OPTIMAL SELECTION OF BAYESIAN VIRTUAL SENSORS FOR DAMAGE DETECTION UNDER VARIABLE ENVIRONMENTAL CONDITIONS

Abstract

Measuring structural vibrations with a large sensor network results in lots of data in structural health monitoring applications. A large number of sensors is advantageous for damage detection and localization. By storing only a few selected Bayesian virtual sensors it is possible to decrease the amount of data and reconstruct the discarded sensor data even with higher accuracy than the original measurements. A method is proposed, in which the stored and reconstructed data are used for damage detection and localization in the time domain. A numerical experiment was performed with a structure having a large number of sensors. The excitation and environmental conditions were variable and unknown. An optimal sensor placement algorithm was applied individually to each measurement to select the appropriate virtual sensors for storage. Less than ten percent of the data were stored, and the signals of all the reconstructed sensors were still more accurate than the actual measurements. The stored and reconstructed data outperformed the actual measurement data in damage detection and localization. Surprisingly, damage detection was also more successful with the stored and reconstructed data than with the full set of virtual sensors.