SSRL - CAMERA Collaboration on Data Analytics for

Accelerating the Pace of New Discoveries

 

 

 

 

 

 

 

 

 

                            Xi-CAM screenshot of an SSRL scattering image with a detector tilt of 28 degrees.

                 The horizontal lines shown in the transformation into q/chi-space demonstrates proper treatment

                                                                     of detector tilt and calibration.

 

 

 

Stanford Synchrotron Radiation Lightsource at SLAC National Accelerator Lab has commissioned a pilot high throughput X-ray diffraction (HiTp) facility to explore and demonstrate how close integration of large scale computation with high throughput experimentation can rapidly decrease the time-to-discovery of new functional materials.  Currently, “humans in the loop” are heavily relied upon to curate (assess data quality and coverage), analyze and manage experimental data, and this manual approach will not keep pace with the rapid increase in data generation.  This separation of data collection from curation results in degradation of data quality and coverage, and separation of data collection and preliminary data analysis defeats the purpose of accelerating data collection.

 

The Goal:

 

The goal of the project is to take predictions from large scale computations to drive high throughput synthesis and characterization, to not only validate the predictions, but by using machine learning and other active learning methods to drive the next round of predictions and discoveries. The SSRL HiTp project will require automated and on-the-fly data reduction, information extraction and visualization tools to succeed.

 

A step in that direction is being provided through a strong collaboration extending the Xi-CAM environment, built by CAMERA (The Center for Advanced Mathematics for Energy Research Applications). Originally developed for small angle scattering experiments at Advanced Light Source. Xi-CAM aims to provide a community driven platform for multimodal analysis in synchrotron science, and is designed to be highly extensible and portable across platforms, beamlines, and facilities. The robust plugin infrastructure of Xi-CAM encourages continuing development to add further functionality.

 

Current Progress:

 

Working together, SSRL and CAMERA scientists are developing the Xi-CAM on-the-fly data reduction platform to meet the needs of a larger x-ray scattering community, most immediately for (GI)WAXS and (GI)SAXS with tilted detectors.

 

Already successful examples of SSRL’s extension of Xi-CAM include:  (1) a 2D image conversion algorithm, originally designed for small scattering angles with detector normal to the incident beam, now made applicable to tilted detector geometry and large scattering angles, and optimized so that 2Kx2K images (4M pixels) can be remeshed into diffraction coordinated (scattering angle Q, and azimuthal angle chi), masked, and integrated to conventional 1D scattering pattern in less than 100 msec on a small server residing at the beamline and (2) Extension and augmentation of  existing noise and artifact suppression features, such as masking, “zinger” (isolated outlier) removal to wide angle scattering patterns.

 

The prototype WAXS Xi-CAM was used for on-the-fly data reduction for a two week HiTp run at (SSRL) BL 1-5.  It was also used for two other high data throughput experiments at BL 11-3 and BL 7-2. 

 

The Future: 

 

In the near future, SSRL and CAMERA scientists plan to (1) implement additional image processing tools  (e.g. smoothing, peaks detection in 1D and 2D patterns, background removal); (2) Detect over, partially over and underexposed diffraction patterns in real time, and (3) Automate detector calibration.

 

In the long term, SSRL sees Xi-CAM as the back-bone for HiTp data analytics and management, and sees an SSRL-customized Xi-CAM as a standard data reduction platform for all high throughput and in-situ experiments at SSRL. SSRL envisions for Xi-CAM output to provide real-time feedback to the data collector, provide input to more advance unsupervised and machining learning clustering, trend and attribute detection engines, and interface with a sophisticated data visualizer.

 

The Team: 

On the SSRL side, the effort includes work of SSRL staff scientists Chris Tassone and Apurva Mehta. SSRL’s Fang Ren is guiding the incorporation of her unique visualization tools for XY-scanning sample plates and reducing large sets of q-spectra in Xi-CAM, and Amanda Fournier is continuing development of peak identification algorithms for feature extraction and optimization of cosmic ray ‘zinger’ detection and masking.

  CAMERA scientists tasked to work with SSRL include Alexander Hexemer,  Dinesh Kumar, and Ron Pandolfi.