Master Student Computer Science, Engineering, Physics or similar (f/m/x)

At the German Aerospace Center's (DLR) Institute of Solar Research, our goal is to make environmentally friendly energy supply both technically and economically viable through cutting-edge Artificial Intelligence (AI) techniques. In the Solar Power Plant Technology department, we focus on developing and testing machine learning methods for intelligent and autonomous solar tower power plants.

In a solar power plant, heliostats—two-axis tracking mirrors—concentrate sunlight onto a central receiver. Each heliostat has unique imperfections and needs regular calibration to ensure accurate sun tracking. Currently, the most common calibration technique is the camera target method. This involves directing heliostats to reflect sunlight onto targets near the receiver and capturing the reflected light with cameras. Image processing is then used to find the centroid of the focal spot and determine the heliostat's alignment. This alignment is compared with a heliostat model to adjust its parameters to match real-world conditions. However, this method's accuracy is limited because the centroid of the focal spot can be affected by mirror surface defects, leading to alignment errors.

To improve this process, we aim to develop an advanced machine learning approach that uses the entire image of the focal spot rather than focusing on just one point. Our plan is to combine two methods from our research group: first, reconstruct the heliostat surface from the focal spot image using a conditional generative deep learning model, and second, use this surface in a differentiable ray tracing environment to create a realistic focal spot. This simulated focal spot will be compared to the actual measurement, and the images will be used to refine heliostat alignment using machine learning techniques.

Your task will be to find the optimal learning procedures, extract the most information from these images, and develop the most effective way to align heliostats based on focal spot images.