Methods and systems for the estimation of the computational cost of simulation
Estimating the computational cost of simulation using a machine learning model. An example method includes inputting a feature data set into a machine learning model. The feature data set includes model geometry metadata and simulation metadata. The method further includes predicting, using the machine learning model, a computational cost characteristic for a simulation process.
1. A computer-implemented method for estimating the computational cost of simulation, comprising:
performing a plurality of simulations to obtain an initial simulation data set;
selecting one or more types of metadata from the initial simulation data set to create a model feature data set, the model feature data set comprising model geometry metadata and simulation metadata wherein:
the model geometry metadata is selected from a group consisting of a number of elements in a mesh, a number of vertices in the mesh, a surface-to-volume ratio of a model, a minimum element size, a mesh quality metric, an element type and element order, an aspect ratio of the model, a number of edges in the model, a number of surfaces in the model, a number of finite volumes in the model, and a number of control points in the model,
the simulation metadata is selected from a group consisting of an analysis type comprising (i) a static analysis, (ii) a quasistatic analysis, or (iii) a dynamic analysis, a solution domain, a linear solver type, a preconditioner, a time integration method, a spatial integration method, a degree of non-linearity, a non-linear solver option, a type and number of levels of adaptive mesh refinement, a time step size or number of frequencies, a boundary condition degrees-of-freedom, a boundary condition type, a material model type, a material property, a number of cores, and a physics-specific parameter, and
training and testing a machine learning model using the model feature data set;
inputting a feature data set into the machine learning model, the feature data set comprising the model geometry metadata and the simulation metadata, wherein inputting the feature data set into the machine learning model comprises (i) inputting the model geometry metadata into a first machine learning model and (ii) inputting the simulation metadata into a second machine learning model; and
predicting, using the machine learning model, a computational cost characteristic for a simulation process, wherein the computational cost characteristic is selected from a group consisting of a memory usage, a time for the simulation process, a time for meshing, a time for preprocessing, a time for solve, and a time for post-processing, and wherein the simulation process is a computer-aided engineering process for simulating a physical human-made object or human-made system represented by computer-aided design.
2. The computer-implemented method of claim 1 , wherein the computational cost characteristic is predicted before performing the simulation process.
3. The computer-implemented method of claim 1 , wherein the model is a computer-aided design (CAD) model.
4. The computer-implemented method of claim 3 , further comprising extracting the model geometry metadata from the CAD model.
5. The computer-implemented method of claim 1 , wherein the computational cost characteristic is at least one of the memory usage or the time for the simulation process.
6. The computer-implemented method of claim 5 , wherein the computational cost characteristic is a time for a portion of the simulation process.
7. The computer-implemented method of claim 5 , further comprising estimating a number of core hours based, at least in part, on the time for the simulation process.
8. The computer-implemented method of claim 1 , wherein predicting, using the machine learning model, a computational cost characteristic for a simulation process comprises a plurality of computational cost characteristics for the simulation process, each respective computational cost characteristic for the simulation process being based on a different set of compute resources and/or solver options.
9. The computer-implemented method of claim 1 , wherein the machine learning model is a deep neural network, a convolutional neural network, gradient boosted decision trees, or Gaussian process regression.
10. The computer-implemented method of claim 1 , wherein the feature data set further comprises computing environment metadata, wherein the computing environment metadata is selected from a group consisting of hardware information, a number of processors, and a bandwidth.
11. The computer-implemented method of claim 1 , further comprising determining the feature data set.
12. The computer-implemented method of claim 1 , further comprising:
performing one or more additional simulations;
supplementing the model feature data set with metadata from the one or more additional simulations; and
training and testing the machine learning model using the supplemented model feature data set.
13. The computer-implemented method of claim 1 , wherein the steps of training and testing the machine learning model are performed iteratively.
14. The computer-implemented method of claim 1 further comprising:
creating a fixed-dimensional representation vector from respective outputs of the first and second machine learning models; and
analyzing the fixed-dimensional representation vector to predict the computational cost characteristic for the simulation process.
15. A system for estimating the computational cost of simulation, comprising:
a computing device comprising a processor and a memory, the memory having computer-executable instructions stored thereon that, when executed by the processor, cause the processor to:
perform a plurality of simulations to obtain an initial simulation data set;
select one or more types of metadata from the initial simulation data set to create a model feature data set, the model feature data set comprising model geometry metadata and simulation metadata, wherein:
the model geometry metadata is selected from a group consisting of a number of elements in a mesh, a number of vertices in the mesh, a surface-to-volume ratio of a model, a minimum element size, a mesh quality metric, an element type and element order, an aspect ratio of the model, a number of edges in the model, a number of surfaces in the model, a number of finite volumes in the model, and a number of control points in the model,
the simulation metadata is selected from a group consisting of an analysis type comprising (i) a static analysis, (ii) a quasistatic analysis, or (iii) a dynamic analysis, a solution domain, a linear solver type, a preconditioner, a time integration method, a spatial integration method, a degree of non-linearity, a non-linear solver option, a type and number of levels of adaptive mesh refinement, a time step size or number of frequencies, a boundary condition degrees-of-freedom, a boundary condition type, a material model type, a material property, a number of cores, and a physics-specific parameter, and
train and test a machine learning model using the model feature data set;
input a feature data set into the machine learning model, the feature data set comprising the model geometry metadata and the simulation metadata, wherein inputting the feature data set into the machine learning model comprises (i) inputting the model geometry metadata into a first machine learning model and (ii) inputting the simulation metadata into a second machine learning model; and
predict, using the machine learning model, a computational cost characteristic for a simulation process, wherein the computational cost characteristic is selected from a group consisting of a memory usage, a time for the simulation process, a time for meshing, a time for preprocessing, a time for solve, and a time for post-processing, and wherein the simulation process is a computer-aided engineering process for simulating a physical human-made object or human-made system represented by computer-aided design.
16. The system of claim 15 , wherein the computational cost characteristic is predicted before performing the simulation process.
17. The system of claim 15 , wherein the model is a computer-aided design (CAD) model.
18. The system of claim 17 , wherein the memory has further computer-executable instructions stored thereon that, when executed by the processor, cause the processor to extract the model geometry metadata from the CAD model.
19. The system of claim 15 , wherein the computational cost characteristic is at least one of the memory usage or the time for the simulation process.
20. The system of claim 19 , wherein the computational cost characteristic is a time for a portion of the simulation process.
21. The system of claim 19 , wherein the memory has further computer-executable instructions stored thereon that, when executed by the processor, cause the processor to estimate a number of core hours based, at least in part, on the time for the simulation process.
22. The system of claim 15 , wherein predicting, using the machine learning model, a computational cost characteristic for a simulation process comprises a plurality of computational cost characteristics for the simulation process, each respective computational cost characteristic for the simulation process being based on a different set of compute resources and/or solver options.
23. The system of claim 15 , wherein the machine learning model is a deep neural network, a convolutional neural network, gradient boosted decision trees, or Gaussian process regression.
24. The system of claim 15 , wherein the feature data set further comprises computing environment metadata, wherein the computing environment metadata is selected from a group consisting of hardware information, a number of processors, and a bandwidth.
25. The system of claim 15 , wherein the memory has further computer-executable instructions stored thereon that, when executed by the processor, cause the processor to determine the feature data set.
26. The system of claim 15 , wherein the memory has further computer-executable instructions stored thereon that, when executed by the processor, cause the processor to:
perform one or more additional simulations;
supplement the model feature data set with metadata from the one or more additional simulations; and
train and test the machine learning model using the supplemented model feature data set.
27. The system of claim 15 , wherein the steps of training and testing the machine learning model are performed iteratively.
28. The system of claim 15 , wherein the memory has further computer-executable instructions stored thereon that, when executed by the processor, cause the processor to:
create a fixed-dimensional representation vector from respective outputs of the first and second machine learning models; and
analyze the fixed-dimensional representation vector to predict the computational cost characteristic for the simulation process.