ARTIST

ARTIST stands for AI-enhanced differentiable Ray Tracer for Irradiation Prediction in Solar Tower Digital Twins.
The ARTIST package provides an implementation of a fully differentiable ray tracer using the PyTorchmachine-learning framework in Python. Leveraging automatic differentiation and GPU computation, it facilitates the optimization of heliostats, towers, and camera parameters within a solar field by combining gradient-based optimization methods with smooth parametric descriptions of heliostats.

Our key contributions include:

• Immediate deployment: ARTIST enables deployment at the beginning of a solar thermal plant's operation, allowing for in-situ calibration and subsequent improvements in energy efficiencies and cost reductions.

• Neural-network driven heliostat calibration: A two-layer hybrid model for most efficient heliostat calibration. It comprises a robust geometric model for pre-alignment and a neural network disturbance model, which gradually adapts its impact via regularization sweeps. In this way, high data requirements of data-centric methods are overcome while maintaining flexibility for modeling complex real-world systems. Check out this paper for more details:

  M. Pargmann, M. Leibauer, V. Nettelroth, D. M. Quinto, & R. Pitz-Paal (2023). Enhancing heliostat calibration on low data by fusing robotic rigid body kinematics with neural networks. Solar Energy, 264, 111962.
  https://doi.org/10.1016/j.solener.2023.111962

• Surface reconstruction and flux density prediction: Leveraging learning Non-Uniform Rational B-Splines (NURBS),ARTIST reconstructs heliostat surfaces accurately using calibration images commonly available in solar thermal power plants. Thus, we can achieve sub-millimeter accuracy in mirror reconstruction from focal spot images,contributing to improved operational safety and efficiency. The reconstructed surfaces can be used for predicting unique heliostat flux densities with state-of-the-art accuracy. Check out this paper for more details:

  M. Pargmann, J. Ebert, D. M. Quinto, R. Pitz-Paal, & S. Kesselheim (2023). In-Situ Solar Tower Power Plant Optimization by Differentiable Raytracing. Under review at Nature Communications.
  https://doi.org/10.21203/rs.3.rs-2554998/v1