Master thesis on characterizing radiation fields for radiation protection measurements via the Timepix3 pixel detector in cooperation with the Computer Graphics Lab at TU Braunschweig

The Timepix3 detector is a spatially resolving pixel detector developed at CERN with a sensor area of 196 mm². The sensor surface is divided into 65,536 pixels with a pixel pitch of 55 μm each. The two semiconductor elements of the hybrid detector (sensor and ASIC chip) are manufactured separately and then connected via bump bonds.

To begin with, measurements were carried out at PTB’s MicroBeam particle accelerator as well as at the beta secondary standard BSS 2 and the 120 kV X-ray facility to train neural networks. Simulations were additionally carried out on PTB’s Linux cluster via Allpix2 and Geant4 to be able to improve the training of neural networks. The results of the neural networks were exclusively validated on real (measured) data.

Figure 1: 2D representation of an alpha radiation event.

The neural networks were developed and trained with the aid of a framework based on Python and pyTorch which was implemented specifically for this purpose. This framework enables performant processing of Timepix3 data by using graphics acceleration.

Figure 2:  Schematic representation of a residual neural network used in this context and based on the spatially and temporally resolved track structure of a radiation event determined by a Timepix3 detector.

The procedures developed within the scope of the master thesis as well as the derived data sets are a milestone on the way to characterizing unknown radiation fields with the aid of a single detector.

Contact

F. Lehner, Department 6.3, Working Group 6.31

J. Roth, Department 6.3, Working Group 6.35

O. Hupe, Department 6.3