Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Reconstruction & Modelling Challenges for Large Volume Liquid Argon Detectors

Reconstruction & Modelling Challenges for Large Volume Liquid Argon Detectors

Talk delivered June 8, 2010 at the IPRD10 (Innovative Particle and Radiation Detectors, Siena 2010) conference.

6e803f34d50b5355d8cf39bdb4ce2932?s=128

Andrew J. Bennieston

June 08, 2010
Tweet

Transcript

  1. Reconstruction & Modelling Challenges for Large Volume Liquid Argon Detectors

    Andrew J. Bennieston University of Warwick IPRD 2010, Siena, Italy 7–10 June 2010
  2. Time Projection Chambers Interested in reconstruction for liquid Argon TPCs

    Detectors on the scale of 100 kiloton required for next-generation ν experiments TPCs use ionisation charge to track particles through gas or liquid volume E Readout Charged Particle Ionisation Charge 2D readout plane (xy) and z coordinate from drift time High-density, fine-grained 3D spatial data
  3. Liquid Argon TPCs (ICARUS, arxiv:0812:2373) LAr TPCs track events with

    bubble-chamber quality High resolution, homogeneous volume: tracks and showers develop side-by-side Automated reconstruction software not established
  4. Reconstruction Challenges Collider Experiments Well-defined primary vertex Sparse hit data

    radiating from this point Multiple scattering mostly at well-defined boundaries Tracks and showers develop separately; track hits easy to find LAr TPCs No well-defined start point for ν interactions High density of hit information throughout Multiple scattering throughout volume Tracks and showers develop together; tracks difficult to find amongst other hits Need to classify hit information (energy deposited) as tracks or showers to do PID and kinematics
  5. Track Reconstruction x -350 -300 -250 -200 -150 -100 -50

    0 y 0 20 40 60 80 100 120 140 160 180 ° Event at 30 x -160 -155 -150 -145 -140 y 128 130 132 134 136 138 140 142 144 (Zoom to vertex region) ° Voxel Data: Event at 30 Procedure: Start with simple geometric tracks; charge deposited into cubic voxels (volume elements) Compare reconstruction algorithms with the same data Algorithms: Hough Transform KDTree Search Corner Detection Clustering
  6. Hough Transform y = mx +c r = x cosθ

    +y sinθ y = − cosθ sinθ x + r sinθ Maps points in (x,y) to sinusoids in (r,θ) Intersect at (r,θ) parameters of straight lines in the image Originally used for bubble chamber images; now widely used in image processing y x r θ
  7. Hough Transform Red Allocated hits Black Unallocated hits Highest peak

    in HT gives line in a projection Hits allocated to line if they fall within some radius of it Iterative HT; allocated points not included in subsequent steps Projections used for HT; 3D data retained for final fits Segments combined from projections based on 3D fits
  8. KDTree (2,5) (6,3) (3,8) (8,9) Data structure: Binary search tree

    built in O(N logN) time Nearest- neighbour search in O(logN) time Algorithm: Pick seed point Find nearest (unallocated) neighbour Build up collection of points Fit line; histogram direction cosines Move onto next line when histogram develops second peak
  9. KDTree: Angle Reconstruction

  10. Clustering (Kinga Partyka, ArgoNeuT) Clustering hits based on density (DBSCANa

    and OPTICSb) Clustering could be used to find tracks Left: Hits coloured by clusters found by DBSCAN Problems with over/underclustering a Sander et al., Data Mining and Knowledge Discovery 2, pp169–194 (1998) b M. Ankerst et al., ACM SIGMOD Int. Conf. on Management of Data, pp49–60 (1999)
  11. Interest-Point Detection Hough Transform provides lines through image but no

    end points KDTree algorithm relies on moving through corners to see gradient changes KDTree algorithm requires a seed point; any will do, but some are better than others Clustering algorithms pick out related points, but some clusters ‘wrap around’ corners Interest-point detection finds corners and endpoints to help tracking algorithms
  12. Corner Finding Harris-Stephens/Plessey1 ‘cornerness’ measure Image I(x,y) has structure tensor

    S(x,y): S(x,y) =   I2 x Ix Iy Ix Iy I2 y   1 C. Harris, M. Stephens, Proc. 4th Alvey Vision Conf. pp147–151 (1988)
  13. Corner Finding GENIE νµ CCQE event (B. Morgan, Warwick) Vertex,

    proton track endpoint and delta electron identified 3D implementation (D. Roythorne, Warwick)
  14. Conclusions High levels of detail in LAr TPC events Promising

    results with spatial data & image processing techniques Ability to tag feature points with high efficiency is required for progress Showers are complex features which appear side-by-side with tracks Feature detection can be used as input to a variety of track & shower fitting algorithms Many more image processing & feature extraction techniques to explore Work is progressing in the UK, Europe and the U.S. — collaborative links are developing