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