Elastic actuators can provide safe human-robot interaction and energy efficient mobility. For this purpose they are ideal for wearable robotic applications. However, such actuators are subject to stiffness faults. We present a stiffness-fault-tolerant control strategy for complex elastic actuators, capable of adapting to changes in output stiffness, and demonstrate it on a smart variable stiffness actuator based on the MACCEPA concept. We develop the dynamics of the actuator and a model-based impedance control scheme. Biomechanical data extracted from the flexion/extension of a real knee joint are used as trajectory reference for the evaluation of the control concept in simulation. Results show that the controlled actuator is capable of tracking a reference trajectory under fault conditions and interaction disturbance while maintaining physical human-robot characteristics.