The paper describes a successful technology transfer of Gaussian Process (GP) modelling, also known as kriging, to the field of industrial metrology. Product compliance to geometrical specifications typically requires an automated inspection cycle operated by a computer controlled machine which sequentially probes the part surface at a small sample of locations. Then the geometric error is computed from the set of point coordinates provided by the machine. Although the inspection plan can be naturally regarded as a statistical experiment, industrial practice generally relies on a deterministic logic both to choose the sample and to compute the geometric error. Opposed to this, we build the inspection plan as an adaptive experiment where the next probing location is selected by criteria based on predictions obtained from a GP model estimated at each step of the procedure. Results show that the good predictive capability of GP models assures an improvement over the current state of the art both in terms of quality of the estimated error and cost of the inspection.
Adaptive inspection plans in coordinate metrology based on Gaussian process models
POLINI, Wilma;
2013-01-01
Abstract
The paper describes a successful technology transfer of Gaussian Process (GP) modelling, also known as kriging, to the field of industrial metrology. Product compliance to geometrical specifications typically requires an automated inspection cycle operated by a computer controlled machine which sequentially probes the part surface at a small sample of locations. Then the geometric error is computed from the set of point coordinates provided by the machine. Although the inspection plan can be naturally regarded as a statistical experiment, industrial practice generally relies on a deterministic logic both to choose the sample and to compute the geometric error. Opposed to this, we build the inspection plan as an adaptive experiment where the next probing location is selected by criteria based on predictions obtained from a GP model estimated at each step of the procedure. Results show that the good predictive capability of GP models assures an improvement over the current state of the art both in terms of quality of the estimated error and cost of the inspection.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
