Managing Big Data From Simulations
As simulation, monitoring, and testing techniques continue to advance toward Digital Engineering ecosystem, many have found themselves drowning in Big Data. As the computational burden continues to increase, analysis becomes intractable. This presentation will focus on two methods that help manage Big Data: the subsampling of a large data set and the sliced sampling of a large data set into smaller, more manageable data sets for training emulators. Subsampling algorithms intelligently considers points for the subset that maintain the characteristics of the whole data set, reducing the potential bias is the subsampled data set. Similarly, sliced sampling intelligently partition the data into slices which allows for training a series of emulators that can be combined into one. A conceptual Digital Twin example will be given to illustrate the capabilities of each method.
"Dr. Mark Andrews is SmartUQ’s UQ Technology Steward where he is responsible for advising SmartUQ on the industry’s uncertainty quantification needs and challenges. He is SmartUQ’s principal investigator for the Probabilistic Secondary Flow and Heat Transfer Model project as part of the Probabilistic Analysis Consortium for Engines (PACE). PACE is developed and managed by the Ohio Aerospace Institute (OAI) under contract with the Air Force Research Laboratory (ARFL) and engine OEMs to develop, apply, and validate advanced probabilistic methods that quantify uncertainty, achieving improvements in engine performance, cost, and reliability.
Prior to working at SmartUQ, Dr. Andrews spent 15 years at Caterpillar where he worked as a Senior Research Engineer, Engineering Specialist in Corporate Reliability, and Senior Engineering Specialist in Virtual Product Development. He is a member of the Probabilistic Methods, a subcommittee of Structures & Dynamics committee for ASME Turbo.