The Real-Time Data Analysis and Decision System for Particle Flux Detection in the LHC Accelerator at CERN.

The superconducting Large Hadron Collider (LHC) under construction at the European Organisation for Nuclear Research (CERN) is an accelerator unprecedented in terms of beam energy, particle production rate and also in the potential of self-destruction. Its operation requires a large variety of instrumentation, not only for the control of the beams, but also for the protection of the complex hardware systems. The Beam Loss Monitoring (BLM) system has to prevent the superconducting magnets from becoming normal conducting and protect the machine components against damages making it one of the most critical elements for the protection of the LHC. For its operation, the system requires 3600 detectors to be placed at various locations around the 27 km ring. The measurement system is sub-divided to the tunnel electronics, which are responsible for acquiring, digitising and transmitting the data, and the surface electronics, which receive the data via 2 km optical data links, process, analyze, store and issue warning and abort triggers. At the surface installation, the processing units (BLMTCs) include Field Programmable Gate Array (FPGA) devices. Each FPGA is treating the beam loss signals collected with a rate of 25 kHz from 16 detectors. It calculates and maintains 192 moving sum windows, giving loss duration integrations of up to the last 84 seconds. For the generation of the abort triggers, demanding the extraction of the circulating beams, it compares the moving sums calculated with threshold values chosen for the given beam energy. Those thresholds can be uniquely programmable for each detector allowing calibration and offset factors to be included. In this thesis, the BLMTC's design is explored giving emphasis to the methods followed in providing very reliable physical and data link communication layers, in merging the data from a Current-to-Frequency converter and an ADC into one value, and in keeping the moving sums updated in a way that gives the best compromise between memory needs, computation, and approximation error.