Image Based Heliostat-Calibration in Solar Tower Power Plants with Differentiable Kinematics and Artificial Intelligence

Abstract

The transformation toward a climate-neutral energy system presents significant technical challenges. Solar tower (ST) power plants represent a promising solution, offering not only electricity generation but also industrial process heat. Unlike wind or photovoltaic systems, they can be equipped with thermal energy storage, enabling dispatchable power delivery. However, one of the key technological limitations is the accurate tracking of heliostats, since even small misalignments can lead to considerable optical energy losses. This thesis investigates a novel approach to heliostat calibration based on computerautomated analysis of flux measurements. The central idea is to extract structural features from measured flux images and use them to guide alignment optimization. In particular, the upper contour of the focal spot is identified as a robust alignment feature, as it is less affected by shading and blocking by neighboring heliostats than the typically used flux centroid. The core contribution of this work is the design of a differentiable optimization framework that operates directly on image-based features. The method is tested using synthetic datasets that include controlled scenarios of partial flux obstruction, allowing for systematic validation under realistic distortions. A custom loss function is developed, integrating multiple alignment objectives and tuned through Bayesian optimization. Results demonstrate that the image-based method yields reliable and accurate calibration performance – even in the presence of severe occlusions. Compared to conventional centroid-based strategies, the proposed approach consistently achieves higher alignment precision. Overall, this thesis lays a conceptual foundation for future calibration systems that incorporate the automated detection of higher-order, structurally meaningful flux features—paving the way for robust and scalable solutions in industrial ST-applications