HUNTSVILLE, Ala. – February 15, 2014 – CFD Research Corporation and the Information Technology and Systems Center (ITSC), at the University of Alabama in Huntsville have been awarded a $750k follow-on contract to develop and demonstrate a general-purpose, fast, and reliable management and learning software tool for analyzing the massive data sets generated by dynamic multi-disciplinary simulations “The key innovations of the project include a hybrid methodology combining advanced data reduction, data mining, and modeling techniques to tackle the ‘big data’ challenge from a mathematically rigorous perspective,” said Dr. Yi Wang, CFDRC Director and Principal Investigator. The result is an integrated, modular, parallelizable software environment to facilitate software autonomy and deployability on various computational platforms (PCs, small-scale clusters, and HPC facilities).
In prior work, key technology elements were developed and proof-of-principle was successfully demonstrated. Data management software encapsulating feature specification algorithms, dimension-reduction engine, feature detection module, and data utilization module were developed in an integrated architecture. USAF relevant case studies were performed to establish the proof-of-principle of the CFDRC tool to accurately capture coherent flow structures with unprecedented data reduction (up to 40X) and focused visualization and learning with <10% of original data.
The figures show the fully automated identification of flow regions of interest (ROI) in a 3D delta wing and the captured events of vortex generation, roll-up, and burst. CFDRC’s data mining technique yields ~25X data reduction for targeted visualization and analysis (total data size is 240 GB). A reduced order model built on system identification technique was developed to predict the flow field and leads to 18,000X speedup over traditional brute force computation.
During the follow-on effort, the developed software will be optimized for performance and functionality. The feature library will be expanded and improved. A more computationally efficient technique will be developed to tackle the massive data sets using limited computational resources. Advanced feature detection algorithms will be developed to enhance accuracy and speed. All of the modules will be integrated onto a high-performance, parallel computing platform. The software will be extensively validated and demonstrated by selected case studies of USAF interest, and the validated software will be delivered to USAF and beta tested at AFRL facilities.
CFD Research Corporation
Manager, Technology Partnerships
256.726.4810 // firstname.lastname@example.org