Master Student Machine Learning, Informatics, Engineering, Physics or similar (f/m/x)

Do you want to actively shape a sustainable future through your research in machine learning? Then you've come to the right place! At the German Aerospace Center's (DLR) Institute of Solar Research, scientists and engineers are working to make environmentally friendly energy supply technically and economically feasible using modern Artificial Intelligence (AI) Techniques.

In the Solar Power Plant Technology department, we conduct machine learning tasks with experimental validation to develop and test methods for an intelligent and autonomous solar tower power plant of the future.

At a solar power plant, a field of dual-axis tracking mirrors, called heliostats, focuses sunlight onto a central receiver. Each heliostats focal spot has a desired aimpoint, which is determined by complex algorithms depending on time and weather. Precision is crucial, as it directly impacts the plant's efficiency. Traditionally, each heliostat undergoes regular calibration to maintain stable precision throughout the year. However, despite continuous efforts, achieving the desired level of accuracy has remained elusive.

We are seeking a talented individual to help us overcome these challenges by developing a neural network which learns how to move all heliostats depending on direct feedback from the receiver. In this exploratory study, a ray tracer is employed to generate superimposed flux densities within a designated area. As the sun moves, the focal points gradually shift out of this area. The network must learn to adjust the heliostats based on the sun’s position and identify which heliostats typically require additional corrections to stay within the target area. Various network architectures, such as GANs or transformers, and learning algorithms, including supervised and reinforcement learning, can be applied for this task.

If the test proves successful, further optimization of the distribution within the area is possible. Additionally, if time permits, experiments at the real solar tower power plant in Jülich could be conducted.

As part of our team, you will have the opportunity to work with cutting-edge GPU clusters and collaborate closely with experienced machine learning developers. This position offers a unique chance to contribute to innovative solutions that will enhance the efficiency and reliability of solar power plants, driving forward the future of renewable energy.

If you are passionate about applying advanced machine learning techniques to solve real-world problems and want to work in a collaborative environment with top-tier technology and expertise, we encourage you to apply.

• choose a neural network architecture and learning procedure (reinforcement or supervised)
• generate synthetic data based on the chosen learning procedure using an existing ray tracing pipeline
• train the neural network for keeping focal spots inside a defined area
• Optional: train for specific aim point distributions
• Optional: test at the Jülich research power plant

You will be part of an interdisciplinary and international team of scientists from the fields of engineering, physics, informatics and mathematics. Our
departments have experts in both modeling machine learning pipelines and experimental validation, who will be happy to support you.