European Southern Observatory Scoring: a novel approach towards automated certification of pipeline products Reinhard Hanuschik, Wolfgang Hummel, Mark Neeser, Burkhard Wolff Data Processing and Quality Control Group European Southern Observatory
Complexity and mass production VLT: 11 instruments VLTI: 2 instruments 2009/10: 2 survey telescopes roughly 50 modes about 150 data types most of them pipeline-supported
Complexity and mass production each data type or instrument component controlled by QC1 parameters metrics for instrument performance extracted by data pipelines stored in database data frequency varies on timescales between a day and a semester
Complexity and mass production two challenges … … the complexity challenge!
Complexity and mass production current VLT instruments: avg: 30 GB/night … the volume challenge! VIRCAM: OMEGACAM: 400 GB/night
The solution: The solution: scoring scoring: well-established outside astronomy … QC process: measure quality compare quality assess quality: score and certify
Scoring: the solution measure quality: done by automatic procedures pipelines other QC procedures implemented for CALIB data progress on SCIENCE data
Scoring: the solution compare quality: trending first step towards assessment: same/different behaviour as others?
Scoring: the solution assess quality: scoring compare new result to trending pick out outliers and flag them
Scoring: the solution required: thresholds statistical thresholds: e.g. +/- 3 sigmas + independent from specs - sensitivity variable control charts specified thresholds: e.g. lower/upper level + stable thresholds - requires careful configuration and thinking saturation noise
Scoring of pipeline products each new pipeline product is scored most relevant QC1 parameters: compare to trending, score as OK: 0 NOK: 1 (outlier) count scores, assign total score, e.g.
Scoring of pipeline products score report:
Scoring of pipeline products score reports for multi-detector instruments (CRIRES: N=4) … detailed information on demand!
Scoring of instrument health all scores go into database re-arrange them by instrument properties detector health calibration lamp performance system efficiency score-based instrument QC cal file1 scores FLAT instrument scores cal LAMPs cal file2 scores ARC
Scoring of instrument health example: detector properties for GIRAFFE instead of …
Scoring of instrument health and finally: www.eso.org/HEALTH
Conclusions scoring: automatic flagging of outliers useful to auto-certify pipeline products quick-look monitoring of instrument health powerful concept to handle large and complex data sets hierarchical approach: from global overview (instrument score) to QC1 parameter per product and detector (6 levels)