A three-level partitioned parallel program architecture for three-dimensional numerical simulation of internal flow in a multi-stage impeller machine

Abstract

The application provides a three-level partition parallel program architecture for three-dimensional numerical simulation of internal flow of a multistage impeller machine, comprising: a fluid domain federation and a rotating-static interface domain federation as a first level; the fluid domain federation comprises fluid solving members as a plurality of second levels, and the fluid solving members correspond to a plurality of spatial partitions as a third level; the rotating-static interface domain federation comprises rotating-static interface members as a plurality of second levels, and the rotating-static interface members correspond to a plurality of spatial partitions as a third level; wherein the fluid domain federation and the rotating-static interface domain federation are solved in parallel, the fluid solving members are solved in parallel, the rotating-static interface members are solved in parallel, the spatial partitions of the fluid solving members are solved in parallel, and the spatial partitions of the rotating-static interface members are solved in parallel. The technical scheme of the application can reduce parallel communication amount to a certain extent, improve parallel efficiency, and reduce calculation time consumption for three-dimensional numerical simulation of internal flow of a multistage impeller machine.

Description

Technical Field

[0001] This application relates to the field of aero-engines, and in particular to a three-level partitioned parallel program architecture for three-dimensional numerical simulation of internal flow in a multi-stage impeller. Background Technology

[0002] Three-dimensional numerical simulation technology for internal flow in multi-stage turbomachinery. With the continuous development and maturation of computational fluid dynamics (CFD), the foundational discipline of three-dimensional numerical simulation technology for internal flow, the rapid improvement of numerical computing capabilities, and the rapid popularization of supercomputers with large-scale parallel computing capabilities in the field of aero-engine engineering design, three-dimensional numerical simulation technology for high Mach number and high Reynolds number flows has been deeply integrated into the aero-engine engineering design system. It is widely used in the aerodynamic design, strength verification, and sound field calculation of major flow passage components of aero-engines such as inlets, compressors, turbines, transition sections, and nozzles, as well as in the overall aerodynamic design, air system design, and whole-engine simulation of aero-engines; it runs through all major design stages, including scheme demonstration, scheme design, technical design, and engineering design.

[0003] In the field of aero-engine turbine engineering design, three-dimensional numerical simulation of internal flow not only needs to meet the accuracy requirements of engineering design, but also must keep the computation time within an acceptable range to satisfy the basic characteristics of rapid and large-scale iteration in the engineering design of flow passage components. Improving the accuracy of flow field numerical simulation can be achieved through theoretical breakthroughs to reduce the inherent biases of mathematical models; on the other hand, it can be achieved by increasing the mesh size to improve spatial resolution and reduce discretization bias. However, increasing the mesh size inevitably increases the solution time for the flow field. Without breakthroughs in fundamental theory, large-scale parallel computing technology based on supercomputing is a realistic and feasible technical choice that effectively controls / reduces computation time while increasing the computational scale.

[0004] A three-dimensional numerical simulation program for the internal flow of multi-stage turbomachinery, possessing large-scale parallel computing capabilities, can fully leverage the potential of supercomputing to achieve high-precision simulations of internal flow in aero-engines from both spatiotemporal dimensions. However, developing a large-scale parallel full three-dimensional internal flow numerical simulation program for solving multi-stage turbomachinery, with a high theoretical starting point and reliable stability, is a typical interdisciplinary project, integrating turbomachinery aerodynamic configuration, physics, mathematics, and computer science expertise, making it very challenging. The program architecture, based on the solution object, parallel strategy, and data management scheme, determines the computational efficiency, parallel performance, scalability, and maintainability of the developed program, and is therefore crucial. Thus, it is necessary to design a program architecture suitable for three-dimensional numerical simulation of internal flow in multi-stage turbomachinery. Summary of the Invention

[0005] To address at least one technical problem in the prior art, this application provides a three-level partitioned parallel program architecture for three-dimensional numerical simulation of internal flow in a multi-stage impeller.

[0006] This application provides a three-level partitioned parallel program architecture for three-dimensional numerical simulation of internal flow in a multi-stage turbomachinery, including:

[0007] As the first-level fluid domain federation and the static-transition interface domain federation;

[0008] The fluid domain federation includes multiple fluid solver members as second-level members, and the fluid solver members include multiple spatial partitions as third-level members.

[0009] The static interface domain federation includes multiple static interface members as second-level members, and each static interface member corresponds to multiple spatial partitions as third-level members.

[0010] Specifically, the fluid domain federation and the transition-static interface domain federation are solved in parallel, the fluid solver members are solved in parallel, the transition-static interface members are solved in parallel, the spatial partitioning of the fluid solver members is solved in parallel, and the spatial partitioning of the transition-static interface members is solved in parallel.

[0011] Optionally, the solution models for all fluid solution members are consistent, and the solution models for all transition-static interface members are consistent.

[0012] Or, at least some of the fluid solvers' solution models are inconsistent;

[0013] Alternatively, at least some of the solution models for the static interface members may be inconsistent.

[0014] Optionally, the three-level partitioned parallel program architecture is built on a parallel framework.

[0015] Optionally, the fluid solver uses the blade row as the unit and solves in parallel between units; the rotary-static interface solver uses the rotary-static interface as the unit and solves in parallel between units.

[0016] Optionally, the blade row includes a stator and a rotor.

[0017] Optionally, the spatial partitioning is implemented by a framework.

[0018] Optionally, the process allocation in the three-level partitioned parallel program architecture is as follows:

[0019] The processes are initially assigned according to member type, and each process is responsible for solving the flow field and the static-to-static interface.

[0020] The processes assigned to members of the same type are redistributed, and the number of processes is allocated to each member according to the proportion of control body units among members of the same type, so as to make the load of each process similar.

[0021] The processes assigned to members are then distributed to each spatial partition unit according to the spatial partition parallel strategy and the proportion of the number of spatial partition control units.

[0022] Optionally, the three-level partitioned parallel program architecture is used for parallel solution of three-dimensional numerical simulation of internal flow in multi-stage impellers.

[0023] The one or more technical solutions provided in the embodiments of this application are conducive to realizing a more flexible load distribution strategy between different types of solution domains and between different solution units of the same type of solution domain. This enables the three-dimensional numerical simulation of internal flow in multi-stage turbomachinery to reduce parallel communication, improve parallel efficiency, and reduce computation time to a certain extent. Attached Figure Description

[0024] The accompanying drawings illustrate exemplary embodiments of the present application and, together with the description thereof, serve to explain the principles of the present application. These drawings are included to provide a further understanding of the present application and are incorporated in and constitute a part of this specification.

[0025] Figure 1 A schematic diagram of a three-level partitioned parallel program architecture for three-dimensional numerical simulation of internal flow in a multi-stage turbomachinery according to an exemplary embodiment of this application is shown.

[0026] Figure 2 Another schematic diagram shows a three-level partitioned parallel program architecture for three-dimensional numerical simulation of internal flow in a multi-stage turbomachinery according to an exemplary embodiment of this application;

[0027] Figure 3 A call hierarchy diagram of a three-level partitioned parallel program architecture based on a parallel framework implemented according to an exemplary embodiment of this application is shown;

[0028] Figure 4 The meshing result of a fluid domain of a single-stage fan according to an exemplary embodiment of this application is shown;

[0029] Figure 5 The spatial partitioning result of a single-stage fan fluid domain according to an exemplary embodiment of this application is shown. Detailed Implementation

[0030] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0031] It should be understood that the steps described in the method embodiments of this application may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this application is not limited in this respect.

[0032] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc., mentioned in this application are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0033] It should be noted that the terms "a" and "a plurality of" used in this application are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0034] The names of the messages or information exchanged between multiple devices in the embodiments of this application are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0035] Characteristics of Three-Dimensional Numerical Solution for Internal Flow in Multi-Stage Turbomachinery. Compared to general flow passage components, the three-dimensional numerical solution for internal flow in multi-stage turbomachinery has distinct characteristics. The numerical solution techniques commonly used in engineering for internal flow in turbomachinery almost all simplify the flow field control equations and their solutions by leveraging the relatively simple geometric boundary motion in the impeller's body coordinate system. However, due to the relative rotation between adjacent impellers, relative rotation inevitably exists between adjacent flow field solution domains that rotate with the impellers, forming a non-physical rotation-stationary interface. This rotation-stationary interface fragments the internal flow solution of a multi-stage turbomachinery into a cluster of internal flow solutions based on impeller cascades. Adjacent impeller cascades are in an upstream-downstream relationship, with the rotation-stationary interface responsible for exchanging flow field information between their upstream and downstream impeller cascades. That is, based on the upstream and downstream flow field calculation results, the boundary conditions of the downstream and upstream flow fields at the rotation-stationary interface are given respectively, ultimately obtaining the numerical calculation results for the entire internal flow field of the multi-stage turbomachinery. In summary, solving the internal flow of a multi-stage turbomachine requires solving two types of governing equations: the flow field governing equations and the flow field information processing and exchange equations at the transition-to-stationary interface (the purpose of which is to provide boundary conditions at the boundary of the flow field at the transition-to-stationary interface; the flow field information processing and exchange equations at the transition-to-stationary interface can be various equations such as difference equations; in actual execution, the appropriate equations are selected and adjusted according to the requirements, making the parallel program architecture of this application more adaptable). The internal flow solution of a multi-stage turbomachine has relative independence based on the blade row as a unit.

[0036] To address the characteristics of three-dimensional numerical solutions for internal flow in multi-stage turbomachineries—namely, the relatively independent solutions to the fluid domain control equations and the rotor-stationary interface control equations (i.e., the fluid domain solution and the rotor-stationary interface domain solution are relatively independent, with the blade row as the solution unit and the rotor-stationary interface domain solution as the solution unit)—a three-level parallel partitioned three-dimensional numerical solution program architecture for internal flow in multi-stage turbomachineries is proposed. This architecture facilitates the construction, compilation, and maintenance of different types of control equation solution processes and mathematical models; it enables more flexible load distribution strategies between different types of solution domains and between different solution units within the same type of solution domain; and it restricts the parallel communication domain to members, thus reducing parallel communication volume, improving parallel efficiency, and reducing computation time to a certain extent.

[0037] For details, see Figure 1This application discloses a three-level partitioned parallel program architecture for three-dimensional numerical simulation of internal flow in a multi-stage turbomachinery, comprising: a fluid domain federation and a transfer-static interface domain federation as the first level; the fluid domain federation includes multiple fluid solver members as the second level, each fluid solver member corresponding to multiple spatial partitions as the third level; the transfer-static interface domain federation includes multiple transfer-static interface members as the second level, each transfer-static interface member including multiple spatial partitions as the third level; wherein the fluid domain federation and the transfer-static interface domain federation are solved in parallel, the fluid solver members are solved in parallel, the transfer-static interface members are solved in parallel, the spatial partitions of the fluid solver members are solved in parallel, and the spatial partitions of the transfer-static interface members are solved in parallel. It is understood that the parallel program architecture of this application can also be used on other platforms similar to multi-stage turbomachinery. Specifically, the three-level partitioned parallel program architecture is built based on a parallel framework. The three-level partitioned parallel program architecture can be built using different tools, such as MPI functions. More specifically, the three-level partitioned parallelism for three-dimensional numerical simulation of internal flow in a turbomachinery can be implemented based on the JASMIN parallel framework. JASMIN, short for Parallel Adaptive Structure Mesh Applications Infrastructure, addresses the need for parallel computation of coupled multi-physics models. It treats all physical models involved in the coupled computation as a "Federation," with each physical model considered a "Federal" within the federation. Each "Federal" can have its own distinct parallel solution model. The "Federals" possess relative independence in parallel solution, which corresponds precisely to the relative independence of flow solutions within a turbomachinery, where blade rows are the unit. Therefore, the "Federals" can be used to build the first and second levels of a three-level partitioned parallel program. The third level of spatial partitioning can be implemented through the parallel framework, but it should be emphasized that it can also be achieved through other mesh partitioning strategies.

[0038] In this step, the three-dimensional numerical simulation of the internal flow of a multi-stage impeller is performed in parallel across three levels of partitions, such as... Figure 2 As shown, in the first stage, the fluid domain and the rotor-static interface domain are solved in parallel; in the second stage, within the fluid domain, the blade row is used as the solution unit, as shown... Figure 2 The aforementioned guide vane isostatic inertial ... Figure 2The transition-to-static interface 1 and transition-to-static interface 2 are shown as solution units, and parallel solutions are implemented between these interfaces. At the third level, within the fluid domain solution unit and the transition-to-static interface domain solution unit, spatial partitioning and parallel solutions are adopted. The solution modules for the fluid unit include: a geometric parameter solution module, a boundary condition solution module, a convection flow solution module, a viscous flux solution module, and a linear equation solution module. The solution modules for the interface unit include: a mixing surface solution module, an incoordination interpolation module, a low-pass / high-pass filtering module, a buffer layer solution module, and a phase delay solution module. That is, the fluid unit solution process follows the corresponding solution processes for the geometric parameter solution module, boundary condition solution module, convection flow solution module, viscous flux solution module, and linear equation solution module; the transition-to-static interface solution process follows the corresponding solution processes for the mixing surface solution module, incoordination interpolation module, low-pass / high-pass filtering module, buffer layer solution module, and phase delay solution module.

[0039] Specifically, firstly, such as Figure 2 As shown, the flow field of a multi-stage turbomachinery is decomposed sequentially into solution units for the fluid domain and solution units for the transition-stationary interface domain. Each solution unit is considered as a member (e.g., member 1, member 2, member 4, member 5). Then, the members are divided into two categories: flow field solution members and transition-stationary interface solution members. All flow field solution members have the same solution model, and all transition-stationary interface solution members have the same solution model; alternatively, at least some of the solution models for the fluid solution members are different, or at least some of the solution models for the transition-stationary interface members are different. Finally, parallel solution processes are assigned to each level of parallel unit according to a process allocation strategy. In this embodiment, the solution models for the transition-stationary interface solution members can be different or the same, providing flexibility in model options. In this application, the solution models for the flow field solution members and the transition-stationary interface solution members can be different, making the program easier to maintain.

[0040] The fluid solver uses the blade array as its element and solves in parallel between elements. It can be known that the element here is the blade array of the corresponding fluid solver. The transition-static interface solver uses the transition-static interface as its element and solves in parallel between elements. It can be known that the element here is the element of the transition-static interface solver.

[0041] Figure 3The diagram shows the call hierarchy of a three-level partitioned parallel program architecture for a three-dimensional numerical simulation of internal flow in a multi-stage turbomachinery, implemented based on a parallel framework. The call process is as follows: prepare the database, restart and set the timer, create the coordinate system and mesh, create the Javavis visualization output class, and create the time integration algorithm and mesh patch strategy accumulation for the federation and members (members) of the mesh layer. Federation Layer: Initializes federation and member components and data slices for specified mesh layers, establishing the main computation flow -> calling the computation flow of functional modules; Functional Module Level Layer: Establishes the computation flow of functional modules such as geometric parameter solving module, boundary condition solving module, flow solving module, viscous flux solving module, and linear equation solving module -> calls the components of the patch layer; Functional Module Patch Layer: Completes the component implementation of functional modules such as geometric parameter solving module, boundary condition solving module, flow solving module, viscous flux solving module, and linear equation solving module -> calls the numerical kernel; Numerical Kernel Layer: Prepares the standard numerical kernels for functional modules such as geometric parameter solving module, boundary condition solving module, flow solving module, viscous flux solving module, and linear equation solving module.

[0042] Figure 4 The image shows the grid partitioning result of a single-stage fan fluid domain network. Figure 5 The image shows the spatial partitioning results of a single-stage fan fluid domain network.

[0043] In this embodiment of the application, the process allocation strategy can be flexibly given according to the specific problem. Taking 24 processes as an example, as shown in Table 1, that is:

[0044] Table 1. Illustration of the three-level partitioning parallel strategy

[0045]

[0046]

[0047] (1) The process is initially allocated according to the member type. Generally speaking, a simpler load allocation strategy is that each process undertakes a certain task of solving the flow field and the static-to-static interface. This allocation strategy is given by the program architecture user according to the specific problem to be solved.

[0048] (2) The processes assigned to members of the same type are redistributed, generally based on the solution scale, such as the proportion of control body units among members of the same type, to allocate the number of processes to each member, so that the workload of each process is similar. This also means that the process groups solving each member are distinct. This level of allocation strategy is given by the program architecture user according to the specific problem being solved. Here, the number of control body units among members of the same type refers to the total number of units contained in a member, such as... Figure 4The number of the smallest unit (grid). (3) The process assigned to a member is then allocated to each spatial partition unit (patch) according to the spatial partition parallel strategy. Generally, it is also based on the solution scale, that is, the proportion of the number of spatial partition control units. At this time, multiple spatial partition units may be assigned to be solved by one process. This level of allocation strategy is completed by the parallel components provided by the parallel framework.

[0049] The program architecture of this application is conducive to the construction, compilation, and later maintenance of solution processes and mathematical models for different types of control equations; it can realize a more flexible load distribution strategy between different types of solution domains and between different solution units of the same type of solution domain; it can reduce parallel communication volume, improve parallel efficiency, and reduce computation time to a certain extent.

Claims

1. A design method for a three-level partitioned parallel program architecture for three-dimensional numerical simulation of internal flow in a multi-stage impeller, characterized in that, include: As the first-level fluid domain federation and the static-transition interface domain federation; The fluid domain federation includes multiple fluid solver members as second-level members, and the fluid solver members include multiple spatial partitions as third-level members. The static interface domain federation includes multiple static interface members as second-level members, and each static interface member corresponds to multiple spatial partitions as third-level members. The fluid domain federation and the static-transition interface domain federation are solved in parallel, the fluid solver members are solved in parallel, the static-transition interface members are solved in parallel, the spatial partitioning of the fluid solver members is solved in parallel, and the spatial partitioning of the static-transition interface members is solved in parallel. The fluid solver uses the blade row as the unit and solves in parallel between units; the rotary-static interface solver uses the rotary-static interface as the unit and solves in parallel between units. The blade row includes a stator and a rotor; The process allocation of the three-level partitioned parallel program architecture is as follows: The processes are initially assigned according to member type, and each process is responsible for solving the flow field and the static-to-static interface. The processes assigned to members of the same type are redistributed, and the number of processes is allocated to each member according to the proportion of control body units among members of the same type, so as to make the load of each process similar. The processes assigned to members are then distributed to each spatial partition unit according to the spatial partition parallel strategy and the proportion of the number of spatial partition control units.

2. The design method of the three-level partitioned parallel program architecture for three-dimensional numerical simulation of internal flow in a multi-stage impeller as described in claim 1, characterized in that, The solution models for all fluid solution members are consistent, and the solution models for all static-transition interface members are consistent. or, The solution models of at least some of the fluid solvers are inconsistent; Alternatively, at least some of the solution models for the static interface members may be inconsistent.

3. The design method of the three-level partitioned parallel program architecture for three-dimensional numerical simulation of internal flow in a multi-stage impeller as described in claim 1, characterized in that, The three-level partitioned parallel program architecture is built on a parallel framework.

4. The design method of the three-level partitioned parallel program architecture for three-dimensional numerical simulation of internal flow in a multi-stage impeller as described in claim 1, characterized in that, The spatial partitioning is implemented using a framework.

5. The design method of the three-level partitioned parallel program architecture for three-dimensional numerical simulation of internal flow in a multi-stage impeller as described in claim 1, characterized in that, The three-level partitioned parallel program architecture was used for parallel solution of three-dimensional numerical simulation of internal flow in multi-stage impellers.

Images