Choose timezone
Your profile timezone:
FRAS: meeting HPA/APC #1
→
Europe/Zurich
926/1-039 - Priority BE-GM (CERN)
Participants: Vivien Rude , Andreas Herty, Jürgen Gutekunst, Francis Klumb, Guillaume Kautzmann
Objectives
Kickstart APC and specially Jurgen's work for the FRAS monitoring.
As part of the discussion / presentation: a non-exhaustive list of subject to cover:
- Hardware : sensors used, control system,...
- Observation scheme and computing process: existing Matlab, strategy for computing…
- Requirements : phase of the project, current delays, frequency, etc…
- On APC side: Jurgen presents his background which can give some ideas.
Minutes
Presentation by Jürgen
Jürgen presented his background in Optimal Control Problem.
Jürgen spoke about the "Initial Value Embedding" technics leading to a faster evaluation of the solution after a measurement is available. The derivative matrix can already be prepared in advance. Discussion followed to see if we can keep an old derivative matrix for speeding up the adjustment process (for instance updating this matrix every ten seconds only). Answer by Jürgen: doable but loosing precision over time. Would need to be evaluated on the spot if a change of variable is needed for adjustment speed issue.
LGC speed issue were raised. Indeed, the combined cases have some performance issue. APC mentioned that the problem is already known and is being investigated by Jürgen.
Jürgen raise a question : "How often the statistics (var cov) are needed?" This operation is indeed cost expensive for fast monitoring.
Presentation by HPA Team
Vivien presented the FRAS plan including the configuration of sensor, the planning and the requirements for APC developments.
For the planning, 3 steps are nowadays considered:
- Single component: A single component without internal cold mass monitoring will be installed. It was agreed that the current MatLab will primarily be used for this step.
- String Test: a mock up on scale 1-1 will be installed. 6 components (including 4 quadripoles with internal FSI cold mass monitoring) compose this test. The global software for the monitoring system is required for February 2024 at the latest.
- HL-LHC planned for LS3
Mentioned by Andreas, the general software to develop must work in these 3 minimum phases:
- "Single observation", a single computation on demand. For instance: having an initial position of the system, etc... The current system can only give an exploitable result on every refresh of NXCAL DB (5 minutes), not practical for alignment operations
- "Active alignment", measurement done while moving the element. Point to take into account: parametrization of the weights of some observation sensor. For instance, the water reference surface have ripples on them during this phase and should have less weight in the computation.
- "Monitoring", measurement while the system is stable. The system should give a result with a 1Hz frequency.
Discussions also mentioned some of the feature required:
- No monitoring of the transient phases of the cold mass is asked by the equipment owner for now.
- The whole system should be parametrizable, scalable and standardized as much as possible (same model applied for each IP). APC mentioned the case of the monitoring of HIE-ISOLDE where everything is configurable via XML configuration files.
- The interface of the general software should also provide an easy way to deactivate sensor or play with some parameters: should logically be integrated in the software provided by BE-ICS.
- An automatic outlier detection must also be performed on the systems. 2 levels of outlier detection: during acquisition / pre-processing of the data and during the adjustment phase. Consequences in terms of observation.
- FESA classes need to be developed (BE-CEM).
- The general software must be integrated in the CERN infrastructure: WInCCOA and NXCAL
The current solution for monitoring passes by GEODE. APC/HPA came to the conclusion that the project is too big and not appropriate to be handled by GEODE. Indeed GEODE should remain the place where final calculated elements positions are stored, but should neither prepare LGC data nor launch the calculations.
About the APC contribution:
APC can take the responsibility of the LS computations (LGC), on the basis of corrected and pre-processed data provided by the system (example: Windows and refractivity index correction for the FSI,...), and will output the results in a structured files. LGC remains a small part of the general software and support can be provided by APC for the calculations relying on it.
HPA raised concerns about the deadline to be ready for the string test (beginning 2024). It involves from the APC side to develop:
- new observation models (HLS, WPS mainly, see if the FSI needs a dedicated model, possibly a sagging effect to develop for frames or objects),
- a reliable outlier detection in the observations,
- a software version addressing the performance issues (the "combined" adjustment model currently involves large matrices!).
Options for the development: L1 norm would be appreciated as well as a Kinematic predictive model (with respect to the result, give input to the motors to move efficiently the element). From APC Side: L1 should be investigated, not simple to develop from scratch, must prospect to see the available libraries... The kinematic predictive model should be implemented by an other module (part of the overlooking software?) with an LGC result as input.
APC mentioned that those goals are achievable in the given time frame. Nevertheless, APC will assess during summer 2022 the partially developed HLS observation model to have a better view on the developments to make. A second FRAS HPA/APC meeting should be organized in autumn 2022 to review the situation and take common decisions on the LGC2 route (continue, need more resources, abandon, etc...).
Also the 3D computation for the monitoring of point 1-5 is put on hold to free time for APC to make this assessment and prioritize the development of the new observation models needed.
Discussions also pointed out that a special LGC branch for monitoring (specific software version) could be necessary. The situation will be evaluated.
General monitoring software
The development of the overlooking software orchestrating all the different sub tasks (acquisition, launching and preparing files for LGC, communicating with DB, etc...) raised worries from both APC and HPA members. The amount of work to be done is huge and clear resources are not yet identified/available. In addition Francis deeply illustrated the complexity of such task using his experience of the HIE-ISOLDE monitoring software development (MATHIS).
HPA/APC will contact their management to raise a warning considering the development of the overlooking manager from acquisition to integration of the data in the DB.
There are minutes attached to this event.
Show them.
