A General-Purpose Software Tool for Multi-disciplinary Simulation Data Management and Learning

ABSTRACT: The overall goal of the proposed effort is to develop and demonstrate a general-purpose, fast, and reliable management and learning software tool for analyzing massive data sets generated by dynamic multi-disciplinary simulations. The salient elements of the software tool are: (1) an innovative combination of Proper Orthogonal Decomposition (POD) and advanced feature detection techniques to address the massive and multi-disciplinary nature of the data from a mathematically rigorous standpoint; (2) feature detection in reduced data sets for improving analysis speed and quality; and (3) a modular software framework to automate the entire data management and learning process and to enable seamless integration with multi-physics simulators. In Phase I, a feature specification module, a POD engine, and a feature detection module encapsulating advanced data mining algorithms, along with facile data exchange interfaces, will be developed in an integrated environment. Feasibility will be demonstrated by selected multi-disciplinary case studies of USAF interest, in which data generated by coupled multi-physics, multi-scale simulation will be analyzed for feature identification, selective visualization, and physics interpretation. The Phase II effort will focus on improving multi-disciplinary compatibility, developing in-line data analysis functionality, optimizing software architecture, integrating our software with USAF simulators for validation and technology demonstration. BENEFIT: The need for next-generation data management and learning tools for deciphering the massive data information generated by large-scale multi-disciplinary simulation is widely recognized. The end product arising from this effort will be a novel, general-purpose management and learning software tool for fast, reliable, automated analysis of massive multi-disciplinary simulation data. The proposed technology will be of direct commercial value in military (DoD, MDA, etc), NASA, and civilian sectors. The product will deliver DoD researchers a valuable tool to: (1) enable accurate feature detection and Region Of Interest (ROI) identification for selective visualization and storage; (2) gain an increased understanding and interpretation of the underlying physics; (3) provide guidance on diagnostics and reconfiguration of the multi-physics model and simulation for targeted applications; and (4) assist in concept evaluation, prediction, optimization, and design of high-performance systems. In the civilian sectors, the proposed data management tool will find widespread uses in various areas due to its capability of handling data from various sources, including multi-physics modeling and simulation, sensor data fusion and mining, experimental data reconstruction and system identification, as well as low-dimensional model extraction for real-time quality control and diagnostics, graphical and image analysis, weather forecasting, bioinformatics for medical diagnosis and therapeutics, etc.