BEGIN:VCALENDAR VERSION:2.0 PRODID:-//CERN//INDICO//EN BEGIN:VEVENT SUMMARY:Exploring the structure of hadronic showers and the hadronic energ y reconstruction with highly granular calorimeters DTSTART;VALUE=DATE-TIME:20230905T094000Z DTEND;VALUE=DATE-TIME:20230905T100000Z DTSTAMP;VALUE=DATE-TIME:20260906T144108Z UID:indico-contribution-547-3262@indico.tlabs.ac.za DESCRIPTION:Speakers: Roman Poeschl (IJCLab\, France)\nPrototypes of elect romagnetic and hadronic imaging calorimeters\ndeveloped and operated by th e CALICE collaboration provide an\nunprecedented wealth of highly granular data of hadronic showers for a\nvariety of active sensor elements and dif ferent absorber materials. In\nthis presentation\, we discuss detailed mea surements of the spatial and\nthe time structure of hadronic showers to ch aracterise the different\nstages of hadronic cascades in the calorimeters\ , which are then\nconfronted with GEANT4-based simulations using different hadronic\nphysics models. These studies also extend to the two different absorber\nmaterials\, steel and tungsten\, used in the prototypes. The hig h\ngranularity of the detectors is exploited in the reconstruction of\nhad ronic energy\, both in individual detectors and combined\nelectromagnetic and hadronic systems\, making use of software\ncompensation and semi-digit al energy reconstruction. The results include\nnew simulation studies that predict the reliable operation of granular\ncalorimeters. Further we show how granularity and the application of\nmultivariate analysis algorithms enable the separation of close-by\nparticles. We will report on the perfor mance of these reconstruction\ntechniques for different electromagnetic an d hadronic calorimeters\, with\nsilicon\, scintillator and gaseous active elements. Granular calorimeters\nare also an ideal testing ground for the application of machine learning\ntechniques. We will outline how these tec hniques are applied to CALICE\ndata and in the CALICE simulation framework .\n\nhttps://indico.tlabs.ac.za/event/112/contributions/3262/ LOCATION: Meeting Room 2.64 - 2.66 URL:https://indico.tlabs.ac.za/event/112/contributions/3262/ END:VEVENT BEGIN:VEVENT SUMMARY:The use of Machine learning to improve quality control for the ATL AS Phase-II Upgrade LVPS bricks at CERN DTSTART;VALUE=DATE-TIME:20230905T092000Z DTEND;VALUE=DATE-TIME:20230905T094000Z DTSTAMP;VALUE=DATE-TIME:20260906T144108Z UID:indico-contribution-547-3202@indico.tlabs.ac.za DESCRIPTION:Speakers: Khathutshelo Phadagi (iThemba Labs)\nAbstract. The T ile Calorimeter (TileCal)\, a sampling hadronic calorimeter covering the c entral region of the ATLAS experiment\, will require new electronics to me et the requirements of the High-Luminosity LHC (HL-LHC). This talk will de monstrate how deep neural networks can improve quality control of the new Low Voltage Power Supply (LVPS) boards in the contest of the ATLAS Phase-I I Upgrade program for HL-LHC. Deep Neural Networks (DNNs) as a machine lea rning algorithm is used to analyze complex data from the LVPS Boards. The first initial testing done on the boards determined their reliability and performance. A total of eleven tests with a binary metric of PASS/FAIL mak e up the initial test station. The measurements are stored in a database a nd the multi-dimensional data is explored and then analyzed by a DNN algor ithm. The DNN model classifies the data and produces significant insights with predictions about the quality of the LVPS boards. These forecasts wil l help the Quality Control of the upgraded TileCal LVPS. Pre-production an d production of the LVPS boards will commence this year generating more da ta than before.\n\nhttps://indico.tlabs.ac.za/event/112/contributions/3202 / LOCATION: Meeting Room 2.64 - 2.66 URL:https://indico.tlabs.ac.za/event/112/contributions/3202/ END:VEVENT BEGIN:VEVENT SUMMARY:Deep learning techniques for energy clustering in the CMS electrom agnetic calorimeter DTSTART;VALUE=DATE-TIME:20230905T100000Z DTEND;VALUE=DATE-TIME:20230905T102000Z DTSTAMP;VALUE=DATE-TIME:20260906T144108Z UID:indico-contribution-547-3176@indico.tlabs.ac.za DESCRIPTION:Speakers: Polina Simkina (CEA)\nThe reconstruction of electron s and photons in the Compact Muon Solenoid (CMS) detector depends on topol ogical clustering of the energy deposited by an incident particle in diffe rent crystals of the electromagnetic calorimeter (ECAL). These clusters ar e formed by aggregating neighbouring crystals according to the expected to pology of an electromagnetic shower in the ECAL.\n\nThe presence of upstre am material causes electrons and photons to start showering before reachin g the ECAL. This effect\, combined with the 3.8T CMS magnetic field\, lead s to energy being spread in several clusters around the primary one. It is essential to recover the energy contained in these satellite clusters to achieve the best possible energy resolution. Historically\, satellite clus ters have been associated to the primary cluster using a purely topologica l algorithm which does not attempt to remove spurious energy deposits from additional pileup interactions (PU). The performance of this algorithm is expected to degrade during LHC Run 3 (2022+) because of the larger averag e PU levels and the increasing levels of noise due to the ageing of the EC AL detector.\n\nNew methods are being investigated that exploit state-of-t he-art deep learning architectures like Graph Neural Networks (GNN) and se lf-attention algorithms. These more sophisticated models improve the energ y collection and are more resilient to PU and noise. This talk will cover the challenges of training the models and the opportunities offered by the deep learning techniques.\n\nhttps://indico.tlabs.ac.za/event/112/contrib utions/3176/ LOCATION: Meeting Room 2.64 - 2.66 URL:https://indico.tlabs.ac.za/event/112/contributions/3176/ END:VEVENT BEGIN:VEVENT SUMMARY:Using NLP for Hardware Quality Control by predicting Alert Signals from Particle Accelerator Detectors DTSTART;VALUE=DATE-TIME:20230905T102000Z DTEND;VALUE=DATE-TIME:20230905T104000Z DTSTAMP;VALUE=DATE-TIME:20260906T144108Z UID:indico-contribution-547-3112@indico.tlabs.ac.za DESCRIPTION:Speakers: Nicholas Perikli ()\n*By Nicholas Perikli \nSchool o f Physics and Institute for Collider Particle Physics\, University of the Witwatersrand\, Johannesburg\, South Africa*\n\nAbstract.\nParticle physic s data consists of patterns in measurements that can be separated into hot topics and more mundane data. This approach is analogous to looking for k eywords or topics in huge text data by separating more specific words and phrases from the generalities of text through the application of NLP. This will be done using DCS alarm data. The NLP models that were constructed o r fine-tuned for text classification included SVM\, BERT- base-cased\, RoB ERTa-base\, as well as stacked LSTM and bi-LSTM. This was done on Google C olab using Pytorch and Python libraries\, and the hyperparameters were opt imised using the WandB platform\, in which an extensive Baye’s optimisat ion search was performed. The idea is to use the best-performing models i. e.\, BERT or RoBERTa and train them by fine-tuning their hyperparameters i n order to classify the alarms\, as well as predict future alarm signals\, and then follow the same procedure for an LSTM model and compare the resu lts. The inputs would contain information about the date and time the alar m was received\, the physical variable involved\, the type of error as wel l as the particular system\, component or sub- component affected. Since this data provides information about the detector components as well as th e abnormal values of the physical variables of their constituent parts dur ing a hardware malfunction\, as well as the length of time that is taken u ntil the issue is resolved\, this data can be used as a correlator for the status of other sub-detector components during a hardware malfunction of another component. Moreover\, the predictive power of this algorithm could avoid fatal errors in the functioning of the hardware and electronic syst ems especially during testing periods and upgrades and allow for faster an d more effective management and advancement of the hardware and electronic systems towards greater technological capabilities.\n\nhttps://indico.tla bs.ac.za/event/112/contributions/3112/ LOCATION: Meeting Room 2.64 - 2.66 URL:https://indico.tlabs.ac.za/event/112/contributions/3112/ END:VEVENT BEGIN:VEVENT SUMMARY:Ultimate precision of a tracking system in future high energy expe riments DTSTART;VALUE=DATE-TIME:20230905T090000Z DTEND;VALUE=DATE-TIME:20230905T092000Z DTSTAMP;VALUE=DATE-TIME:20260906T144108Z UID:indico-contribution-547-2764@indico.tlabs.ac.za DESCRIPTION:Speakers: Gang LI (Institute of High Energy Physics\, Chinese Academy of Sciences)\nOne of the top goals of a high energy experiment is to perform precision tests on the Standard Model and probe new physics bey ond the Standard Model. Therefore\, it is essential to precisely measure t he momenta and impact parameters of charged tracks. Because of the rapid a dvancement of technology\, excellent tracking systems could be built. The most accurate silicon pixel tracker is approaching the spatial resolution of micron-level and a material budget of sub-permille-level. As a result\, the trade-off between spatial resolution and material budget becomes crit ical. Analytical calculation and fast simulation are used to examine the m aximum accuracy of a tracking system with restricted resolution and materi al budget. These conclusions could be beneficial for future experiments.\n \nhttps://indico.tlabs.ac.za/event/112/contributions/2764/ LOCATION: Meeting Room 2.64 - 2.66 URL:https://indico.tlabs.ac.za/event/112/contributions/2764/ END:VEVENT END:VCALENDAR