AI-enhanced differentiable Ray Tracer for Irradiation Prediction in Solar Tower Digital Twins

What ARTIST can do for you

The ARTIST package provides an implementation of a differentiable ray tracer using the PyTorch machine-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 contributions include:

• 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. On 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.

• 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.

• 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.

• Optimized flux density: Coming soon so stay tuned 🚀!

Installation

We heavily recommend installing the ARTIST package in a dedicated Python3.8+ virtual environment. You can install ARTIST directly from the GitHub repository via:

pip install git+https://github.com/ARTIST-Association/ARTIST

Alternatively, you can install ARTIST locally. To achieve this, there are two steps you need to follow:

1. Clone the ARTIST repository:

   git clone https://github.com/ARTIST-Association/ARTIST.git

2. Install the package from the main branch. There are multiple installation options available:
   • Install basic dependencies: pip install .
   • Install with mpi4py to enable distributed computing: pip install ."[mpi]"
   • Install with optional dependencies to run the tutorials: pip install ."[tutorials]"
   • Install an editable version with developer dependencies: pip install -e ."[dev]"

Structure

The ARTIST repository is structured as shown below:

.
├── artist # Parent package
│   ├── field # Objects in the field, e.g. heliostats and receivers
│   ├── raytracing
│   ├── scene # Light sources and factors influencing the surroundings
│   └── util
├── measurement_data # Real measurements that can be used in ARTIST
├── scenarios # Scenarios describing a field that can be loaded by ARTIST
├── tests/
│   ├── field
│   ├── raytracing
│   ├── scene
│   └── util
└── tutorials # Tutorials to help you get started with ARTIST

Documentation

You can check out the full ARTIST documentation at https://artist.readthedocs.io/en/latest/index.html 🚀! The ARTIST documentation includes:

• Installation instructions
• Tutorials
• Some theoretical background information
• API reference

How to contribute

Check out our contribution guidelines if you are interested in contributing to the ARTIST project 🔥. Please also carefully check our code of conduct 💙.

Acknowledgments

This work is supported by the Helmholtz AI platform grant.

---