About BayWa r.e.
BayWa r.e. is the home for change makers. We r.e.think energy – how it is produced, stored and can be best used to enable the global renewable energy transition that is essential to the future of our planet. At BayWa r.e. we effect change globally, being a leading global developer, service provider, distributer, and solutions provider. With a focus on wind and solar energy, we have, so far, brought over 6 GW of energy online and manage over 10 GW of assets.
 

About the Project
Improving the management of existing renewable energy projects is vital. It boosts efficiency, cuts costs, and extends their lifespan, enhancing their contribution to the energy transition. To achieve this, BayWa r.e. relies on continuous data monitoring, extensive collection of historical data and the development of corresponding advanced analytics to gain profitable insights.
In alignment with this, the task of this project will be to gain valuable insights into the root causes of wind turbine downtimes, particularly in terms of specific component failures within the turbine system. By analyzing historical telemetry data, especially temperature and pressure measurements, the objective is to develop a deep learning method for time series classification, that categorizes each failure event (i.e. the respective telemetry signals) to its underlying cause and is able to predict future failures following the concept of predictive maintenance. The results of this can be used to improve the assessment of turbine degradation, gain better insights into failure indicators, and support data driven decisions for maintenance strategies.

The data basis for this project of failure event analysis, is a set of telemetry data which includes both general operational data as well as specific sensor measurements. Key operational parameters include, for instance, wind speed, wind direction, turbine orientation, blade angle or produced power, all of which are essential for understanding the overall performance of the turbine. In addition, the dataset also contain measurements from various sensors, such as temperature and pressure sensors, that monitor critical components of the turbine. The entire dataset encompasses historical data from a total of over 900 wind turbines, which were gathered over the last 4 years with a time resolution of 10 min. The classification labels for the failure events, which are essential for supervised learning, are extracted from the status codes of the wind turbines, that can be mapped to specific failure events. Nevertheless, to improve the respective knowledge base for a more detailed failure analysis, supplementary Natural Language Processing (NLP) techniques may be required, to extract additional insights from various text sources such as service reports of component replacements or repairs. Targeting a cloud-based solution, from a technical perspective, this project provides the great opportunity to engage with the latest state-of-the-art platform technologies, specifically Azure Databricks. This analytics platform was optimized for the Microsoft Azure cloud services and facilitates the seamless integration of data analytics, machine learning workflows, and scalable computing resources, ensuring an efficient processing and analysis of the extensive time series dataset of wind turbines.

Apply to this project here