An intelligent design and simulation system for flow-through components of fluid machinery

Abstract

The present application relates to the technical field of fluid machinery design, and discloses an intelligent design simulation system for fluid machinery flow components, including: a multi-physics modeling module for constructing fluid mechanics, structural mechanics, and heat conduction models based on preset boundary conditions and physical properties; an intelligent optimization decision module for constructing a flow performance optimization objective function; a collaborative calculation module for constructing a comprehensive performance evaluation model based on the candidate design parameter set; a virtual experiment feedback module for reconstructing the corresponding flow component structural model according to the multi-objective evaluation results; and a scheduling control module for controlling the algorithm path selection and computing resource allocation during the optimization process. The present invention adopts a method that combines multi-physics coupling modeling with a dynamic feedback mechanism, and continuously adjusts the design parameters through real-time feedback simulation data, so that the optimization process can be dynamically updated according to the actual operating conditions, thereby improving the optimization convergence speed and the effect of global performance.

Description

Technical Field

[0001] The present invention relates to the technical field of fluid machinery design, and in particular to an intelligent design simulation system for flow-through components of a fluid machinery. Background Art

[0002] With technological advancements, fluid machinery flow components are increasingly used in various engineering systems, particularly in high-precision fields such as aerospace, energy, and petrochemicals. The design of these components is directly related to system performance and safety. Traditional fluid machinery design often relies on empirical methods, requiring gradual iteration and manual adjustments to meet design requirements.

[0003] Existing multi-physics coupled simulation methods have been applied to the design of flow-through components in fluid machinery. Traditional computational fluid dynamics (CFD) techniques provide information such as fluid flow and pressure distribution, while structural mechanics simulation helps designers predict material strength and deformation, and heat conduction simulation analyzes temperature distribution. These techniques typically utilize the finite element method (FEM) and finite volume method (FVM) to separately handle the calculations for each physical field, helping engineers optimize designs.

[0004] However, existing simulation technologies for fluid machinery flow components often overlook coupling phenomena, making it difficult to accurately predict their performance under actual operating conditions. Optimization methods often rely on manual intervention and are prone to falling into local optima during multi-objective optimization, lacking a global perspective. Even when advanced optimization techniques such as genetic algorithms are employed, these methods often have fixed initial parameters and optimization paths, making them difficult to adjust in real time based on simulation results. Therefore, the present invention provides an intelligent design and simulation system for fluid machinery flow components to address these shortcomings. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides an intelligent design and simulation system for flow-through components of fluid machinery, which solves the problems of insufficient multi-physical field coupling modeling, inflexible optimization path adjustment, and inaccurate virtual experiment feedback.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent design and simulation system for flow-through components of fluid machinery, comprising:

[0007] Multi-physics modeling module, which is used to build fluid mechanics, structural mechanics, and heat conduction models based on preset boundary conditions and physical properties, and output coupled simulation parameters;

[0008] An intelligent optimization decision module is configured to receive the coupled simulation parameters, construct and define a flow performance optimization objective function, and generate an optimization path and a set of candidate design parameters through a behavior-driven optimization strategy and a deep reinforcement learning method;

[0009] A collaborative computing module is used to call a multi-physics field numerical simulator to jointly calculate the response results of each physical field based on the candidate design parameter set, build a comprehensive performance evaluation model, and generate a multi-objective evaluation result corresponding to each candidate design parameter;

[0010] A virtual experiment feedback module is used to reconstruct the corresponding flow component structure model based on the multi-objective evaluation results corresponding to each candidate design parameter, simulate the operating conditions in a virtual simulation environment, and generate real perception feedback data as a basis for performance verification;

[0011] An adjustment module is configured to construct a design error function based on the real perception feedback data, reversely adjust the optimization strategy and dynamically update the design parameters by comparing the preset expected performance indicators with the multi-objective evaluation results corresponding to the current candidate design parameters, and drive the iterative convergence of the optimization process;

[0012] The scheduling control module is used to dynamically schedule local optimization or global search strategies according to the optimization process iteration stage and performance convergence status output by the adjustment module, and control the algorithm path selection and computing resource allocation during the optimization process.

[0013] Preferably, the multi-physics field modeling module includes:

[0014] A fluid mechanics modeling unit, used to construct a fluid flow model through mass conservation equations and momentum conservation equations based on preset boundary conditions and initial design parameters;

[0015] A structural mechanics modeling unit is used to construct a structural strength model of the flow-through component based on the fluid flow model and in combination with material property parameters, and to calculate stress and deformation fields;

[0016] The heat conduction modeling unit is used to construct a temperature distribution model of the flow-through component based on the stress and deformation field and the fluid flow model, and generate coupling simulation parameters.

[0017] Preferably, the intelligent optimization decision module includes:

[0018] An objective function definition unit, configured to extract flow performance indicators, structural response indicators, and thermal management indicators based on the coupled simulation parameters, and dynamically construct a multi-objective optimization objective function;

[0019] A dynamic strategy optimization unit, configured to initialize an optimization path based on the multi-objective optimization objective function and to adjust a candidate design parameter generation strategy in real time using a behavior-driven optimization strategy;

[0020] A deep reinforcement learning unit is used to use the candidate design parameter generation strategy as an initial state, gradually train the optimization strategy network through simulation feedback interaction with the collaborative computing module, and dynamically generate a new round of high-performance candidate design parameter sets.

[0021] Preferably, the collaborative computing module includes:

[0022] The multi-physics simulation unit is used to call the modeling process of the multi-physics modeling module to jointly simulate each candidate parameter based on the generated new round of high-performance candidate design parameter sets, and output a data set containing fluid, structural and thermal field results;

[0023] A performance evaluation unit, configured to construct a comprehensive performance scoring model based on the data set comprising fluid, structural, and thermal field results, and dynamically map the impact of design parameters on performance indicators;

[0024] The data integration unit is used to construct multi-objective evaluation results corresponding to candidate design parameters based on the influence relationship between the design parameters and the performance indicators, and dynamically adjust the weight factors of the evaluation indicators.

[0025] Preferably, the modeling process of calling the multi-physics field modeling module to perform joint simulation on each candidate parameter includes:

[0026] Building a multi-physics field joint simulation task list based on a new round of high-performance candidate design parameter sets generated by the deep reinforcement learning unit;

[0027] Dynamically configure the inlet velocity boundary conditions of the fluid mechanics simulation, the load distribution conditions of the structural mechanics simulation, and the heat source distribution conditions of the heat conduction simulation according to the design parameters in the simulation task list;

[0028] The fluid pressure field, structural stress field and temperature field data in the joint simulation results are synchronized to the performance evaluation unit in real time, triggering the comprehensive performance scoring model to update the weight factors and generate multi-objective evaluation results.

[0029] Preferably, the virtual experiment feedback module includes:

[0030] A model reconstruction unit, configured to dynamically generate a three-dimensional geometric model according to the multi-objective evaluation results corresponding to the candidate design parameters;

[0031] a dynamic simulation unit, configured to call a multi-physics field simulator to simulate the operating state of the flow component under typical operating conditions based on the three-dimensional geometric model, wherein the boundary conditions loaded by the multi-physics field simulator are dynamically configured by the joint simulation data output by the multi-physics field simulation unit;

[0032] The feedback generation unit is used to extract pressure distribution, temperature distribution and structural strain data from the simulation data configured by the dynamic simulation unit to construct a real perception feedback data set.

[0033] Preferably, calling a multi-physics field simulator to simulate the operating state of the flow-through component under typical working conditions includes:

[0034] Extracting candidate design parameters with optimal flow efficiency based on the multi-objective evaluation results and mapping them to a virtual simulation environment as typical operating condition parameters;

[0035] Based on the three-dimensional geometric model, calling a multi-physics field simulator to load the typical working condition parameters and perform a fluid-structure-thermal coupled transient simulation;

[0036] During the fluid-structure-thermal coupled transient simulation, the inlet pressure pulsation data of the flow components, the trailing edge temperature gradient data, and the blade strain peak data are collected in real time to construct a structured feedback data set.

[0037] Preferably, the adjustment module includes:

[0038] an error analysis unit, configured to compare and analyze the real perception feedback data generated by the virtual experiment feedback module with the flow performance optimization objective function defined in the intelligent optimization decision module, and construct a design error function;

[0039] A strategy updating unit, configured to adjust the optimization strategy network of the deep reinforcement learning unit using a back-propagation algorithm based on the error function result and generate new strategy parameters;

[0040] The parameter iteration unit is used to dynamically adjust the current candidate design parameter set based on the new strategy parameters, and drive the optimization process in the intelligent optimization decision module to iterate again to form a closed-loop optimization process.

[0041] Preferably, the scheduling control module includes:

[0042] A state monitoring unit is used to monitor the design parameter update trend and performance convergence speed of the parameter iteration unit in real time, and dynamically determine the current optimization stage state;

[0043] an algorithm scheduling unit, configured to dynamically select a global optimization algorithm to be used in the global search phase or a local optimization algorithm to be used in the local optimization phase according to the current optimization phase state;

[0044] A resource allocation unit is used to dynamically adjust the computing resource allocation strategy of each simulation task based on the local optimization algorithm, and the resource allocation strategy is fed back to the multi-physics field simulation unit to optimize the simulation execution efficiency.

[0045] A method for intelligent design and simulation of flow-through components of a fluid machinery is also provided, comprising the following steps:

[0046] Based on preset boundary conditions and physical properties, a multi-physics coupling model is constructed through the mass conservation equation, momentum conservation equation, and heat conduction equation to output coupling simulation parameters of fluid mechanics, structural mechanics, and heat conduction;

[0047] Based on the coupled simulation parameters of fluid mechanics, structural mechanics, and heat conduction, the objective function for optimizing flow performance is dynamically defined, an initial set of candidate design parameters is generated through a behavior-driven optimization strategy, and a deep reinforcement learning mechanism is triggered to iteratively adjust the parameter generation strategy.

[0048] Based on the initial candidate design parameter set, a multi-physics field numerical simulator is called to perform joint calculations to generate response data sets of fluid, structure, and thermal field, and a comprehensive performance evaluation model is constructed to output multi-objective evaluation results;

[0049] Based on the multi-objective evaluation results, a parametric modeling tool is driven to reconstruct a three-dimensional geometric model of the flow component, typical operating parameters are loaded in a virtual simulation environment to simulate dynamic operating behavior, and real perception feedback data including pressure, temperature, and strain distribution is generated;

[0050] Based on the difference analysis between the real perception feedback data including pressure, temperature and strain distribution and the preset performance indicators, a design error function is constructed, and the deep reinforcement learning strategy network is updated through the back-propagation algorithm to trigger a new round of optimization iterative convergence;

[0051] According to the new round of optimization iteration stage and performance convergence status, the combination strategy of global search algorithm and local optimization algorithm is dynamically scheduled to control the priority of computing resource allocation.

[0052] The present invention provides an intelligent design and simulation system for flow-through components of fluid machinery. It has the following beneficial effects:

[0053] 1. This invention combines multi-physics field coupling modeling with a dynamic feedback mechanism. This method continuously adjusts design parameters using real-time simulation data, enabling the optimization process to be dynamically updated based on actual operating conditions. This significantly improves optimization convergence speed and global performance. Compared to conventional static single-field optimization methods, this method addresses the problem of the optimization process being prone to falling into local optima and failing to accurately reflect actual operating conditions.

[0054] 2. The present invention introduces deep reinforcement learning and behavior-driven optimization strategies, adopts an automatically adjusted candidate design parameter generation mechanism, and achieves a high global convergence success rate through automatic exploration and real-time adjustment of the optimization path. Compared with the optimization strategies in the prior art that are easily limited by human experience, the strategies of the present invention are more dynamic and adaptable, significantly improving the design quality and efficiency.

[0055] 3. This invention utilizes a combination of multi-physics simulation and virtual experiment feedback to generate real-world sensory data in real time, establishing a novel feedback control mechanism that accurately analyzes errors during the design process and promptly adjusts optimization strategies. This achieves higher simulation accuracy and real-time performance, and solves the problem of inaccurate performance judgments caused by noise interference in simulation results, compared to conventional methods that struggle to efficiently process complex feedback data. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a system architecture diagram of the present invention;

[0057] Figure 2 The figure is a flow chart of the method steps of the present invention. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0059] Please see the attached Figure 1 The embodiment of the present invention provides an intelligent design and simulation system for flow-through components of fluid machinery, comprising:

[0060] Multi-physics modeling module, which is used to build fluid mechanics, structural mechanics, and heat conduction models based on preset boundary conditions and physical properties, and output coupled simulation parameters;

[0061] An intelligent optimization decision module is configured to receive the coupled simulation parameters, construct and define a flow performance optimization objective function, and generate an optimization path and a set of candidate design parameters through a behavior-driven optimization strategy and a deep reinforcement learning method;

[0062] A collaborative computing module is used to call a multi-physics field numerical simulator to jointly calculate the response results of each physical field based on the candidate design parameter set, build a comprehensive performance evaluation model, and generate a multi-objective evaluation result corresponding to each candidate design parameter;

[0063] A virtual experiment feedback module is used to reconstruct the corresponding flow component structure model based on the multi-objective evaluation results corresponding to each candidate design parameter, simulate the operating conditions in a virtual simulation environment, and generate real perception feedback data as a basis for performance verification;

[0064] An adjustment module is configured to construct a design error function based on the real perception feedback data, reversely adjust the optimization strategy and dynamically update the design parameters by comparing the preset expected performance indicators with the multi-objective evaluation results corresponding to the current candidate design parameters, and drive the iterative convergence of the optimization process;

[0065] The scheduling control module is used to dynamically schedule local optimization or global search strategies according to the optimization process iteration stage and performance convergence status output by the adjustment module, and control the algorithm path selection and computing resource allocation during the optimization process.

[0066] For the multi-physics field modeling module, in this embodiment, it is used to establish physical field models such as fluid mechanics, structural mechanics and heat conduction. In order to ensure the consistency of boundary conditions and model accuracy in intelligent design tasks, the multi-physics field modeling module must first complete the initial parameter setting and regional grid division, and then realize model construction and parameter output through the physical law solver.

[0067] Generally speaking, this module takes into account the multi-physics field coupling mechanism involved in the engineering objectives based on the preset design boundary conditions and initial design parameters, and dynamically generates a multi-source input parameter set while maintaining calculation stability, including but not limited to inlet flow velocity, pressure boundary, material property distribution, initial temperature field, etc.

[0068] Specifically, the multi-physics modeling module includes the following components:

[0069] In one possible implementation, the fluid mechanics modeling unit is used to establish a flow field model based on the mass conservation equation and the momentum conservation equation. Taking incompressible flow as an example, the mass conservation equation can be expressed as , where represents the fluid velocity vector. The momentum conservation equation is further expressed as:

[0070] ;

[0071] Where, is the fluid density; For pressure; is the fluid velocity vector; is the body force term; is the pressure gradient, which indicates the rate at which pressure changes with spatial position; is the Laplace operator of flow velocity, which represents the curvature or diffusion of the fluid velocity field; is the time derivative of flow velocity, which indicates the change of fluid velocity with time; is a nonlinear term that represents the convection effect of the fluid and the change in fluid velocity. The equations are discretely solved by a numerical solver to obtain the spatial distribution of the velocity field and pressure field.

[0072] As an option, the structural mechanics modeling unit is activated synchronously after the fluid modeling is completed, which is used to analyze the stress-deformation response behavior of the component. The stress-strain model follows linear elastic theory or nonlinear constitutive relations.

[0073] The heat conduction modeling unit is used to supplement the analysis of the coupled effects of the thermal field on the fluid and structural responses, and to solve the temperature field distribution based on the steady-state heat conduction or transient heat conduction equations. Under steady-state conditions, the control equation is:

[0074] ;

[0075] Where, is thermal conductivity; is the temperature field variable; is the volume heat source term; represents the temperature gradient; is the gradient operator.

[0076] In this example, a multi-field unified physics interface coordinates variable transfer, enabling modeling of coupled relationships between multiple physical quantities. For example, the temperature field affects the material properties of the structural mechanics module (e.g., the elastic modulus varies with temperature), while structural deformation affects the boundary conditions of the flow field (e.g., changes in the flow path).

[0077] In some embodiments, the model is solved by combining the finite element method (FEM) and the finite volume method (FVM), wherein FEM is used for stress analysis and heat conduction sub-models, and FVM is used for discretization of flow field variables.

[0078] Specifically, an automatic mesh refinement strategy can be introduced into the building block to enhance simulation accuracy in complex areas such as vortex regions and wall boundaries. Specific refinement criteria can be dynamically triggered based on the rate of change of field quantities such as velocity gradient and pressure gradient.

[0079] For 3D simulations, the module supports an adaptive time-step adjustment mechanism to ensure computational convergence and physical time consistency. This mechanism makes iterative adjustments based on local error estimates, effectively improving the stability and accuracy of multi-field simulations.

[0080] In this embodiment, the intelligent optimization and decision-making module receives coupled simulation parameters output by the multi-physics modeling module and implements intelligent screening and iterative optimization of candidate design parameters by constructing a multi-objective optimization function and a dynamic strategy generation mechanism. This module maps the physical field simulation results to the design space through a data-driven approach. Combining behavior-driven optimization strategies with deep reinforcement learning algorithms, it forms an adaptive decision-making framework for complex engineering problems.

[0081] The intelligent optimization decision module includes an objective function definition unit, a dynamic policy optimization unit, and a deep reinforcement learning unit. The objective function definition unit extracts key performance indicators based on coupled simulation parameters and constructs a multi-objective optimization function. The dynamic policy optimization unit generates an initial optimization path based on the objective function evaluation results. The deep reinforcement learning unit dynamically adjusts the parameter generation strategy by training the policy network through interaction with simulation feedback.

[0082] Specifically, the objective function definition unit defines the comprehensive performance index through the following formula:

[0083] ;

[0084] Where, is the design parameter vector (such as blade geometric parameters, material properties, etc.); is the flow efficiency; is the maximum equivalent stress; is the maximum temperature field value; is the weight coefficient, satisfying ; is the objective function. This formula normalizes aerodynamic efficiency, structural strength, and thermal management indicators to achieve multi-objective collaborative optimization.

[0085] In some embodiments, the weight coefficient can be dynamically adjusted according to the design stage. For example, in the aerodynamic performance priority stage, ; During the structural reliability verification phase, it is adjusted to The dynamic adjustment mechanism is implemented through an external configuration file or an interactive interface.

[0086] The dynamic policy optimization unit uses a behavior-driven optimization algorithm to generate an initial candidate parameter set. Its policy function can be expressed as:

[0087] ;

[0088] Where, represents the initial strategy selection function; represents the performance vector corresponding to the design parameters; is the target performance vector; it represents the The parameter vector that minimizes the objective function; For the The unit solves the above optimization problem by gradient descent method or genetic algorithm and outputs a set of candidate parameters that meet the constraints. .

[0089] The deep reinforcement learning unit builds a policy network based on the Actor-Critic framework. The Actor network generates parameter adjustment actions, and the Critic network evaluates the state value function. , where the status Contains current design parameters, objective function deviation, and historical optimization trajectory. Network updates follow the following policy gradient rules:

[0090] ;

[0091] Where, Represents the parameters of the policy network (Actor); represents the policy objective function; Indicates that based on the current policy parameters The obtained status Take action probability; represents the logarithmic probability of the strategy in the current action selection; Indicates that the policy function is Conditions for parameters The gradient (derivative) of represents the action-value function; Indicates that in the strategy The expectation under the current strategy (the expected value calculation is based on the trajectory samples generated by the current strategy). This unit obtains simulation feedback data through interaction with the collaborative computing module and uses the experience replay pool to store transfer samples. , among which rewards It is obtained by calculating the improvement rate of the objective function.

[0092] Alternatively, the deep reinforcement learning unit can gradually converge to the optimal policy in the early stages of training. The network structure can be a multi-layer perceptron (MLP) or a graph neural network (GNN), the latter of which is suitable for design parameter spaces with topological correlations.

[0093] Specifically, when the design parameters include the coordinates of parameterized control points of the blade geometry, the graph neural network encodes the spatial adjacency relationships between the control points into a graph structure, capturing the impact of local features on global performance through a message passing mechanism. This implementation improves the policy network's ability to generalize to complex geometric variations.

[0094] In one possible implementation, the module introduces an early stopping mechanism to prevent overfitting. When the average reward on the validation set does not improve for five consecutive epochs, training is terminated and the current optimal policy is saved. This mechanism is implemented using a monitoring window, the size of which can be adjusted based on the dataset size.

[0095] In this embodiment, the collaborative computing module is used to call a multi-physics numerical simulator to jointly calculate the response results of each physical field based on the candidate design parameter set generated by the intelligent optimization decision module, construct a comprehensive performance evaluation model, and output multi-objective evaluation results. Through multi-physics joint simulation and data integration, this module achieves dynamic mapping between design parameters and performance indicators, providing verification basis for the virtual experiment feedback module.

[0096] The collaborative computing module includes a multi-physics simulation unit, a performance evaluation unit, and a data integration unit. The multi-physics simulation unit generates joint simulation tasks based on a set of candidate design parameters, the performance evaluation unit constructs a scoring model based on the simulation results, and the data integration unit dynamically adjusts the output of multi-objective evaluation results through weighting factors.

[0097] Specifically, the multi-physics simulation unit uses fluid mechanics, structural mechanics, and heat conduction numerical simulators for joint calculations. Fluid mechanics simulation is based on the Reynolds-averaged Navier-Stokes equations, structural mechanics simulation uses the finite element method to solve the stress field, and heat conduction simulation calculates the temperature field distribution based on Fourier's law.

[0098] The performance evaluation unit constructs a comprehensive performance scoring model using the following formula:

[0099] ;

[0100] Where, For the Comprehensive score of group design parameters; For the Item performance indicator value; and is the normalized boundary of the indicator; is the dynamic weight coefficient. The weight coefficient is dynamically adjusted according to the requirements of the design stage; Indicates the total number of performance indicators.

[0101] The data integration unit maps the physical field response results to a unified design space. In one implementation, principal component analysis (PCA) is used to reduce dimensionality and extract key performance features to reduce data redundancy. The mathematical process can be expressed as:

[0102] ;

[0103] Where, is the original high-dimensional dataset; is the eigenvector matrix; is the design space coordinate after dimensionality reduction.

[0104] When a simulation diverges due to mesh distortion, it automatically flags abnormal parameter combinations and triggers local mesh refinement, dynamically adjusting the mesh size to ensure simulation stability. This mechanism, based on the residual convergence threshold and the order of precision of the discrete format, ensures reliable calculations of complex parameter combinations.

[0105] In this embodiment, the virtual experiment feedback module reconstructs the flow component structural model based on the multi-objective evaluation results output by the collaborative computing module. Simulating operating conditions in a virtual simulation environment generates real-world sensory feedback data as a basis for performance verification. This module uses dynamic modeling and multi-physics coupled simulation to virtually verify candidate design parameters, providing a foundation for error analysis in the adjustment module.

[0106] The virtual experiment feedback module includes a model reconstruction unit, a dynamic simulation unit, and a feedback generation unit. The model reconstruction unit drives a parametric modeling tool to generate a three-dimensional geometric model based on key geometric feature parameters from the multi-objective evaluation results. The dynamic simulation unit invokes a multi-physics simulator to load boundary conditions from the co-simulation data and perform dynamic response simulations under typical operating conditions. The feedback generation unit extracts pressure, temperature, and strain distribution data from the simulation results to construct a structured feedback dataset.

[0107] Generally speaking, the model reconstruction unit generates the geometric model through the following steps:

[0108] Analyze the blade profile control point coordinates, flow channel cross-sectional dimensions, and material distribution parameters in the candidate design parameters;

[0109] Construct the three-dimensional surface of the blade based on non-uniform rational B-spline (NURBS) curves;

[0110] Discretize the geometric model into a finite element mesh and associate material properties with boundary identifiers.

[0111] In one possible implementation, the dynamic simulation unit uses the following thermal-fluid-solid coupling control equation to describe the operating conditions:

[0112] ;

[0113] Where, is the fluid density; is the velocity field; For pressure; is the dynamic viscosity; is the coefficient of thermal expansion; is the temperature field; is the reference temperature; is the gravitational acceleration vector; is the Laplace operator of flow velocity; is the time derivative of the flow velocity. This equation couples the influence of the temperature field on the flow and reflects the thermal convection effect.

[0114] As an option, the structural response analysis uses the following constitutive relation to describe the stress-strain behavior:

[0115] ;

[0116] Where, is the stress tensor; is the elastic stiffness tensor; is the strain tensor; is the coefficient of thermal expansion; is the temperature change; is the Kronecker symbol. This equation quantifies the effect of temperature gradient on structural deformation.

[0117] Specifically, the dynamic simulation unit performs the following key operations:

[0118] Based on the multi-objective evaluation results, the candidate parameters with the best flow efficiency are selected and mapped to the inlet flow velocity, temperature boundary and structural constraints of the virtual environment;

[0119] Call the transient solver to perform fluid-structure-thermal coupled simulation, with the time step adaptively adjusted based on the Courant-Friedrichs-Lewy (CFL) condition;

[0120] Real-time monitoring of vortex shedding frequency and structural resonance risk triggers local grid encryption strategy to ensure computational stability.

[0121] The feedback generation unit constructs the verification basis through the following data processing process:

[0122] Extract the pressure pulsation spectrum characteristics of the fluid domain and calculate the main frequency energy ratio;

[0123] Quantify the maximum temperature gradient and spatial distribution uniformity in the trailing edge region;

[0124] Identify the strain concentration area on the blade surface and count the frequency of peak strain.

[0125] In this embodiment, the adjustment module constructs a design error function based on real-world sensory feedback data generated by the virtual experiment feedback module. By comparing preset performance indicators with the multi-objective evaluation results of the current candidate design parameters, it reversely adjusts the optimization strategy and dynamically updates the design parameters, driving the iterative convergence of the optimization process. This module, through a closed-loop mechanism of error analysis, strategy update, and parameter iteration, ensures dynamic optimization and performance convergence within the multi-physics coupled design space.

[0126] The adjustment module includes an error analysis unit, a strategy update unit, and a parameter iteration unit. The error analysis unit receives the pressure distribution, temperature gradient, and structural strain data output by the virtual experiment feedback module, compares and analyzes them with the flow performance optimization objective function defined in the intelligent optimization decision module, and constructs a design error function. Based on the error function results, the strategy update unit uses a backpropagation algorithm to adjust the parameters of the deep reinforcement learning strategy network and generate new optimization strategy instructions. The parameter iteration unit dynamically adjusts the candidate design parameter set based on the updated strategy parameters and triggers the intelligent optimization decision module to initiate a new round of optimization iterations.

[0127] Specifically, the error analysis unit quantifies the design error through the following formula:

[0128] ;

[0129] Where, Design parameters Error index; For the Simulation values ​​of performance indicators (such as flow efficiency, maximum stress value); is the preset target value; is the error weight coefficient; is the total number of error indicators. This formula eliminates dimensional differences through normalization and weightedly integrates multi-objective performance deviations.

[0130] The policy update unit uses the back-propagation algorithm to update the parameters of the deep reinforcement learning policy network. The network parameter update rule is:

[0131] ;

[0132] Where, For the Policy network parameters for iterations; is the learning rate; is the regularization coefficient. The second regularization term is used to prevent sudden changes in strategy parameters and improve optimization stability.

[0133] In one possible implementation, the parameter iteration unit introduces an adaptive step size mechanism. The parameter update step size is dynamically adjusted according to the magnitude of the error gradient: when the gradient is large, a smaller step size is used to avoid oscillation, and when the gradient is small, a larger step size is used to accelerate convergence. This mechanism is controlled by the gradient threshold and the step size coefficient to ensure a stable and efficient optimization process. Deep reinforcement learning unit

[0134] In this embodiment, the scheduling control module dynamically schedules global search or local optimization strategies based on the optimization iteration stage and performance convergence status, controlling computing resource allocation and algorithm path selection. This module, through real-time sensing of design parameter update trends and simulation task load characteristics, coordinates data exchange between the multi-physics simulation unit and the intelligent optimization decision module, ensuring efficient system convergence in complex parameter spaces.

[0135] The scheduling control module consists of a state monitoring unit, an algorithm scheduling unit, and a resource allocation unit. The state monitoring unit tracks the design parameter update sequence output by the parameter iteration unit in real time and dynamically determines the current optimization phase status by quantifying the convergence rate and performance fluctuation. The algorithm scheduling unit invokes a global search or local optimization algorithm based on the phase status. The resource allocation unit dynamically allocates computing resources based on task complexity and historical execution efficiency, forming a closed-loop control logic.

[0136] In some embodiments, the state monitoring unit calculates the design parameter convergence rate using the following formula:

[0137] ;

[0138] Where, For the The design parameter vector of the iteration; For the The design parameter vector of the iteration; is a very small positive number (usually ) to prevent the denominator from being zero. ( When it is a preset threshold, such as 0.01), the system is determined to have entered the local optimization stage and the algorithm switching mechanism is triggered.

[0139] Specifically, the resource allocation unit dynamically configures computing resources through the following functions:

[0140] ;

[0141] Where, For the The resource allocation ratio of each simulation subtask, is the task priority weight (calculated by the objective function deviation rate), is the historical average simulation time, is the nonlinear adjustment factor (usually 0.5-1.2), is the total number of current parallel tasks. Strengthen resource allocation to time-consuming tasks to alleviate simulation bottlenecks.

[0142] In one possible implementation, the algorithm scheduling unit adopts a two-tier decision-making mechanism: when the performance change rate (The calculation formula is the weighted average of the rate of change of each objective function value) When the threshold is exceeded, the global search algorithm is forced to reset the optimization path; when three consecutive iterations When the number of switches exceeds 0, the algorithm switches to a local optimization algorithm such as the quasi-Newton method. This mechanism is implemented through the status flag register to avoid oscillation caused by frequent switching.

[0143] As an option, the resource allocation unit supports elastic scaling. When a simulation subtask is detected to be taking longer than 150% of its estimated duration, it automatically triggers a compute node scaling command. This strategy is implemented through the Kubernetes container orchestration framework, ensuring load balancing in a distributed simulation environment.

[0144] The intelligent design and simulation method for a fluid machinery flow-through component described below and the intelligent design and simulation system for a fluid machinery flow-through component described above may refer to each other.

[0145] Please see the attached Figure 2 The present invention also provides a method for intelligent design and simulation of flow-through components of a fluid machinery, comprising the following steps:

[0146] Based on preset boundary conditions and physical properties, a multi-physics coupling model is constructed through the mass conservation equation, momentum conservation equation, and heat conduction equation to output coupling simulation parameters of fluid mechanics, structural mechanics, and heat conduction;

[0147] Based on coupled simulation parameters of fluid mechanics, structural mechanics, and heat conduction, the objective function for optimizing flow performance is dynamically defined. An initial set of candidate design parameters is generated through a behavior-driven optimization strategy, and a deep reinforcement learning mechanism is triggered to iteratively adjust the parameter generation strategy.

[0148] Based on the initial candidate design parameter set, a multi-physics numerical simulator is called to perform joint calculations to generate response data sets for fluid, structure, and thermal fields. A comprehensive performance evaluation model is then constructed to output multi-objective evaluation results.

[0149] Based on the multi-objective evaluation results, the parametric modeling tool is driven to reconstruct the 3D geometric model of the flow-through component. Typical operating parameters are loaded into the virtual simulation environment to simulate dynamic operating behavior and generate realistic sensory feedback data including pressure, temperature, and strain distribution.

[0150] Based on the difference analysis between the real sensory feedback data including pressure, temperature and strain distribution and the preset performance indicators, a design error function is constructed, and the deep reinforcement learning strategy network is updated through the back-propagation algorithm, triggering a new round of optimization iterative convergence;

[0151] According to the new round of optimization iteration stage and performance convergence status, the combination strategy of global search algorithm and local optimization algorithm is dynamically scheduled to control the priority of computing resource allocation.

[0152] The method of this embodiment can be used to execute the above system embodiment. Its principles and technical effects are similar and will not be described in detail here.

[0153] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent design and simulation system for flow-through components of fluid machinery, characterized in that: include: Multi-physics modeling module, which is used to build fluid mechanics, structural mechanics, and heat conduction models based on preset boundary conditions and physical properties, and output coupled simulation parameters; An intelligent optimization decision module is configured to receive the coupled simulation parameters, construct and define a flow performance optimization objective function, and generate an optimization path and a set of candidate design parameters through a behavior-driven optimization strategy and a deep reinforcement learning method; A collaborative computing module is used to call a multi-physics field numerical simulator to jointly calculate the response results of each physical field based on the candidate design parameter set, build a comprehensive performance evaluation model, and generate a multi-objective evaluation result corresponding to each candidate design parameter; A virtual experiment feedback module is used to reconstruct the corresponding flow component structure model based on the multi-objective evaluation results corresponding to each candidate design parameter, simulate the operating conditions in a virtual simulation environment, and generate real perception feedback data as a basis for performance verification; An adjustment module is configured to construct a design error function based on the real perception feedback data, reversely adjust the optimization strategy and dynamically update the design parameters by comparing the preset expected performance indicators with the multi-objective evaluation results corresponding to the current candidate design parameters, and drive the iterative convergence of the optimization process; A scheduling control module is used to dynamically schedule local optimization or global search strategies according to the optimization process iteration stage and performance convergence status output by the adjustment module, and control the algorithm path selection and computing resource allocation during the optimization process; The multi-physics modeling module includes: A fluid mechanics modeling unit, used to construct a fluid flow model through mass conservation equations and momentum conservation equations based on preset boundary conditions and initial design parameters; A structural mechanics modeling unit is used to construct a structural strength model of the flow-through component based on the fluid flow model and in combination with material property parameters, and to calculate stress and deformation fields; a heat conduction modeling unit, configured to construct a temperature distribution model of the flow-through component based on the stress and deformation field and the fluid flow model, and generate coupling simulation parameters; The collaborative computing module includes: The multi-physics simulation unit is used to call the modeling process of the multi-physics modeling module to jointly simulate each candidate parameter based on the generated new round of high-performance candidate design parameter sets, and output a data set containing fluid, structural and thermal field results; A performance evaluation unit, configured to construct a comprehensive performance scoring model based on the data set comprising fluid, structural, and thermal field results, and dynamically map the impact of design parameters on performance indicators; A data integration unit is used to construct multi-objective evaluation results corresponding to candidate design parameters based on the influence relationship between the design parameters and the performance indicators, and dynamically adjust the weight factors of the evaluation indicators; The modeling process of calling the multi-physics field modeling module to perform joint simulation on each candidate parameter includes: Build a multi-physics co-simulation task list based on a new round of high-performance candidate design parameter sets generated by the deep reinforcement learning unit; Dynamically configure the inlet velocity boundary conditions of the fluid mechanics simulation, the load distribution conditions of the structural mechanics simulation, and the heat source distribution conditions of the heat conduction simulation according to the design parameters in the simulation task list; Synchronize the fluid pressure field, structural stress field, and temperature field data in the joint simulation results to the performance evaluation unit in real time, trigger the comprehensive performance scoring model to update the weight factors, and generate multi-objective evaluation results; The virtual experiment feedback module includes: A model reconstruction unit, configured to dynamically generate a three-dimensional geometric model according to the multi-objective evaluation results corresponding to the candidate design parameters; a dynamic simulation unit, configured to call a multi-physics field simulator to simulate the operating state of the flow component under typical operating conditions based on the three-dimensional geometric model, wherein the boundary conditions loaded by the multi-physics field simulator are dynamically configured by the joint simulation data output by the multi-physics field simulation unit; a feedback generation unit, configured to extract pressure distribution, temperature distribution, and structural strain data from the simulation data configured by the dynamic simulation unit to construct a real perception feedback data set; The adjustment module includes: an error analysis unit, configured to compare and analyze the real perception feedback data generated by the virtual experiment feedback module with the flow performance optimization objective function defined in the intelligent optimization decision module, and construct a design error function; A strategy updating unit, configured to adjust the optimization strategy network of the deep reinforcement learning unit using a back-propagation algorithm based on the error function result and generate new strategy parameters; The parameter iteration unit is used to dynamically adjust the current candidate design parameter set based on the new strategy parameters, and drive the optimization process in the intelligent optimization decision module to iterate again to form a closed-loop optimization process.

2. The intelligent design and simulation system for flow-through components of fluid machinery according to claim 1, characterized in that: The intelligent optimization decision module includes: An objective function definition unit, configured to extract flow performance indicators, structural response indicators, and thermal management indicators based on the coupled simulation parameters, and dynamically construct a multi-objective optimization objective function; A dynamic strategy optimization unit, configured to initialize an optimization path based on the multi-objective optimization objective function and to adjust a candidate design parameter generation strategy in real time using a behavior-driven optimization strategy; A deep reinforcement learning unit is used to use the candidate design parameter generation strategy as an initial state, gradually train the optimization strategy network through simulation feedback interaction with the collaborative computing module, and dynamically generate a new round of high-performance candidate design parameter sets.

3. The intelligent design and simulation system for flow-through components of fluid machinery according to claim 1, characterized in that: The calling of the multi-physics field simulator to simulate the operating state of the flow-through component under typical working conditions includes: Extracting candidate design parameters with optimal flow efficiency based on the multi-objective evaluation results and mapping them to a virtual simulation environment as typical operating condition parameters; Based on the three-dimensional geometric model, calling a multi-physics field simulator to load the typical working condition parameters and perform a fluid-structure-thermal coupled transient simulation; During the fluid-structure-thermal coupled transient simulation, the inlet pressure pulsation data of the flow components, the trailing edge temperature gradient data, and the blade strain peak data are collected in real time to construct a structured feedback data set.

4. The intelligent design and simulation system for flow-through components of fluid machinery according to claim 1, characterized in that: The scheduling control module includes: A state monitoring unit is used to monitor the design parameter update trend and performance convergence speed of the parameter iteration unit in real time, and dynamically determine the current optimization stage state; an algorithm scheduling unit, configured to dynamically select a global optimization algorithm to be used in the global search phase or a local optimization algorithm to be used in the local optimization phase according to the current optimization phase state; A resource allocation unit is used to dynamically adjust the computing resource allocation strategy of each simulation task based on the local optimization algorithm, and the resource allocation strategy is fed back to the multi-physics field simulation unit to optimize the simulation execution efficiency.

5. A method for intelligent design and simulation of flow-through components of a fluid machinery, applied to an intelligent design and simulation system for flow-through components of a fluid machinery according to any one of claims 1 to 4, characterized in that: The following steps are involved: Based on preset boundary conditions and physical properties, a multi-physics coupling model is constructed through the mass conservation equation, momentum conservation equation, and heat conduction equation to output coupling simulation parameters of fluid mechanics, structural mechanics, and heat conduction; Based on the coupled simulation parameters of fluid mechanics, structural mechanics, and heat conduction, the objective function for optimizing flow performance is dynamically defined, an initial set of candidate design parameters is generated through a behavior-driven optimization strategy, and a deep reinforcement learning mechanism is triggered to iteratively adjust the parameter generation strategy. Based on the initial candidate design parameter set, a multi-physics field numerical simulator is called to perform joint calculations to generate response data sets of fluid, structure, and thermal field, and a comprehensive performance evaluation model is constructed to output multi-objective evaluation results; Based on the multi-objective evaluation results, a parametric modeling tool is driven to reconstruct a three-dimensional geometric model of the flow component, typical operating parameters are loaded in a virtual simulation environment to simulate dynamic operating behavior, and real perception feedback data including pressure, temperature, and strain distribution is generated; Based on the difference analysis between the real perception feedback data including pressure, temperature and strain distribution and the preset performance indicators, a design error function is constructed, and the deep reinforcement learning strategy network is updated through the back-propagation algorithm to trigger a new round of optimization iterative convergence; According to the new round of optimization iteration stage and performance convergence status, the combination strategy of global search algorithm and local optimization algorithm is dynamically scheduled to control the priority of computing resource allocation.

Images