The offshore wind industry has grown rapidly over the last decade and drivetrains are increasing in size to reduce the cost of energy. These turbines are operating in a harsh environment. Adopting a preventive maintenance strategy is important to achieve an as high as possible availability of the farm and reduce the cost of maintenance. A well performing condition monitoring system that utilizes SCADA data from the wind farm can enable this strategy without the need in additional cost in hardware. This master thesis focusses on the development of a framework that can be utilized for this task. This framework can process raw operational SCADA data collected at the Egmond aan Zee offshore wind farm to create a clean dataset to train supervised machine learning models on. This work provides an insight in the correlation between different SCADA signals using a mathematical approach and from a understanding of the system integration of drivetrain components. Bearing temperatures are modelled using a data driven approach to describe the temperatures under healthy conditions. Several models are evaluated for this task and it was concluded that a decision tree supervised machine learning regression model resulted in the lowest error between predicted and measured values. Anomalies are detected and tracked with a normal behaviour model and a Sherward and CUSUM control chart that are applied on the residual error between modelled and measured temperature signals. 4 anomalies could be identified in the gearbox bearings using the developed framework. Abnormal behaviour of the drivetrain could be identified as early as 1 month before the turbine was taken out of productions. This highlights that temperature based condition monitoring that utilizes SCADA data can be used for early detection of faults by combining the accuracy of supervised machine learning methods with different fault detection methods like the CUSUM control chart. This work also investigates the relation between experienced wake of a wind turbine and the influence on the drivetrain component temperatures. The wake conditions at Egmond aan Zee, modelled with an Ishahara wake model, and the component temperature measurements from the SCADA data are used for this analysis. The bearing temperature distributions under different operational and wake conditions can be compared by clustering over the wind speed and the velocity deficit or turbulence intensity at turbine level. It is concluded from this work that wake effects do not result in a change in drivetrain component temperatures. The effects of asymmetric wake conditions opposed to wake experienced over the entire rotor is analysed by comparing the temperature distributions under these conditions in a cluster where the turbine is partially waked. A small shift towards higher component temperatures can observed on a limited amount of data for turbines under asymmetric loading conditions.