Presented are online adaptive models for ventricular assist devices (VADs). Such devices are used to assist failing hearts or in the case considered here to create a total artificial heart. Adaptive models are developed to estimate cardiac output (CO) and power consumption of the VAD. These parameters are critical to physicians during patient care as well as in diagnosing VAD operation. The online adaptive nature of these models will be used to estimate effective blood viscosity in real-time and to create a mechanism whereby specific VAD diagnostics, important to robust CO delivery, can be identified, isolated and estimated. The experiments conducted were ex-vivo in a mock circulation loop in which the precise nature of the working fluid properties can be controlled and measured.