Advanced data mining tool for feature detection in turbulent flow simulations

While DNS of complex flow phenomena is a routine practice nowadays, the analysis tools lack the sophistication to utilize the abundance of data generated, for they are devoid of physics-based data extraction and feature-detection protocols. The current effort proposes an automated and intelligent co-processing data-mining framework, in the form of an API, that detects and tracks flow features of significance in DNS calculations. The salient features of the proposed framework are a) wavelet based multi-resolution data compression algorithm for extracting regions of interest b) intelligent data-accumulation and monitoring methods for assessing specific terms in the governing equations of turbulent and aero-acoustic applications c) efficient feature-detection algorithms for coherent structures. The advantages of the proposed framework are a) co-processing of the data allows access to transient correlations that are unavailable for post-processing tools b) the physics-based choice of the mined-data addresses the deficiencies of the turbulent and aero-acoustic modeling c) the feature detection and data extraction are efficient due to scalable algorithms combining wavelet-filters and Fourier-transforms. The Phase I work demonstrates the applicability of this tool to representative flow fields in turbulence and aero-acoustic applications. Enhancing the framework’s features and algorithm fine-tuning will comprise the tasks for Phase II work. BENEFIT: The proposed technology will provide a fast and efficient means to mine data from unsteady DNS calculation. The generic design of the API will enable easy integration with a large range of commercial and research codes. The feature-detection algorithms are adaptable to any of the existing Visualization packages. A potential integration into CFDRC’s MDICE, which is part of many government and industry codes, will greatly help in eventual commercialization.