Vibration-based Structural Health Monitoring of operating wind turbines is challenging, as those structures are characterized by complex non-stationary response and are subject to environmental and operational uncertainties. FS-TARMA parameter based methods are ideal for this problem since they are capable of summarizing the non-stationary dynamics within a small parameter set. In this work, robust FS-TARMA parameter based fault detection methods are pursued by including several FS-TARMA models in the estimation of the statistical model used for posterior decision making. Different combination rules for the different FS-TARMA models are defined, analyzed and compared within the problem of vibration based fault detection on operating wind turbines using simulated data obtained from the FAST aeroelastic simulation code. Results demonstrate the improvement in terms of accuracy and reliability provided by the multiple model approach.