Katharina Dort ; Dr. Jens Soeren Lange

2019
II. Physics Institute Giessen, Germany

Abstract: The Belle II experiment located at the high-energy research facility KEK in Tsukuba (Ibaraki) Japan started operation in 2018. Compared to the predecessor experiment Belle, Belle II plans to increase the peak luminosity by a factor of 40, by employing nanobeam technology in the interaction region. A novel Pixel Detector (PXD) featuring two layers of DEPFET silicon sensors has been installed in Belle II for tracking of charged particles close to the interaction region. The DEPFET technology allows to thin down the sensitive detector region to only 75 μm. The low material budget reduces multiple scattering and eliminates the demand for active cooling. With pixel sizes ranging between 50 μm to 85 μm the PXD has an excellent spatial resolution, which is crucial for the tracking capabilities of Belle II. The PXD employs an online data reduction system in order to cope with high background occupancy. Clusters generated by charged particles, which do not leave a signal in the outer sub-detectors, are discarded allowing for a data reduction by a factor of 10. Despite being an efficient filter against background, the PXD reduction system also prevents the detection of signal particles, which are not registered by the outer sub-detectors. Among these, Highly Ionizing Particles (HIPs) possess a characteristically severe energy loss limiting their penetration depth into the detector. In particular, magnetic monopoles and anti-deuterons as possible HIP candidates are considered in this thesis. Without a signal in the outer sub-detectors, the particles remain undetected resulting in a loss of information about HIPs. In this thesis the possibility of identifying HIPs solely with information provided by the PXD is presented. PXD clusters generated by HIPs and background particles are compared and differences are analyzed. The separation between HIP and background clusters is performed with neural network algorithms operating in a multi-dimensional parameter space of PXD pixel information. The neural networks are found to outperform a sequence one dimensional cuts on PXD cluster properties. The influence of the input vector set is analyzed and the performance of the networks is critically reviewed. The prospect of employing neural networks trained on PXD data for offline and online applications is discussed.

Note: Presented on 25 04 2019
Note: MSc