Predictive analytics for satellite instrumentation

Predictive analytics for satellite instrumentation


A space-borne earth observation instrument uses a miniature gas discharge lamp for internal reference. The aging of the lamps must be predicted with accuracy in order to select lamps that will operate flawlessly over the entire duration of the satellite. How do you choose the lamp to use in your 100 M$ satellite?


Using deep neural networks and physical parameters derived from images taken at given time intervals we are able to model characteristics that allow to distinguish between good and bad lamps.


  • Data-driven ranking of the lamps
  • Performance¬†prediction over a 10-year period
  • Reduction of one of the most important risk of the project, as stated by NASA