arXiv subjects

K. Lamichhane

Publications and source records attributed to K. Lamichhane.

3 recordsLinked to original sources

arXiv2025

Vertex Imaging Hadron Calorimetry Using AI/ML Tools

The fluctuations in energy loss to processes that do not generate measurable signals, such as binding energy losses, set the limit on achievable hadronic energy resolution in traditional energy reconstruction techniques. The correlation between the number of hadronic interaction vertices in a shower and invisible energy is found to be strong and is used to estimate invisible energy fraction in highly granular calorimeters in short time intervals (<10 ns). We simulated images of hadronic showers using GEANT4 and deployed a neural network to analyze the images for energy regression. The neural network-based approach results in significant improvement in energy resolution, from 13 % to 4 % in the case of a Cherenkov calorimeter for 100 GeV pion showers. We discuss the significance of the phenomena responsible for this improvement and the plans for experimental verification of these results and further development.

physics.ins-det↗

arXiv2024

High-granularity Dual-readout Calorimeter: Evolution of a Classic Prototype

The original dual-readout calorimeter prototype (DREAM), constructed two decades ago, has proven instrumental in advancing our understanding of calorimetry. It has facilitated a multitude of breakthroughs by leveraging signals from complementary media (Cherenkov and scintillation) to capture fluctuations in electromagnetic energy fraction within hadronic showers. Over the years, extensive studies have shed light on the performance characteristics of this module, rendering it exceptionally well-understood. Drawing on this wealth of experience, we have embarked on enhancing the detectors' capabilities further by integrating fast silicon photomultipliers (SiPMs) with finer transverse segmentation, $\sim$1 cm$^2$, as well as longitudinal segmentation by timing measuring better than 10 cm. This configuration will allow us to image hadronic showers with high granularity (HG-DREAM). We argue that the spatial information provided by such a granular detector in a short time window ($\approx$5 ns) leads to substantial enhancement in energy resolution when advanced neural networks are employed in energy reconstruction. We briefly present the current status of work, new concepts that have been introduced to the detector, and expectations from simulations.

physics.ins-det↗

arXiv2022

Deep learning applications for quality control in particle detector construction

The growing complexity of particle detectors makes their construction and quality control a new challenge. We present studies that explore the use of deep learning-based computer vision techniques to perform quality checks of detector components and assembly steps, which will automate procedures and minimize the need for human interventions. This study focuses on the construction steps of a silicon detector, which involve forming a mechanical structure with the sensor and wire bonding individual cells to electronics for reading out signals. Silicon detectors in high energy physics experiments today have millions of channels. Manual quality control of these and other high channel-density detectors requires enormous amounts of labor and can be prone to errors. Here, we explore computer vision applications to either augment or fully replace visual inspections done by humans. We investigated convolutional neural networks for image classification and autoencoders for anomalies detection. Two proof-of-concept studies will be presented.

hep-ex↗