B2G4: Integrating 3D Blender models in Geant4 simulations for synthetic muography data generation

Kurzfassung

Particle simulation software allows physicists to simulate the interactions of particles with matter at atomic and subatomic levels, providing means to study these interactions in a diverse range of scenarios and environments. Geant4 is one the most prominent toolkit for high-energy physics simulations, which can deal with particle transport through a wide range of materials for most standard model particles as well as simulate the subsequent detector response. Yet, one of the current limitations of Geant4 is its ability to create complex simulation geometries. Creating these geometries requires a significant amount of manual input, which can be time-consuming and error-prone. This work presents Blender2Geant4 (B2G4), a novel synthetic data workflow that translates visually generated 3D scenes in Blender to Geant4. B2G4 is built on the premise that Blender allows users to create sophisticated 3D scenes, which can then be exported as a geometry input for the Geant4 framework. The proposed B2G4 workflow uses the powerful modeling tools in Blender to position objects intuitively, construct layered objects, and assign realistic material properties that are compatible with Geant4. Further, the properties of each object, such as position or shape, can be modified and tailored to the simulation at hand using the randomization tools available in Blender. Additionally, B2G4 offers a range of tools to perform structural geometry checks to generate more realistic scenarios and improve the quality of the simulations. The included scripts and bindings in B2G4 allow for faster iteration on different simulation scenarios with Geant4, making it an efficient and effective approach to real-world scenarios. An emergent application of cosmic ray tomography is identifying potential threats for security domains such as borders or port facilities. Albeit deep neural networks can be trained to detect anomalies, the difficulty of acquiring real muography data and ground truth annotations remains the main limitation in the security domain. The applicability of B2G4 is demonstrated with the generation of complex scenes relevant to the security domain: with minimal manual coding, B2G4 enables the creation of a large volume of muography data based on feedback from customs authorities. The generated scenes demonstrate that B2G4 is a valuable framework for synthetic data generation, particularly when real muography data is unavailable or challenging to obtain.