FY2010 - Oak Ridge National Laboratory
FY2010 - Oak Ridge National Laboratory
FY2010 - Oak Ridge National Laboratory
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Seed Money Fund—<br />
Computational Sciences and Engineering Division<br />
05888<br />
Development of Real-Time Optimization Methods for Neutron<br />
Scattering Experiments—Where to Measure and When to Stop<br />
Y. Jiao, K. An, P. F. Peterson, X.-L. Wang, and S. D. Miller<br />
Project Description<br />
In advanced neutron scattering, there are strong interests in the research community to study in situ<br />
dynamic or kinematic behaviors of materials under varying temperature, magnetism, pressure, and<br />
electric fields. The major challenge is that materials often react to changing conditions in heterogeneous<br />
and unpredictable ways: parts of the material may exhibit interesting behavior, while the rest does not;<br />
characteristics of a material may change dramatically within a very small range of temperatures, but do<br />
not vary much outside that range; as the pressure increases, molecular structural changes of a material<br />
may appear and then disappear; and the list goes on. We developed a new iterative, dynamic sampling<br />
approach that adjusts an experiment’s plan on the fly, based on the actual observations, to dynamically<br />
decide the area of interest (where to measure) and the minimum measurement time required to obtain<br />
sufficient data (when to stop). The novelty of our approach is threefold: (1) processing of streaming event<br />
data in real time, (2) developing robust algorithms that can handle various material samples and<br />
experimental settings, and (3) interfacing and controlling the instrument in real time. While experiment<br />
optimization on neutron instruments faces the same fundamental challenges, specific solutions are highly<br />
dependent on the instrument. For the proof of principle, we choose an engineering diffraction beam line,<br />
VULCAN at the Spallation Neutron Source, as our test bed with a real neutron diffraction experiment.<br />
Mission Relevance<br />
Neutrons are used by scientists to explore a wide spectrum of problems in the fields of physics, biology,<br />
chemistry, materials science, etc. Ideally, scientists want to be able to capture the intriguing moments<br />
when transformation occurs inside the material and to obtain sufficient amounts of data so that<br />
statistically significant conclusions can be drawn. Due to the exploratory nature of scientific experiments,<br />
scientists usually do not possess the precise knowledge of where or when these moments occur, nor do<br />
they know exactly the minimum measurement time required to obtain sufficient data prior to the<br />
experiment. As a result, precious beam time may be wasted and scientific discoveries may be delayed.<br />
One of the long-term goals of the neutron scattering sciences community is to be able to conduct<br />
experiments the way radiologists do magnetic resonance imaging oday: data are visualized and analyzed<br />
as the procedure is progressing, and decision support software, in real time, helps doctors to decide where<br />
they should investigate further. This project aims to take a step toward that goal. The success of this<br />
project can help accelerate scientific discoveries and can potentially save the research community<br />
millions of dollars.<br />
Results and Accomplishments<br />
We have successfully designed, implemented, and deployed software modules that provide real-time<br />
uncertainty analysis and optimization support to scientists. A Restful web service interface seamlessly<br />
connects the optimization components to the existing data acquisition system (DAS). The design of the<br />
software architecture follows three main principles: simple, modular, and extensible. First, we choose the<br />
RESTful web service paradigm instead of a SOAP-based web interface for its simplicity and stateless<br />
nature. This feature significantly reduces the amount of bookkeeping on the server side and allows for<br />
faster processing. Second, our fitting, visualization, and optimization tools are implemented as separate<br />
modules that can be mixed and matched to solve specific problems. Finally, we decoupled data analysis<br />
from data acquisition to allow both systems to expand independently. This work combines experiment<br />
198