The thrust of nuclear physics is studying the universe down to its smallest subatomic parts. Now, two physicists at the Department of Energy’s Thomas Jefferson National Accelerator Facility have secured more than $2 million in federal funding dedicated to research projects that harness the power of data analytics to make that work faster and more efficient.
David Lawrence and Chris Tennant are among 14 scientists at seven DOE national laboratories whose proposals were awarded a total of $37 million to be allocated over three years.
“Artificial Intelligence and machine learning have the potential to transform a host of scientific disciplines and to revolutionize experimentation and operations at user facilities in the coming years,” Chris Fall, director of DOE’s Office of Science, said in announcing the funding. “These awards will help ensure America remains on the cutting edge of these critical technologies for science.”
Industry has extensively used AI and machine learning to revolutionize daily life: from building the ability to instantly pull up web pages on request to identifying the sound and meaning of a speaker’s voice, from recognizing faces and objects to translating languages.
The world of physics is also using the power that these algorithms offer. Lawrence and Tennant, for instance, will be working on ways to use machine learning and data analytics to help optimize Jefferson Lab’s research program. Their projects aim to improve operations of Jefferson Lab’s Continuous Electron Beam Accelerator Facility, a DOE user facility. These projects will improve the CEBAF accelerator’s efficiency by speeding up equipment calibrations and by streamlining the final physics analysis phase of research. These improvements may allow nuclear and accelerator physicists to shave off months of research labor from start to finish.
Both AI and machine learning are driven by the ability to efficiently and automatically analyze data. Tennant, a Jefferson Lab staff scientist in the Center for Advanced Studies of Accelerators, said he began the transition to machine learning a few years ago when he recognized its potential to extract meaningful information from the data-rich field of accelerator physics.
“Machine learning can keep us from getting into a situation where we are data rich, but information poor,” Tennant said.
Typically, physicists have taken on the computer programming necessary for extensive data analysis by default.
“Certainly in our field, most all of the software that’s written to do this work has been done by physicists,” said Lawrence, a Jefferson Lab staff scientist and Experimental Physics Scientific Computing Infrastructure group lead. “Most of them do not have computer science degrees. I don’t even have a computer science degree, even though I’ve done so much programming I’ve really made my career off of software and computing.”
His project, “AI Assisted Experiment Control and Calibration,” was funded for $810,000 — largely to hire a physicist and a computer scientist who will collaborate closely for three years to enhance the use of data analytics in research.
“We really want to try to get some resonance going between computer science and physics,” Lawrence said. “More so than we’ve done in nuclear physics before.”
The project’s goal is to use data analytics to calibrate detectors in the lab’s experimental halls, enabling scientists to adjust a whole host of factors in real time to enhance performance and generate better-quality data. This would save months of arduous post-experiment calibration work.
“The bottom line is, we’re hoping to get more science for the buck when we take the data,” Lawrence said. “The DOE funding really is the difference from being able to do this kind of thing or not.”
“The lab has a yearly budget and we squeeze everything we can out of that,” he added. “But sometimes, when you want to do new, really big things, we go for these programs.
“To me, this could be one of those things that we push on with this project and, once we learn how to do it, then it becomes standard. And then it will be absorbed as regular operations.”
Tennant’s proposal, “AI for Improved SRF Operation at CEBAF,” was awarded $1.35 million. The foundation for his project was laid by work on a project led by Anna Shabalina, a Jefferson Lab staff member and principal investigator on a proposal funded by Jefferson Lab’s Laboratory Directed Research and Development program. Begun in 2013, the LDRD program encourages and provides opportunities for staff to make rapid and significant contributions to critical national science and technology problems.
Tennant’s project consists of three parallel paths. A majority of machine learning models are implemented on static, off-line data. But accelerator facilities are continuously streaming many thousands of signals. One part of Tennant’s project will develop machine learning and data-analytic models that work with continuous streaming data to predict –in real-time – if the accelerating cavities that propel the electron beams in CEBAF are about to fail.
“That’s really key,” said Tennant. “Because when you have a cavity fault, it means the beam gets turned off. If the beam gets turned off, it means that experiments or users aren’t getting the data that they need. That’s highly disruptive.”
A second project will expand the capabilities of a data acquisition system to collect information-rich data from accelerating cavities. The data will leverage machine learning’s ability for pattern recognition to detect when a cavity becomes unstable. Currently, this is a laborious and time-consuming task for machine operators.
Finally, he plans to use machine learning to optimize the distribution of accelerating cavity gradients in CEBAF’s two linear accelerators to minimize the impact of field emission, which can cause damage to the accelerator cavities and other components. Doing so would have an immediate benefit for reaching higher beam energy, while reducing damage to beamline components.
Tennant said he was “absolutely thrilled” to receive the DOE funding and is happy to have subproject co-leads Riad Suleiman, Anna Shabalina and Dennis Turner working with him on the project.
“I think we’re just at the tip of the iceberg on what machine learning can do in the field of accelerator physics,” Tennant said. “I’m really excited to be a part of that moving forward.”
Tennant is from central New York. He did his doctoral research at Jefferson Lab from 2001-2006 and stayed on as a staff scientist.
Originally from Oklahoma, Lawrence came to Jefferson Lab in 1998. Co-leads on his project are Thomas Britton, a staff scientist at the lab, and Naomi Jarvis, a research associate in the physics department at Carnegie Mellon.
The DOE AI and machine learning proposals were chosen by competitive peer review in the program fields of Basic Energy Sciences, High Energy Physics and Nuclear Physics. This year’s projects target automating facility operations and managing data modeling, acquisition, data mining and analysis.