RadiaSoft LLC, Boulder, Colorado, USA
ORNL RAD, Oak Ridge, Tennessee, USA
ORNL, Oak Ridge, Tennessee, USA
Access to neutron scattering centers, such as Oak Ridge National Laboratory (ORNL) and the NIST Center for Neutron Research, has provided beam energies to investigating a wide variety of applications such as particle physics, material science, and biology. In these experiments, the quality of collected data is very sensitive to sample and beam alignment, and stabilization of the experimental environment, requiring human intervention to tune the beam. While this procedure works, it is inefficient and time-consuming. In the work we present progress towards using machine learning to automate the alignment of a beamline in neutron scattering experiments. Our algorithm uses convolutional neural network to both learn a surrogate of the image data of the sample and to predict the sample contour using a u-net. We tested our algorithm on neutron camera images from the H2-BA powder diffractometer and the Topaz single crystal diffractometer beamlines of ORNL.
RadiaSoft LLC, Boulder, Colorado, USA
JLab, Newport News, Virginia, USA
RadiaSoft and Jefferson Laboratory are working together to develop a machine-learning-based smart alarm system for the CEBAF injector. Because of the injector’s large number of parameters and possible fault scenarios, it is highly desirable to have an autonomous alarm system that can quickly identify and diagnose unusual machine states. We present our work on artificial neural networks designed to identify such undesirable machine states. In particular, we test both auto-encoders and inverse models as possible tools for differentiating between normal and abnormal states. These models are being developed using both supervised and unsupervised learning techniques, and are being trained using CEBAF injector data collected during dedicated machine studies as well as during regular operations. Lastly, we discuss tradeoffs between the two types of models.
RadiaSoft LLC, Boulder, Colorado, USA
BNL, Upton, New York, USA
We present recent developments on control of x-ray beamlines for synchrotron light sources. Effective models of the x-ray transport are updated based on diagnostics data, and take the form of simplified physics models as well as learned models from scanning over mirror and slit configurations. We are developing this approach to beamline control in collaboration with several beamlines at the NSLS-II. By connecting our online models to the Blue-Sky framework, we enable a convenient interface between the operating machine and the model that may be applied to beamlines at multiple facilities involved in this collaborative software development.
RadiaSoft LLC, Boulder, Colorado, USA
BNL, Upton, New York, USA
Over the past several years machine learning has increased in popularity for accelerator applications. We have been exploring the use of machine learning as a diagnostic and tuning tool for transfer line from the AGS to RHIC at Brookhaven National Laboratory. In our work, inverse models are used to either provide feed-forward corrections for beam steering or as a diagnostic to illuminate quadrupole magnets that have excitation errors. In this talk we present results on using machine learning for beam steering optimization for a range of different operating energies. We also demonstrate the use of inverse models for optical error diagnostics. Our results are from studies that use combine simulation and measurement data.
THBR05
RadiaSoft LLC, Boulder, Colorado, USA
LBNL, Berkeley, California, USA
High repetition-rate, ultrafast laser systems play a critical role in a host of modern scientific and industrial applications. We present a prototype diagnostic and correction scheme for controlling laser focal position for operation at 10s of Hz. Our strategy is to couple fast wavefront sensor measurements at multiple positions to generate a focal position prediction. We then train a neural network to predict the specific adjustments to adaptive actuators along the beamline to provide the desired correction to the focal position at 10s of ms timescales. Our initial proof-of-principle demonstrations leverage pre-compiled data and pre-trained networks operating ex-situ from the laser system. We then discuss the application of a high-level synthesis framework for generating a low-level hardware description of ML-based correction algorithms on FPGA hardware coupled directly to the beamline. Lastly, we consider the use of remote computing resources, such as the Sirepo scientific framework*, to actively update these correction schemes in the presence of new data
*M.S. Rakitin et al., ’Sirepo: an open-source cloud-based software interface for X-ray source and optics simulations," Journal of Synchrotron Radiation25, 1877-1892 (Nov 2018).
THPV017
RadiaSoft LLC, Boulder, Colorado, USA
Modern particle accelerator facilities generate large amounts of data and face increasing demands on their operational performance. As the demand on accelerator operations increases so does the need for automated tuning algorithms and control to maximize uptime with reduced operator intervention. Existing tools are insufficient to meet the broad demands on controls, visualization, and analysis. We have developed a cloud based toolbox featuring a generic virtual accelerator control room for the development of automated tuning algorithms and the analysis of large complex datasets. This framework utilizes tracking codes combined with with algorithms for machine drift, low-level control systems, and other complications to create realistic models of accelerators. These models are directly interfaced with control toolboxes allowing for rapid prototyping of tuning algorithms. In this paper, we will provide an overview of our interface and demonstrate its utility for building beamline controls displays directly from accelerator simulation lattices. We will also demonstrate the use of our interface for testing online optimization and control algorithms.