A reliable detection and characterisation of the different operating modes of a wind turbine is essential for the correct understanding of its behaviour and production performance. Wind turbines are usually installed in fleets, which offers richer datasets to exploit. However, blindly applying machine learning approaches to such datasets may mask turbine-specific features. Moreover, data originating from wind turbines often contain multiple heterogenous subsets of parameters, which not always can be considered simultaneously during analysis. We propose a method for advanced exploration of wind fleet data by uniformly converting the time series sensor measurements across the fleet into a discrete code of operating mode labels by use of the data exploration framework proposed by Dhont et al. (2020a). Subsequently, we illustrate the potential of the method to derive relevant insights about the turbine performance on a real-world dataset of a fleet of 4 turbines.