Condition Monitoring (CM) has been considered as an effective method to enhance the reliability of wind turbines and implement cost-effective maintenance. Thus, adopting an efficient approach for condition monitoring of wind turbines is desirable. This paper presents a data-driven model-based CM approach for wind turbines based on the online sequential extreme learning machine (OS-ELM) algorithm. A physical kinetic energy correction model is employed to normalize the temperature change to the value at the rated power output to eliminate the effect of variable speed operation of the turbines. The residual signal, obtained by comparing the predicted values and practical measurements, is processed by the physical correction model and then assessed with a Bonferroni interval method for fault diagnosis. Models have been validated using supervisory control and data acquisition (SCADA) data acquired from an operational wind farm, which contains various types of temperature data of the gearbox. The results show that the proposed method can detect more efficiently both the long-term aging characteristics and the short-term faults of the gearbox.