For many applications in science and engineering a machine learning (ML) model must be able to predict a field of output values. For example, an ML model trained on CFD results may need to predict temperature or velocity as a function of location. An ML model trained on FEA results may need to predict stress as a function of location rather than for example a single scalar output such as maximum stress. Such spatially distributed problems can fall into two general categories:
Training samples all share a common grid. No matter how the input parameters are adjusted there are always the same number of outputs at the exact same spatial locations. SmartUQ currently has a Spatial/Temporal Emulator capable of handling such problems. This model further allows the spatially distributed outputs to be predicted as a function of time, for example temperature of a component as a function of location in 3D space and time
Different training samples may have a different grid both in terms of number and location of spatial points. Such a situation may occur for example if some of the problem’s input parameters alter the physical geometry and in turn the computational domain. SmartUQ has recently developed a module to handle such problems.
SmartUQ features a number of predictive models called emulators. Join us for this webinar to learn more about SmartUQ’s Spatial/Temporal Emulator and Varying Geometry module.
