Inactive

TECHNOLOGY LICENSING OPPORTUNITY Radiation Spectra Analysis for Environmental Changes using Reinforcement Learning An accurate and repeatable analysis of radiation spectra using an unsupervised learni... TECHNOLOGY LICENSING OPPORTUNITY Radiation Spectra Analysis for Environmental Changes using Reinforcement Learning An accurate and repeatable analysis of radiation spectra using an unsupervised learning technique where the agent directly learns a policy via deep learning. Opportunity: Idaho National Laboratory (INL), managed and operated by Battelle Energy Alliance, LLC (BEA), is offering the opportunity to enter into a license and/or collaborative research agreement to commercialize this new energy resolution upscaling technique. This technology transfer opportunity is part of a dedicated effort to convert government-funded research into job opportunities, businesses and ultimately an improved way of life for the American people. Overview: Various industries including nuclear, medical, and non-proliferation, have been using radiation spectroscopic analysis for decades. Despite its wide use, this technology has always struggled to provide accurate spectroscopy measurements for their applications. Efforts have been made, and continue, to improve the mathematical models for peak-fitting and subsequent analysis but the core concept of spectroscopy remains the same. Conventional methods commonly take many days to perform because of the substantial human factor involved in the analysis, just for the analysis to be inaccurate. All of these factors contribute to an increase in the financial and time expense, which leads user to either accept lower quality results, or forgo measurements all together. Description: Researchers at Idaho National Laboratory have developed an accurate and repeatable analysis method of radiation spectra using an unsupervised learning technique known as reinforcement learning where the agent directly learns the policy via deep learning. It performs a training procedure using spectra analysis data with both radionuclide identification and quantification serving as training data, executes a Markov Decision Process with a reward structure that includes negative rewards during training, and then infers the spectra analysis when provided unknown spectra. This inferred analysis returns both radionuclide identification and quantification within milliseconds at high accuracy. This new unsupervised learning algorithm can account for changes in the collection environment of the spectra including background. The algorithm can train against a wide array of experimental spectra collected over large timescales and incorporate changes in the experimental environment when executing inference without expert intervention. Benefits: The advantage over supervised learning approaches is that the policy can robustly handle changes in experimental acquisition environment. Provides a single policy that can perform spectra analysis without sacrificing inference accuracy. Highly accurate Applications: Any spectroscopic analysis including, but not limited to: Neutron Gamma Ray X-ray Charged Particle Chemical Development Status: TRL 5, technology has been demonstrated in a laboratory environment with real-world conditions. IP Status: Provisional Patent Application No. 63/320,674, “Reinforcement Machine Learned Spectrum Analysis,” BEA Docket No. BA-1256. INL is seeking to license the above intellectual property to a company with a demonstrated ability to bring such inventions to the market. Exclusive rights in defined fields of use may be available. Added value is placed on relationships with small businesses, start-up companies, and general entrepreneurship opportunities. Please visit Technology Deployment’s website at https://inl.gov/inl-initiatives/technology-deployment for more information on working with INL and the industrial partnering and technology transfer process. Companies interested in learning more about this licensing opportunity should contact Andrew Rankin at td@inl.gov.