Photovoltaics Informatics: Harnessing Energy Science via Data-driven Approaches
- PDF / 20,350,932 Bytes
- 11 Pages / 612 x 792 pts (letter) Page_size
- 3 Downloads / 154 Views
Photovoltaics Informatics: Harnessing Energy Science via Data-driven Approaches Changwon Suh1, Kristin Munch1, David Biagioni2, Stephen Glynn1, John Scharf1, Miguel A. Contreras1, John D. Perkins1, Brent P. Nelson1, and Wesley B. Jones1 1 2
National Renewable Energy Laboratory, 1617 Cole Blvd., Golden, CO 80401, USA Department of Applied Mathematics, University of Colorado, Boulder, CO 80309, USA
ABSTRACT We discuss our current research focus on photovoltaic (PV) informatics, which is dedicated to functionality enhancement of solar materials through data management and data mining-aided, integrated computational materials engineering (ICME) for rapid screening and identification of multi-scale processing/structure/property/performance relationships. Our current PV informatics research ranges from transparent conducting oxides (TCO) to solar absorber materials. As a test bed, we report on examples of our current data management system for PV research and advanced data mining to improve the performance of solar cells such as CuInxGa1xSe2 (CIGS) aiming at low-cost and high-rate processes. For the PV data management, we show recent developments of a strategy for data modeling, collection and aggregation methods, and construction of data interfaces, which enable proper archiving and data handling for data mining. For scientific data mining, the value of high-dimensional visualizations and non-linear dimensionality reduction is demonstrated to quantitatively assess how process conditions or properties are interconnected in the context of the development of Al-doped ZnO (AZO) thin films as the TCO layers for CIGS devices. Such relationships between processing and property of TCOs lead to optimal process design toward enhanced performance of CIGS cells/devices. INTRODUCTION Terms such as data deluge, massive data, digital data, and multi-scale data integration are being referenced with increasing frequency in various news reports and articles (such as [1, 2]) relative to current progress in data-intensive science. All of these reflect the importance of scientific data itself as well as the manipulation of data in science and engineering. The issue is more critical in materials science and engineering because the number of possible materials that can be created is too numerous to follow a trial-and-error approach in the hope that the desired material is fortuitously developed. Here, the multi-scale data often range from scalar, categorical, and spectral to N-dimensional (N-D) hyper-spectral image data [3]. Therefore, while the objective of many design activities for functional materials focuses on finding the best combinations of processing routes to meet required properties, it is also highly desirable to develop computational frameworks including high-throughput (HT) data analysis using data mining for screening new candidate materials or predicting a material’s functionality. Such data-driven approaches in materials science and engineering (i.e., materials informatics) encompass a wide range of typical data-mining appli
Data Loading...
