Christian FreudeORCID iD, Lukas LippORCID iD, Matthias Zezulka, Florian Rist, Michael WimmerORCID iD, David HahnORCID iD
Inverse Simulation of Radiative Thermal Transport
Computer Graphics Forum, April 2025.

Information

Abstract

The early phase of urban planning and architectural design has a great impact on the thermal loads and characteristics of constructed buildings. It is, therefore, important to efficiently simulate thermal effects early on and rectify possible problems. In this paper, we present an inverse simulation of radiative heat transport and a differentiable photon-tracing approach. Our method utilizes GPU-accelerated ray tracing to speed up both the forward and adjoint simulation. Moreover, we incorporate matrix compression to further increase the efficiency of our thermal solver and support larger scenes. In addition to our differentiable photon-tracing approach, we introduce a novel approximate edge sampling scheme that re-uses primary samples instead of relying on explicit edge samples or auxiliary rays to resolve visibility discontinuities. Our inverse simulation system enables designers to not only predict the temperature distribution, but also automatically optimize the design to improve thermal comfort and avoid problematic configurations. We showcase our approach using several examples in which we optimize the placement of buildings or their facade geometry. Our approach can be used to optimize arbitrary geometric parameterizations and supports steady-state, as well as transient simulations.

Additional Files and Images

No additional files or images.

Weblinks

BibTeX

@article{freude-2025-iso,
  title =      "Inverse Simulation of Radiative Thermal Transport",
  author =     "Christian Freude and Lukas Lipp and Matthias Zezulka and
               Florian Rist and Michael Wimmer and David Hahn",
  year =       "2025",
  abstract =   "The early phase of urban planning and architectural design
               has a great impact on the thermal loads and characteristics
               of constructed buildings. It is, therefore, important to
               efficiently simulate thermal effects early on and rectify
               possible problems. In this paper, we present an inverse
               simulation of radiative heat transport and a differentiable
               photon-tracing approach. Our method utilizes GPU-accelerated
               ray tracing to speed up both the forward and adjoint
               simulation. Moreover, we incorporate matrix compression to
               further increase the efficiency of our thermal solver and
               support larger scenes. In addition to our differentiable
               photon-tracing approach, we introduce a novel approximate
               edge sampling scheme that re-uses primary samples instead of
               relying on explicit edge samples or auxiliary rays to
               resolve visibility discontinuities. Our inverse simulation
               system enables designers to not only predict the temperature
               distribution, but also automatically optimize the design to
               improve thermal comfort and avoid problematic
               configurations. We showcase our approach using several
               examples in which we optimize the placement of buildings or
               their facade geometry. Our approach can be used to optimize
               arbitrary geometric parameterizations and supports
               steady-state, as well as transient simulations.",
  month =      apr,
  articleno =  "e70048",
  doi =        "10.1111/cgf.70048",
  issn =       "1467-8659",
  journal =    "Computer Graphics Forum",
  pages =      "14",
  publisher =  "WILEY",
  keywords =   "Ray tracing, Physical simulation, Computer-aided design",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2025/freude-2025-iso/",
}