Data Mining for Scientific and Engineering Applications

Due to advances in information technology and high performance computing are that are employed in the scientifc and engineering simulation, very large data sets are increasingly becoming available in many computational mechanics disciplines. The rate of production of such data sets far outstrips the ability to analyze them manually. More specifically, there is often information hidden in the data that cannot be discovered manually. Such information is mostly disguised as important associations and associations that are not readily evident from the governing model of physical phenomena. A deeper understanding of these underlying associations may lead to further improvements in the modeling and analysis of engineering problems. Thus, there is an increasing interest in various scientific communities in exploring the use of data mining techniques to analyze the large data sets emanating from scientific simulation code running on High Performance Computing platforms. Given the success of data mining in non-scientific areas, such as knowledge discovery in web data and prediction powers in business and stock market data it seems likely that data mining techniques can be used to develop new and sophisticated analysis tools which will automatically analyze scientific data and allow engineers and scientists to gain fundamental insights into the underlying mechanisms of the physical processes involved.
Interactive, virtual prototyping and testing has the potential of significantly reducing the cost and time required to the meet budgetary and time constraints of short design cycles. To enable such an environment, the design process needs to be equipped with a new generation of design tools for rapid modeling, fast analysis, and 3D physics-based visualization, prototyping and testing. Data mining techniques can play an equally important role in this design process for a variety of science and engineering problems such as structural, thermal and flow design problems. As the complexity of the simulated phenomena increases, both the parameter space on which the underlying model is tested, as well as the size of the resulting data becomes excessively large. This makes it difficult and cumbersome to manually analyze and correlate the data generated by the large number of simulations. Data mining techniques are viable tools for determining interesting patterns, clustering the parameter space, detecting anomalies in the simulation results, and for designing improved physical models.

Related Selected Publications

Data mining for Damage Detection in Engineering Structures, Encyclopedia of Data Warehousing and Mining, Ed. J. Wang, Idea Group Publishing, 2005 (with A. Lazarevic)

Damage Detection Employing Novel Data Mining Techniques, in New Generation of Data Mining Applications, Eds. M. Kantardzic and J. Zurada, IEEE 2005 (A. Lazarevic, C. Kammath and V. Kumar).

Predicting Very Large Number of Target Variables Using Hierarchical and Localized Clustering, Journal, (in review), (with A. Lazarevic, C. Kamath and V. Kumar).

Effective Localized Regression for Damage Detection in Large Complex Mechanical Structures, 10th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining Seattle, WA, August 22--25, 2004, (with A. Lazarevic, C. Kamath and V. Kumar).

Localized Prediction of Continuous Target Variables Using Hierarchical Clustering, 3rd IEEE International Conference on Data Mining, Melbourne, Florida, November 19--22, 2003. (with A. Lazarevic, C. Kamath and V. Kumar).

Damage Prediction in Structural Mechanics Using Hierachical Localized Clustering-based Approach, Data Mining and Knowledge Discovery: Theory, Tools, and Technology V, Orlando, Florida, April 21-25, 2003. (with A. Lazarevic, C. Kamath and V. Kumar).

A Dimension Reduction Data Mining Model for Damage Detection in Large Scale Complex Structures: Sub-structure approach, Mathematical Challenges in Scientific Data Mining, IPAM, January 14-18, 2002, (with S. S. Sandhu, C. Kamath and V. Kumar).

A Sub-Structuring approach Via data Mining for damage Prediction and Estimation in Complex Structures, SIAM International Conference on Data Mining, Arlington, VA, April 11--13, 2002. (with S. S. Sandhu, C. Kamath and V. Kumar).

Determination of an Initial Mesh Density for Finite Element Computations via Data Mining, The 7th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining/4th Workshop on Mining Scientific Datasets San Francisco, California, USA, August 26 - 29, 2001, (with C. Kamath and V. Kumar).

Damage Prediction and Estimation in Structural Mechanics Based on Data Mining, The 7th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining/4th Workshop on Mining Scientific Datasets San Francisco, California, USA, August 26--29, 2001, (with S. S. Sandhu, C. Kamath and V. Kumar).