Alignment Closing Strategy Silvia Borghi 6 th April
Alignment, Closing Strategy Silvia Borghi 6 th April 2008 Silvia Borghi
Outline Alignment Silvia Borghi, Marco Gersabeck, Stefano De Capua (PVSS use of resolver) n Perform alignment on TED data n Motion System position stored in database and used to update alignment constants n Implement alignment monitoring , add Green/Red light from alignment for data processing n Improve primary vertex alignment method for relative alignment of the two VELO halves n Alignment run in FEST n Test/tune HLT alleys for halo tracks (lower priority) Closing Malcolm John, Stefano De Capua (PVSS) n Test HLT vertex reconstruction and display in online monitoring with FEST data n Test PVSS closing procedure n Setup online presenter pages for beam position monitoring 2 Silvia Borghi 6 th April 2008
Alignment Silvia Borghi, Marco Gersabeck, Stefano De Capua (PVSS use of resolver) 6 th April 2008 Silvia Borghi
Perform Alignment for TED data Module alignment constants evaluated by different methods using Aug. and Sept. data samples n n Diff. of alignment const. obtained using Aug. and Sept. data samples for X and Y translations Millepede considering X & Y trans and Z rot, Kalman considering X & Y trans. The detector displacement from metrology is less than 10 m Module alignment precision is 5 m for X and Y translation and 200 rad for Z rotation 4 Silvia Borghi 6 th April 2008
Motion System position Resolver position from the motion system for x and y position of each VELO half n n n automatically updated into online Cond. DDB at beginning of run if it is changed from the previous one used by the HLT automatically updated into a private Cond. DDB each time it changes VELO position used in the reconstruction in LHCb reference system is determined by a function that combines the Alignment constants and the resolver position (transparent for the user). 5 Silvia Borghi 6 th April 2008
Align. const. and resolv. pos. in Cond. DB Align Const in Offline Cond. DB Read Module & Sensor Align. Const 2 halves Align const. not used! Resol. Pos in Online Cond. DB Closing procedure Read Velo Resol. Pos from PVSS Read Physics Data Taking Run Alignment Yes Changed? Write Read No Write Read Offline Reconstruction Read 6 Silvia Borghi 6 th April 2008
Implement alignment monitoring: Sensor Online monitoring: n n n Residual distribution for each sensor Overview: Residuals versus sensor number Histogram of Mean and RMS of residual distribution for each sensor Offline monitoring: n n n Measure size of misalignment from residual vs Φ distribution Analyse the plots by fitting sinusoidal functions using python scripts Script has two levels: warnings and alarms o o o n Warnings are counted Alarms are counted and printed for each module that causes an alarm Thresholds depend on fit values (≈ misalignments) and their significances Produce overview plots of both 7 Silvia Borghi 6 th April 2008
Implement alignment monitoring: Sensor Traces of a ‘Gauss bug’ in only 98 events Misalignment Significance Without Metrology = full effect of bug With Metrology = less affected by bug 8 Silvia Borghi 6 th April 2008
VELO alignment monitoring Monitoring of the VELO half alignment n n n Monitor VELO half alignment through primary vertices by Offline PV Tool Reconstruct PV position with tracks of only one half at a time Plot A and C side PV position Plot A-C side 2 D difference in PV position Script for warnings and alarms should be implemented 9 Silvia Borghi 6 th April 2008
VELO alignment monitoring Results on the FEST data week 2 nd -6 th March 2009 Only sensible conclusion from an alignment point of view: n PV left x – PV right x VELO halves are 8 μm apart Mean 8 m Beam pos. (x, y) = (15, 30) is nothing that can be used as an alignment constant n n ambiguous interpretation not at all indication of problems Mean 19 m Mean 11 m PV right x [mm] Silvia Borghi PV left-right x [mm] Mean 30 m PV left x [mm] 6 th April 2008 PV Y [mm] 10
VELO half alignment Velo half alignment via MILLEPEDE [minimisation done with single matrix inversion]: Overlaps tracks: n n n Measurements: point coordinate (x, y, z) Local variable: track slope and intercept (a, b, c, d) Global variable: half misalignment parameters ( x, y, z, , , ) Require a PR in the global frame to determine overlaps also in the case the VELO not completely closed PV method: n n n Measurements: track slope and intercept (a, b, c, d) Local variable: vertex coordinate (vx, vy, vz) Global variable: half misalignment parameters ( x, y, z, , ) 11 Silvia Borghi 6 th April 2008
PV method Improvement of PV method for the VELO half alignment, based on Millepede: Selection of event with only one PV found using only the tracks reconstructed in each half PV offline tool used to determine the tracks coming from the same PV Track slope and intercept are evaluated by a linear fit based on least square method PV coordinates are the local variables evaluated in Millepede (the PV position evaluated by the PV offline tool is not used) Performance of the method is under study. 12 Silvia Borghi 6 th April 2008
Alignment with FEST data Update of software-options for both methods Changes will be available in the new release of Escher FEST data on 2 nd-6 th March 2009 week n Input misalignment 10 micron misalignment along x between the two halves o o Kalman results 11 -12 m Millepede result 8 m It should be tested to be able to run on calibration farm 13 Silvia Borghi 6 th April 2008
HLT Alley: An alignment-friendly Pat. Rec The need for tracks parallel to the beam axis has been emphasised many times n n No suitable pattern recognition exists to easily do this and to be used at HLT level A new fast, efficient, and pure pattern recognition was developed The idea of the new pattern recognition: n n n Look for a certain number (small range) of space-points in r-phi projection of all modules (both halves) Check that this number is not produced by combinatorics on a single module Ensure that the number of space-points is a local maximum Optional: Check whether track candidate has a minimum number of spacepoints in both halves to detect overlap tracks Don’t fit the track candidate as this is thought to be a filter only, to select useful events for alignment 14 Silvia Borghi 6 th April 2008
HLT Alley: An alignment-friendly Pat. Rec The status: n Filter implemented as Pat. Velo. Align. Track. Filter in Tf/Pat. Velo o o n Two ‘alleys’ for generic parallel tracks and for overlap tracks specifically Can pre-scale differently and hence enhance overlap sample Outputs of these ‘alleys’ should end up in calibration stream Future plan n Tune the parameters on Montecarlo minimum bias events and ‘halo’ events 15 Silvia Borghi 6 th April 2008
Conclusion on Alignment n Perform alignment on TED data n Motion System position stored in database and used to update alignment constants n Implement alignment monitoring add Green/Red light from alignment for data processing Green/Red light should be tuned. The procedure should be integrated in the general alignment and data quality procedure n Improve primary vertex alignment method for relative alignment of the two VELO halves First results are promising but performance should be evaluated n Alignment run in FEST Missing test to run it on the calibration farm n Test/tune HLT alleys for halo tracks (lower priority) 16 Silvia Borghi 6 th April 2008
VELO closing Malcolm John, Stefano de Capua with Paula Collins, Eddy Jans, Kurt Rinnert 6 th April 2008 Silvia Borghi
What the VELO needs Safety — Detector integrity is paramount n The strategy is to use all information available from the detector and the machine to make a self-consistent decision Speed — Don’t waste any more stable-beam time than necessary motion takes 150 sec from OPEN to CLOSE. n Aim that monitoring, validation and decision-making < 30 seconds (total) n Want quasi-real-time feedback on the beam position (~2 s/measurement) n Strategy — Use the HLT for the CPU intensive 3 D reconstruction n Leaves the monitoring job “just” filling histograms. NB: In very early running, with lower luminosity, refresh rates will be slower 18 Silvia Borghi 6 th April 2008
HLT-resident, 3 D vertex reconstruction Decode. Lite. Cluster, Pat. Velo. Space, Pat. Velo. General, Pat. PV 3 D, HLTVelo. Closing. Decision Two independant lines for A-side and C-side. Vertex measure w. r. t. each half The resolver measurement of the opening is NOT broadcast to the HLT - too slow. HLTVelo-accepted event Vertex stored in a Raw. Bank Histogramming in the online monitoring Online monitoring job that uses a selective trigger mask (0 x 00 0 x 08 0 x 00) Unpacks Hlt. Vertex. Report from the Raw. Event and histograms vertex x, y, z & n. Trks Every 1000 events, the vertex distribution is fitted by a Gaussian and then cleared. The mean, RMS, mu and sigma are written to a PVSS data-point DIM data-point Luminous region position PVSS-based closing manager The core of the closing logic. Takes all available information, makes a judgement on its consistency and suggests the next move. . . Instruction sent to motion control DIM data-point 19 Silvia Borghi 6 th April 2008
Screenshot: ‘live’ VELO-open FEST data Latest 1000 events Accumulation Luminous region xposition w. r. t. the A-side = -30 mm Screenshot taken after ~250000 FEST minimum bias events (i. e. 1/40 th of a second of nominal running) N(trks)/vertex DIM data point monitoring N(evts): ~21700 triggered in HLT VELO lines ~12300 had a vertex found by A-side LOTS! But only requiring >4 trks/vertex here, In reality, will require 10 to improve resolution. 20 Silvia Borghi 6 th April 2008
PVSS closing manager Real-time vertex monitoring Resolver measurement LHC beam position information at IP 8 Silicon bias current Performs consistency checks between motion and measurements as well as monitoring the key hardware. Its output can be: a) Retract and park! b) Request human! c) Move to new position! BCM relative flux 21 Silvia Borghi 6 th April 2008
SCREEN-SHOT 22 Silvia Borghi 6 th April 2008
1 st version graphical page for closing manager Graph for A and C sides n n n X resolver position X vertex position X sigma of vertex position Graph n n n Beam position A side: X Res. Pos. – X Vert. Pos. C side: X Res. Pos. – X Vert. Pos. Res – PV for A side 14 mm – (-13 mm) =27 mm Beam position = 1 mm Res – PV for C side 23 Silvia Borghi 6 th April 2008
Conclusion for Closing n Test HLT vertex reconstruction and display in online monitoring with FEST data Online monitoring task keeps up with 100% of the HLT output Demonstrates the selective trigger mask and use of Hlt. Vertex. Reports Velo HLT lines work as expected on FEST data n PVSS closing manager extensively tested n Setup online presenter pages for beam position monitoring n Add graphical front page to closing manager to ease human digestion Plans n n n Revisit the closing strategy documentation based on recent updates. Write Twiki. Liaise with LHC to confirm the provision of their beam-position measurement and understand what information the wish from us Add new information - from tracks traversing both VELO halves. Investigate a new PR algorithm that forms space-points from overlapping phi-sensors when the VELO is <1. 7 mm open. 24 Silvia Borghi 6 th April 2008
- Slides: 24