The problem of damage detection in an operating wind turbine under normal operating conditions is addressed. This is characterized by difficulties associated with the lack of measurable excitation(s), the vibration response non-stationary nature, and its dependence on various types of uncertainties. To overcome these difficulties a stochastic approach based on Random Coefficient (RC) Linear Parameter Varying (LPV) AutoRegressive (AR) models is postulated. These models may effectively represent the non-stationary random vibration response under healthy conditions and subsequently used for damage detection through hypothesis testing. The performance of the method for damage and fault detection in an operating wind turbine is subsequently assessed via Monte Carlo simulations using the FAST simulation package.