An Artificial Intelligence Framework to Design Radiation Shielding for Spacecraft and Satellites Using Machine Learning and Topology Optimization
Doctor of Philosophy (PhD)
Department of Physics & Astronomy
Purpose: To develop an artificial intelligence based generative design framework to optimize radiation shielding for satellites and spacecraft using a genetic algorithm combined with a generative adversarial network and compare the performance to using only a genetic algorithm.
Methods: The optimization framework is developed and tested on three different models, a simple cube, the SpaceX Dragon vehicle, and the STPSat-4 satellite. Each model's geometry is created in our Monte Carlo particle transport code with an added allowed shielding volume. Each model is exposed to a simulated radiation source using the Monte Carlo code. We use a genetic algorithm, GA, where the material distribution is represented by individuals in the algorithm. GAs are evolutionary algorithms primarily used to perform optimizations. The goal of the optimization is to minimize the dose to a specified volume inside the vehicle. Halfway through the optimization, the results are used to train a generative adversarial network, GAN, which is a type of generative model based on adversarial training. The GAN then produces new shielding solutions that are fed into the optimization process. The shielding solutions produced by the GAN-assisted-GA are compared to those produced by the GA only, and both are compared to the reference baseline of aluminum shielding.
Results: Our optimization framework was able to reduce the internal dose by up to 20% for the generic satellite, 14% for the Dragon vehicle, and up to 45% for the STPSat-4 for equal weights of uniform aluminum. We found the GAN-assisted-GA was able to improve the overall optimization performance by 6% for the simple cube and 8% for the Dragon vehicle.
Conclusions: Our genetic algorithm based optimization framework was able to produce radiation shielding profiles with reduced dose compared to aluminum references of the same weight. We also showed increased performance by incorporating a generative adversarial network into the genetic algorithm optimization process. This work demonstrates the ability of our approach to be used to design radiation shielding for spacecraft and satellites.
Taylor, Jared Joseph, "An Artificial Intelligence Framework to Design Radiation Shielding for Spacecraft and Satellites Using Machine Learning and Topology Optimization" (2023). LSU Doctoral Dissertations. 6167.
Available for download on Friday, May 17, 2030