Optimization of sensor distribution using Gaussian processes

Abstract

Experiments are undertaken both to test a hypothesis and, as is usually the case in metrology, to estimate the values of quantities. Experimental design is the process of planning the experiment in order to ensure that measurements gathered are of the right type and are of sufficient quality and quantity to achieve the aims of the experiment as efficiently as possible. In this paper, we discuss the design of experiments for systems modelled in terms of Gaussian Processes that account for spatially (and/or temporally) correlated random effects. In particular, we look at how the number of measurements required depends on the degree of spatial/temporal correlation. These methods have been applied to determine the placement of a network of sensors on a thermal plate in order to determine the heat profile of the plate to sufficient accuracy with as few sensors as possible.

Keywords

• Gaussian processes
• Design of experiment
• Measurement uncertainty
• Sensor networks