Mixture of soft sensors for monitoring air ambient parameters

Monitoring the physical or chemical conditions of the materials composing a monument can be achieved in a not invasive way by using trained neural networks. Soft sensors based on Elman neural networks have been developed to provide virtual measurements at locations of the monument surface using only the measurements acquired by an Air Ambient Monitor Station located nearby the monument. Here we improve the accuracy of the virtual measurements by using averaging techniques or mixture of such soft sensors. The accuracy of these virtual instruments is analyzed and compared from a metrological and statistical point of view.