Sensor placement strategy to inform decisions

Abstract

This paper introduces a computational strategy to determine optimal sets of sensor locations to support real-time operational decisions. We exploit unsupervised learning strategies (specifically self-organizing maps) to identify the most informative locations to place sensors. The sensor placement procedure is then combined with a Multi-Step-Reduced Order Modeling approach that exploits the low-dimensional map between the sparse sensed data and the decisions at hand. The approach is demonstrated for the real-time assessment of an unmanned aircraft wing panel undergoing structural degradation. For this application, we compare the optimal sets of sensor locations with random placements for a variety of sensor availabilities. By adopting our placement strategy, we achieve improvements in accuracy and robustness of capability predictions, even when measured data are sparse and cover less than 10% of the reference data.