Machine Learning for Changing Geometries

Schedule a SmartUQ Demo

Webinars

Machine Learning for Spatial and Temporal Response Prediction with Varying Geometry
This webinar will introduce SmartUQ’s Spatial/Temporal Emulator and Varying Geometry Module.

Watch Here

Machine Learning For Functional Response Prediction
This Webinar will introduce SmartUQ’s Functional Response Emulator.

Watch Here

White Papers

SmartUQ: Pushing the Boundaries of Engineering Analytics
As simulation, monitoring, and testing techniques become more sophisticated and produce increasing quantities of data, the need for a powerful engineering analytics tool to accurately and efficiently extract information and insights from the data has never been greater.

What is Spatial Emulation?

Spatial/temporal emulation leverages SmartUQ's functional emulation techniques to capture more functional dimensions. This allows users to build emulators that predict outputs dependent on up to 4-D functional inputs in addition to regular continuous inputs. These types of spatial or field responses are common in all variaties of engineering simulation and test. For example, CFD simulations often calculate pressure vs time and location or temperature vs position, FEA simulations generate stress/strain data on a mesh, and flow experiments produce velocity fields. By directly predicting the response profiles, users can explore design spaces without the need to create summary statistics, preserving more information for analysis.

The challenges of Varying Geometry Systems

Simulating domains with related but different geometries is a crucial piece of many workflows. They include geometry optimization, tolerance analysis, calibration to physical measurements, and uncertainty quantification. Depending on what users change between runs in a simulation, the coordinates may change. Anything that changes the simulation domain, e.g., changing the geometry contained within it, will almost certainly change the coordinates.

SmartUQ's Spatial-Temporal Emulator requires that all different sets of inputs utilize the exact same coordinates, i.e., the same number and the same location of spatial coordinates limitting its utility to situations where the topology is the same and coordinates can be transposed.

SmartUQ Solutions for Varying Geometry Emulation

SmartUQ's Varying Geometry Emulator extends the capabilities of SmartUQ's spatial emulator, allowing users to build and make predictions with kernel-based models using training data with changing geometries and varying coordinates. Users are no longer limited by the requirements to maintain mesh consistency, topology, or geometry. This advancement opens fast and accurate surrogate modeling for all kinds of simulations and test data sets.

Example

Changing Geometry, Mesh, and Topology
In this example, an emulator was built using a FEA simulation with variations in geometry and mesh dependent on the continuous inputs. This simulation also resulted in missing data at some mesh points.

Benefits of SmartUQ's Varying Geometry Emulation

SmartUQ's varying geometry emulation technology presents a novel ability to predict field responses on varying geometries opening opportunities in predictive modeling:

Predict response fields resulting from larger changes in geometry including topology variations.
Handle irregular meshes resulting from remeshing or adaptive meshing simulations
Build empirical models from irregular collections of points from testing or scans.
Fit models to data sets with missing or incomplete spatial information.
Avoids the need to interpolate, resample, or re-mesh data sets prior to model construction reducing preprocessing requirements.

To learn more about changing geometry applications, check out SmartUQ white papers and webinars.