In this work, we present a data-driven methodology to overcome actuator faults in empirical model-based feedback control. More specifically, we introduce the use of a moving horizon error detector that quantifies prediction errors and triggers updating of the model used in the controller online when significant prediction errors occur due to the loss of one of the actuators. Model reidentification is conducted online using the most recent input/output data collected after the fault occurrence. The error-triggered online model identification approach can be applied to overcome various types of actuator faults, including the case where the value at which the actuator is stuck is known and the case where the value at which the actuator is stuck is unknown. The proposed methodology is applied in an economics-based feedback controller, termed economic model predictive control (EMPC), that uses a model obtained either from first-principles or from plant data to optimize plant economics online. Two different chemical process examples are considered in order to demonstrate the application of the proposed strategy. In the first example, application of the proposed scheme for the case where the value at which the actuator is stuck is known is demonstrated through a benchmark catalytic chemical reactor example where the actuator faults occur in the heat input causing shifts and variations in plant operating conditions. The second example demonstrates the case where the value at which the actuator is stuck is unknown. The proposed scheme was able to compensate for the variations in the plant caused by the actuator loss by obtaining more accurate models that are suitable for the new conditions and updating them in the EMPC architecture. Improved economic performance was obtained as the updated models were able to capture the process dynamics under the new conditions and provide better state predictions.