Robust Monitoring of an Industrial IT System in the Presence of Structural Change

摘要: This research was motivated by two industrial problems in microelectronics and environment, where one has to reconstruct a spatial variable on disk from few number of experiments. In this context Gaussian process (GP) regression, or Kriging, is often used, sometimes coupled with Zernike polynomials which are orthogonal the disk. However, usual GP models do not take into account geometry their covariance structure (or kernel), which may be drawback at least for technological physical processes involving diffusion center disk, rotation. In talk we introduce so-called polar GPs, defined space coordinates. Their kernels obtained algebraically combining [0,1] circle. Their efficiency illustrated applications, where radial angular patterns visible. In second time, consider design experiments (DoE) issue GPs. We show how adapt construction optimal space-filling designs coordinates. Two new DoEs introduced compared D-optimal polynomials. Their prediction also assessed set various toy functions models.