A unified sampling-based framework for optimal sensor placement considering parameter and prediction inference

We present a Bayesian framework for model-based optimal sensor placement. Our interest lies in minimizing the uncertainty on predictions of particular response quantity interest, with parameter estimation being an intermediate step this purpose. By developing methodology that targets prediction inference rather than inference, we prioritize reduction parameters matter most actual interest. Currently available placement methods focus and might therefore yield suboptimal solutions inference. opt unifying where case is merely special Following quantification, model are treated as random variables their before data collection described by prior probability density function. The updated to posterior using measured depends chosen locations. This then converted uncertainty. As scalar measure uncertainty, use determinant covariance matrix. general type metric which can be used both Using expectation respect distribution possible objective function, locations optimized minimize expected or required matrices evaluated Monte Carlo sampling approach. verify procedure simple test example (simplified) study from structural dynamics modal show how differ those obtained In general, difference will depend uncertainties, way experimental parameters, parameters. Significant differences occur when well local nature optimizing allows adapting such they informative relevant subset.