the original aerodynamic load distribution, you can fix the outlet angle or passage area distribution in the constraints.

[0058] Introduce structural disciplines in the optimization setting: add "minimum maximum equivalent stress" and "avoid frequency band" as objectives or constraints; add FEM grid and calculation log to the results layer directory; you can use a "first aerodynamic, then coupling" hierarchical strategy or a multi-objective joint optimization strategy.

[0059] Four, algorithm and numerical setting suggestions 1、 Experimental design and surrogate model: the number of initial samples should be 5-10 times the number of variables; Kriging uses anisotropic correlation length, and hyperparameters are estimated by maximum likelihood or cross-validation; when the proxy prediction error exceeds the threshold, additional real simulation samples are added adaptively.

[0060] 2、 Optimizer parameters: NSGA-II uses crowding distance to maintain diversity; discrete variables (such as the number of blades) use hybrid encoding and repair operators; for individuals that violate hard constraints, use feasibility priority or penalty function strategy.

[0061] 3、 Grid and solution: increase the tip clearance area; control the wall size as needed (such as y+ less than a set threshold or use wall functions); give the total pressure / total temperature at the inlet and the static pressure at the outlet or mass flow feedback; the convergence criterion is the stability of the residual and key objectives.

[0062] 4. Consistency check: geometric check (minimum thickness, hub-to-tip ratio, no self-intersection), thermophysical property state within library range, one-dimensional estimation and post-processing with unified reference surface and weighting method.

[0063] 5. Playback and audit: record software version, script hash, random seed, optimization and grid hyperparameters to Replay.json, any sample can be reproduced with one key; export report containing variable set, objective / constraint, convergence curve and geometry snapshot.

[0064] The following will take the design optimization of an axial flow fan as an example to explain the present application in detail. The specific functions and mutual relationships of each module are highlighted to realize the automation and high efficiency of axial flow fan design.

[0065] The goal of the axial flow fan multi-objective optimization platform (TurboXOPT) is to perform preliminary design of turbomachinery based on basic parameters (such as mass flow, total pressure ratio, etc.), then parameterize the blades through blade modeling technology, and finally optimize the preliminary designed blades from the aspects of aerodynamics and structure using optimization algorithms (or direct optimization from known blade parameters), so as to obtain high-performance blades that meet the requirements of industrial use. The software includes six parts: basic parameters, mean line design, blade geometry generation, blade parameterized modeling, optimization settings, and optimization results.

[0066] (1) Basic Parameters: Machine type: Select different types of turbines, such as fans or compressors, etc. Design point: Set the design operating condition information, such as flow rate, rotational speed, pressure ratio, etc. Working medium: Specify the working medium (air, steam, or other gases).

[0067] Rotational direction: Determine the direction of impeller rotation (clockwise / counter-clockwise). The output of this step is generally the basic input file or basic data required for preliminary design, providing a calculation benchmark for subsequent modules.

[0068] (2) Meanline Design: Define the stage parameters: According to the design operating condition information, further determine key parameters such as stage number, inlet / outlet angle, aerodynamic load distribution, etc.

[0069] Select the airfoil: Choose an appropriate airfoil based on a database or user-defined preferences, providing a reference shape for subsequent blade design. Generate configuration file & RUN: Generate a configuration file based on input parameters, followed by one-dimensional / mean line design calculations to obtain performance estimates. This step can output relevant result files, such as flow rate, efficiency, and pressure distribution after mean line design, providing references for the blade geometry generation module.

[0070] (3) Blade Geometry: Plot 2D / 3D airfoil: Based on mean line design and selected airfoils, generate and visualize 2D or 3D blade shapes. Plot different component curves with different spans: Show geometric curves at multiple cross-sections from the root to the tip (span-wise) to observe changes in blade shape. Output: 2D curve file / 3D geometry file: Save generated geometric information as files to prepare for subsequent parameterized modeling and meshing. Through the Blade Geometry module, users can initially check and modify blade shapes to ensure they meet design requirements.

[0071] (4) Blade Parameterization: Read curve file: Obtain the original blade surface or curve from the geometry file output by the previous module. Select the parameterization method: Define design parameters: such as camber distribution, thickness distribution, and twist angle distribution. Plot: 2D / 3D baseline and modified blade: Real-time visualization of the blade shape before and after parameterization to determine the appropriate parameter range. Output: configuration file (Design variables): Write all design variables into the configuration file for easy access during optimization.

[0072] (5) Optimization Setup: Generate initial samples according to the range of design variables and defined sampling algorithm (e.g. LHS). Select the optimization type: Single-objective or multi-objective optimization algorithm can be chosen, such as genetic algorithm, particle swarm optimization, etc. Define the objectives and constraints: e.g. maximize aerodynamic efficiency, minimize structural stress, constrain outlet flow angle, etc. Generate configuration file & RUN: Generate configuration file containing objectives, constraints and algorithm parameters. The software will automatically create separate folders for each candidate blade design, run numerical simulation or analysis (e.g. fluid dynamics analysis, vibration analysis, etc.), and save the results (curves, geometry, mesh, etc.). User can stop the optimization at any time.

[0073] (6) Optimization Results: Select the No. of blade designs to compare: From all the completed candidate designs, select several optimal or interesting solutions. Plot: Objectives and design variables: Compare the differences in aerodynamic efficiency, stress distribution, etc. of each design, and understand the changes in key geometric parameters. Plot: 2D / 3D blade: Visualize the differences in the shape of each candidate design. Output: 2D / 3D plot: Export and save the comparison results or the final design shape. In Optimization Results, the user can quickly locate the best or compromise solution, and conduct in-depth analysis of the performance and shape differences of different designs.

[0074] In the interface shown, the user can perform fine geometric design and parameterization settings for the blade shape of the axial fan. In the drop-down menu, the specific method of parameterized modeling can be selected. The example in the figure selects "Traditional", which is based on traditional geometric features (chord length, thickness distribution, twist, etc.) to parameterize the blade. "FileName" specifies the curve file to be read or output (e.g. small_fan.curve), and the unit used (e.g. mm) can be set.

[0075] Parameterization parameters include: (1) Basic parameters (Unit: %) Chord (chord length), ThickMax (maximum thickness), ThickLE (leading edge thickness), ThickTE (trailing edge thickness), etc.: These parameters determine the thickness distribution and chord length variation of the blade at different blade heights. LowerBound / UpperBound: Respectively represent the minimum and maximum allowable change amount (for example -0.3 to 0.3) of the parameter during optimization or design adjustment, used to limit the modification range of the design variable. NumPoints: Used to specify the number of discrete points when the blade is segmented and interpolated or polynomially fitted in the spanwise direction. (2) Engineering parameters Twist (twist), Sweep (forward sweep / backward sweep), Lean (axial inclination), CamberPosition (camber line position), etc. These parameters can be used to control the overall shape of the blade in three-dimensional space from the root to the tip, for example: how does the twist angle change with the spanwise direction, how does the blade centerline lean or sweep. Also contains LowerBound / UpperBound and NumPoints, so that users can flexibly define the adjustment range and resolution during design or optimization. (3) Advanced parameters TipClearance (top clearance), NumBlades (number of blades), FilletRadius (root fillet), etc. More engineering-related geometric features can also be included in the parameterization range, so as to more realistically simulate and evaluate the entire fan system.

[0076] The buttons on the right side, such as "T1Add", "T1Remove", "T2Add", "T2Remove", etc., facilitate the addition and deletion of specified design variables in the 2D or 3D parameter table. For example, to add a thickness adjustment for a specific blade height, "T1Add" or "T2Add" can be used for insertion; if not needed, it can be deleted. "NumLayers" indicates the number of layers divided for the blade when using parameterized modeling; "Num_DVs" indicates the total number of design variables selected for this design / optimization.

[0077] Two-dimensional Graphics: Display the shape of the blade section in the 2D plane. For example Figure 2As shown, the blue curve generally represents the original original airfoil, and the red curve represents the new airfoil after adding or modifying several design variables, which allows users to instantly view the change effect. Three-dimensional Graphics: Display the shape of the current blade in 3D space, and can be refreshed in real time with the update of the design variable. Users can rotate and translate the view to observe the geometric changes of the blade, the changes of the leading and trailing edges, and the position relationship in the overall flow passage.

[0078] Based on these settings and visualization means, TurboXOPT can help users quickly build a blade shape model suitable for axial flow fan applications and allow it to be adjusted and optimized in multiple dimensions. By considering key geometric parameters such as chord length, thickness, twist, sweep, root fillet, etc., users can accurately control the aerodynamic characteristics and structural performance of the blade, laying a solid foundation for subsequent numerical calculation and multi-objective optimization.

[0079] In the "Optimization Setup" interface of TurboXOPT, the axial flow fan blade can be defined with isentropic efficiency and pressure ratio as optimization objectives. The following is a more detailed description of its settings and processes: In the "Optimization targets" area on the left, you can select the physical quantity of interest from the "Catalog" drop-down menu, such as aerodynamic performance or structural stress. For this example, we will focus on two metrics within "Aerodynamic": isentropic efficiency and pressure ratio. In the "Objectives" panel, use the "Obj_add" button to add an objective function. Efficiency has already been added in this example. To add pressure ratio as another objective, you can add another objective, such as pressure_ratio or a similarly named objective. In the "Constraints" area, you can set constraints for the design process. For example, you can maintain certain flow characteristics, geometric dimensions, or mechanical properties within a specified range. The total_pressure constraint shown in the figure, with both lower_bound and upper_bound set to 0, is for demonstration purposes only. In practice, you can set it to meet minimum / maximum total pressure (or pressure ratio) requirements. Sampling_methods: You can select Latin hypercube sampling, etc., to generate initial sample points for conducting experimental designs within the multidimensional design space. Num_samples: Determines the number of initial sample points. More sample points provide richer initial training data, but also increase the computational complexity. Surrogate_models: You can select "Kriging" (often used to capture complex nonlinear relationships) or other surrogate models. This model learns the mapping between the objective function and design variables based on the numerical simulation results of the initial samples, accelerating the subsequent optimization process. In this example, MOGA (Multi-Objective Genetic Algorithm) is used as the solution strategy. It can handle multiple objectives simultaneously and iteratively obtain a series of non-dominated solutions (Pareto Front). Users can select different algorithms (such as PSO and NSGA-II) using the drop-down menu and adjust parameters such as crossover and mutation probability in the configuration file. After clicking the "Generate" button to generate the optimization configuration file, press "RUN" to begin the automated optimization process. The software iterates and evaluates multiple blade designs (creating separate folders to store the geometry, mesh, and analysis results for each design).

[0080] During the calculation process, the changes of blade geometry can be viewed in real time, and the effects of twist angle, thickness, leading edge, and trailing edge on the aerodynamic shape can be understood. During the iteration process, the software records the objective function values obtained in each evaluation, helps to observe the convergence trend and the distribution of solutions, and is presented in the Calculation Process chart. If a surrogate model is used, the system will evaluate the model accuracy according to the difference between the real simulation and the agent prediction, and continuously update the agent model to improve the prediction ability, which is presented in the Model Accuracy chart. After the calculation is completed, the “post-process” or the corresponding result output button can be clicked to compare the performance of different iterations of the blade design in terms of isentropic efficiency and pressure ratio, or to view the distribution of each solution on the Pareto frontier. Finally, the user can select several designs in the “Optimization results” module for detailed comparison, including viewing the target values, design variables, and 2D / 3D blade shape, so as to screen the most suitable scheme for the actual working condition.

[0081] In summary, by setting the isentropic efficiency and pressure ratio as the two aerodynamic indicators in the optimization interface of TurboXOPT, combining with appropriate surrogate models and multi-objective genetic algorithms, the candidate blade designs can be continuously iterated, searched, and evaluated in the controllable design variable space, and finally the blade scheme with high efficiency and high pressure ratio for the axial flow fan can be obtained.

[0082] In the Optimization Results module, the software provides multiple visualization methods to help users compare and analyze the performance and geometry differences of different blade designs. The main parts include: (1) Define No. Blade Design: Users can input or select the blade design numbers they want to compare (such as BladeNo 10, Design-No21, etc.) in this section, and click the "Plot" button to generate the corresponding graphs. When multiple designs are selected, the system will display their differences in various objective functions and design variables simultaneously, making it easy for users to quickly identify which solution is better or more suitable for their needs. (2) Sensitivity Analysis: In the interface, there is a mixed column chart and line chart representing the "value" (column) and "sensitivity" (line) of each design variable. "Sensitivity" represents the degree of influence of the variable on the objective function or performance indicator. The larger the value, the more significant the small changes in the variable will affect the target. By comparing the variables of Design-No21 and Design-No10, such as chord length, maximum thickness, 100% blade height twist, 80% blade height sweep, and lean, users can intuitively see the variable differences and their impact on performance.

[0083] The column chart displays the specific values of the two designs in the objective functions (e.g., efficiency and tp). By color differentiation (blue and orange), users can quickly compare the high and low of each target and evaluate the trade-off relationship in the multi-objective scenario. 2D-Comparison: This mainly shows the airfoil profile comparison on the same section. 3D-Comparison: The difference in blade shape is more intuitive in a three-dimensional perspective. Through color mapping, users can view the changes in blade surface between different designs, such as the change in twist angle, camber, thickness, and other regions. The "Span" input box in the interface allows users to select specific sections to view the comparison of blade height, or to browse the changes from the root to the tip of the blade layer by layer. The "Output_2D" / "Output_3D" buttons can export the current visualization results as graphic files or data files, making it easy to record and share. Users can also perform further operations, such as viewing numerical results, reading flow field data, or applying sensitivity analysis results to the next stage of multi-objective optimization.

[0084] Overall, by comparing the performance indicators, geometric variables, and sensitivity of different design schemes, users can quickly identify the advantages and disadvantages of each design, and select or continue to adjust the final blade scheme that best meets industrial needs.

[0085] The above is the description of the embodiments of the present application. Through the above description of the disclosed embodiments, those skilled in the art can implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0086] Those skilled in the art will appreciate that embodiments of the present application can be devised for use with methods, systems, or computer program products. Accordingly, the present application can be embodied in a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0087] The present application is described with reference to the flowcharts and / or block diagrams of the methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowcharts and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flowchart or flows and / or block or blocks. Figure 1 one or more functions specified in the flowchart or flows and / or block or blocks.

[0088] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the functions specified in the flowcharts and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flowchart or flows and / or block or blocks. Figure 1 one or more functions specified in the flowchart or flows and / or block or blocks.

[0089] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowcharts and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flowchart or flows and / or block or blocks. Figure 1 one or more functions specified in the flowchart or flows and / or block or blocks.

[0090] In one typical arrangement, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0091] Memory can include non-persistent memory and / or volatile memory, such as a random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM), EPROM, and / or a flash memory. Memory is an example of computer-readable media.

[0092] Computer-readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile discs (DVDs) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media, such as modulated data signals and carrier waves.

Claims

1. An optimization design method for turbomachinery, characterized in that: The method comprises the following steps: S1) Obtain the machine model, design point and medium information and write them into the parameter layer; S2) performing inter-stage aerodynamic distribution and one-dimensional performance estimation based on the parameter layer artifacts in the average line modeling module, and writing the obtained results into the configuration layer; S3) based on the configuration layer artifact, converting the average line result into a controllable blade parameterized model through a defined geometric mapping relationship, generating a two-dimensional cross section and a three-dimensional blade shape, and writing the point set / mesh into the geometry layer; S4) In the optimization scheduling module, the design variables of the parameterized model are used as decision vectors, and a combined strategy of stratified sampling initialization and multi-objective population evolutionary optimization is adopted to solve the problem. At the same time, a sample-level independent directory is established for each design evaluation. The parameter layer, configuration layer, geometry layer artifacts, evaluation logs, and target / constraint values ​​are archived in the sample-level independent directory to form a result layer. S5) The results analysis module visualizes the targets and variables of multiple samples, including Pareto front and geometric overlay comparison, and supports local reoptimization of the neighborhood of the labeled samples while maintaining consistency across four layers of artifacts.

2. The method according to claim 1, characterized in that In step 2), the thermophysical property library is called through a cross-language external interface to calculate the medium thermophysical parameters such as density, specific heat, and viscosity to participate in the average line equation system. The cross-language external interface is a function call interface of an external language. Through this interface, the thermophysical property library is accessed to calculate the state quantity of air, steam, or refrigerant. The calculation strategy of the state quantity shares the same implementation in the average line modeling and subsequent evaluation to ensure the consistency of the physical quantity caliber of multiple modules. The calculation strategy includes table interpolation or state equation.

3. The method according to claim 1, characterized in that The geometric mapping relationship in step 3) includes the chord length, camber line, maximum thickness position, and their interpolated distribution along the span. A hybrid parameterization of segmented Bezier control and spanwise interpolation is adopted: basic parameters include chord length, leading edge thickness, trailing edge thickness, and maximum thickness position; engineering parameters include torsion, axial offset, radial offset, and camber; advanced parameters include top clearance, number of blades, and fillet radius. Changes in these parameters are written back to the geometry layer artifact in real time to maintain the association with the result layer.

4. The method according to claim 1 or 3, characterized in that The stratified sampling initialization described in step 4) adopts Latin hypercube sampling or Sobol sequence to cover the design space; the multi-objective population evolutionary optimization adopts a combination of NSGA-II and genetic algorithm, and is executed in parallel on a multi-core processor. The optimization process has interruption-recovery capability, and the hyperparameters at different stages are recorded in the result layer artifact; the geometric layer artifact includes two-dimensional section curves and three-dimensional blade meshes, which support output in a standard exchange format and are used for third-party mesh generation; in a multi-stage unit scenario, the geometric generation traverses the rotor and stator and interpolates on at least three spanwise sections to control the section smoothness and thickness distribution.

5. The method according to claim 4, characterized in that The optimization scheduling module described in step 4) couples external physical solvers, including computational fluid dynamics and finite element analysis, with a black box adapter. The adapter converts the geometry layer artifact into a mesh or solution input, and writes the key indicators of the solution results back to the result layer artifact in a structured manner, while retaining the mesh and log to support auditing and recalculation. The key indicators include efficiency, pressure ratio, stress, deformation and noise indicators.

6. The method according to claim 1, characterized in that Steps 1) to 5) are organized by a unified data bus mechanism. The four intermediate artifacts of the parameter layer, configuration layer, geometry layer, and result layer are all serializable standardized data structures, realizing lossless transmission across modules, deterministic reproducible experiments, and process replay.

7. The method according to claim 1, characterized in that The user interface layer and the script engine layer share the unified model of the four layers of artifacts. Each parameter interaction on the interface side triggers the incremental serialization and version number update of the corresponding artifact to achieve interoperability of graphical interaction / command line batch processing and cross-mode consistency verification.

8. A design optimization system for turbomachinery, characterized in that: The system is used to implement the method according to any one of claims 1 to 7, comprising: Input and average line module, used to execute S1 and S2 and generate parameter layer and configuration layer artifacts; Geometry parameterization module, used to execute S3 and generate geometry layer artifacts; Optimize the scheduling and evaluation module to execute S4 and build sample-level catalogs and result-level artifacts; Results analysis module, used to execute S5 and visualize and locally reoptimize the objectives, variables, and geometry; The data bus and artifact management module is used to maintain four layers of serializable intermediate artifacts and their versions, dependencies, and playback control; The external interface module is used to access the thermal property library and external physics solvers through a cross-language interface.

9. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the computer implements the method according to any one of claims 1 to 7.

10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images