Neural Network Combustion Simulation Mesh Coupling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current simulation methods for combustion chambers in thermodynamic rotary machines, such as turboshaft engines, face significant computation time challenges due to the need for interpolation between computational fluid dynamics (CFD) solvers and neural networks, leading to inaccuracies and inefficiencies in determining optimal design parameters.
Innovation Solution
A method involving a multilayer neural network trained on a given mesh with local combustion quantities, allowing direct input and output with a solver, and using a coarser mesh for large-scale simulations to reduce computational complexity and improve accuracy, while maintaining the accuracy of computational fluid dynamics techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If interpolation steps are used between solver and neural network, then compatibility between different mesh types is achieved, but computation time increases significantly and processing efficiency decreases
Solution Approach 1:
The patent extracts and removes the interpolation steps from the simulation workflow. By training the neural network directly on solver data without interpolation, the method eliminates these unnecessary processing steps while maintaining compatibility between the solver and neural network through direct data exchange on the same mesh structure
Solution Approach 2:
The patent segments the simulation process into distinct phases: solver execution on a given mesh, direct neural network prediction on the same mesh, and iterative refinement. This segmentation allows each component to operate independently on its optimal mesh without requiring interpolation between different mesh representations
2Ease of operation
If interpolation and re-interpolation steps are performed at each iteration, then data exchange between solver and neural network is enabled, but digital processing cost increases and a large part of computation time is absorbed
Solution Approach 1:
The patent removes the interpolation operations from the iterative data exchange process. The neural network is trained to operate directly on solver output data using the same mesh, eliminating the need for interpolation and re-interpolation steps while maintaining full data exchange capability between components
Solution Approach 2:
The patent introduces a standardized data interface that directly connects the solver and neural network without requiring interpolation as an intermediary step. This interface allows efficient data exchange by maintaining consistent mesh representations throughout the simulation process
3Device complexity
If a fixed mesh is used for neural network operations, then network computation simplicity is maintained, but accuracy decreases when working with adaptive discretization from solvers
Solution Approach 1:
The patent makes the given mesh universal by using it for both solver computation and neural network operations. This single mesh structure serves multiple functions: it provides adaptive discretization for the solver while simultaneously serving as the operational mesh for the neural network, eliminating the need for separate fixed and adaptive meshes
Solution Approach 2:
The patent changes the operational parameters of the neural network to accept and process data directly in the solver's mesh format. By modifying the network's input expectations and training data format to match the adaptive mesh structure, the system maintains both simplicity and accuracy without requiring mesh conversion
Data Source
AI summary
A method for simulating the combustion of a fluid in a combustion chamber, for the design of said combustion chamber, which includes: discretizing the space of the chamber into a given mesh; training a neural network by means of a learning set associating a graph corresponding to the mesh, the vertices of which have, as a value, progress variables predicted by a computational fluid dynamics simulation with local combustion quantities at these vertices; and an iterative simulation phase, where: the values predicted by the neural network of a local combustion quantity at the vertices of the mesh are provided as input to a solver, in order to obtain a value of a progress variable at each vertex of said mesh, and a graph corresponding to the vertices of the mesh is provided to said neural network, each vertex having a corresponding value of said progress variable, obtained by said solver, in order to obtain predicted values of the local combustion quantity at said vertices.