Semiautomatic dimensional screening of plastic scintillator cubes using image analysis and robotics

Atsuko K. Ichikawa; Mao Tani; Tatsuya Kikawa; Tomohisa Ogawa; Tsunayuki Matsubara; Tsuyoshi Nakaya

REVIEW 2 major objections 5 minor 16 references

A camera-and-robot system screens scintillator cubes to 10 μm and sorts them into 48 hole-position groups with only 3.1% rejected.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-13 21:47 UTC pith:FYQ73SNL

load-bearing objection Working multi-camera + robot QC for SuperFGD-type cubes: 10 μm hole-position reproducibility, >80% match to manual rod screening, and 3.1% reject via 48-way orientation-preserving sort on ~6500 cubes. the 2 major comments →

arxiv 2603.26732 v2 pith:FYQ73SNL submitted 2026-03-20 physics.ins-det hep-ex

Semiautomatic dimensional screening of plastic scintillator cubes using image analysis and robotics

Tatsuya Kikawa , Mao Tani , Atsuko K. Ichikawa , Tsunayuki Matsubara , Tsuyoshi Nakaya , Tomohisa Ogawa This is my paper

classification physics.ins-det hep-ex

keywords Neutrino detectorPlastic scintillatorQuality controlImage analysisRoboticsWavelength-shifting fibersDimensional screening

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Future neutrino detectors may need millions of identical 1 cm³ plastic scintillator cubes whose holes must line up well enough for wavelength-shifting fibers to pass through long rows. Manual rod screening is slow, subjective, and hard to scale. This paper builds and validates a semiautomatic alternative: six cameras and a rotating stage measure cube size, surface bumps, and hole positions; image analysis classifies each cube; and a six-axis robotic arm places accepted cubes into 48 orientation-preserving bins grouped by similar hole offsets. On a prototype the measurements are reproducible to about 10 μm and agree with manual screening more than 80% of the time. On roughly 6,500 cubes the finished system rejects only 3.1% while keeping within-group hole scatter below the geometric limit needed for fiber insertion. The result is a quantitative, non-contact, and potentially fully automatable quality-control path for large detector arrays.

Core claim

A completed inspection system that combines multi-camera imaging, calibrated image analysis, and a six-axis robotic arm can measure 1 cm³ scintillator cubes to 10 μm reproducibility, match manual rod screening on more than 80% of cubes, and sort them into 48 orientation-preserving hole-position categories so that the overall rejection rate falls to 3.1% while within-group hole variation stays under 500 μm.

What carries the argument

The 48-category, orientation-preserving sort performed by the six-axis robotic arm: among the three faces the one with the smallest hole offset is designated Z, the remaining two faces are binned into four intervals each, and the arm places the cube into the matching slot of a 48-slot box so that cubes with similar hole shifts can be stacked together.

Load-bearing premise

That keeping hole-position scatter inside each of the 48 groups below half a millimetre is enough to guarantee that fibers will actually pass through long rows of those grouped cubes, even though the paper never reports an end-to-end fiber-insertion test on the sorted product.

What would settle it

Assemble multi-meter rows from each of the 48 robot-sorted groups, attempt to thread 1 mm fibers through the 1.5 mm holes, and measure the fraction of holes that refuse insertion or show high stress; a substantial failure rate would falsify the claim that the 48-group scheme is sufficient.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

Working multi-camera + robot QC for SuperFGD-type cubes: 10 μm hole-position reproducibility, >80% match to manual rod screening, and 3.1% reject via 48-way orientation-preserving sort on ~6500 cubes.

read the letter

This is a practical instrumentation paper that delivers a built and run system, not a concept sketch. They put six cameras, a gravity-aligned rotating stage, OpenCV contour/Hough/circle fitting, lighting-alignment corrections, and later a 6-axis arm into a working line that measures 1 cm³ etched cubes and sorts them.

What is new is the full engineering package for this specific part: passive three-wall platforms that roll the cube to expose all faces, multi-configuration linear corrections that bring hole-position reproducibility from ~23 μm to ~10 μm (Fig. 9, 96 cubes), and the switch from binary accept/reject to 48 orientation-preserving bins so that cubes with similar hole offsets can still form insertable rows. The prototype already runs ~6 s/cube and matches independent manual rod labels at 82.5%/87.6% (Table 1). The completed robot system drops the reject rate to 3.1% on ~6500 cubes while keeping within-bin scatter under the 0.5 mm geometric clearance (Fig. 14).

The soft spots are real but secondary. Classification thresholds were tuned to the manual rod distributions, so the agreement numbers are not fully independent of that heritage. More importantly, they never actually thread fibers or rods through multi-cube rows drawn from a single robot-sorted bin; they argue from geometry and SuperFGD experience. That is the weakest link, but it does not break the measurement claims themselves. No public code or raw data is a minor production caveat, not a soundness problem.

Math and data look solid for an instrumentation paper: reproducibility quantified, failure modes stated (<0.5% roll failures), throughput and reject rates measured on large samples. Citations are appropriate (SuperFGD, OpenCV, Hough). This is for people building or planning large granular scintillator trackers (T2K-scale and DUNE near-detector concepts). It will not change detector physics, but it is useful engineering that a serious referee should see. I would send it to peer review.

Referee Report

2 major / 5 minor

Summary. The manuscript presents a semiautomatic quality-control system for 1 cm^{3} plastic scintillator cubes of the SuperFGD type, combining a motorized rotating stage, six cameras, OpenCV-based image analysis (contour detection, Hough transform, circle fitting, lighting/alignment corrections), and later a 6-axis robotic arm. The prototype measures cube size, surface protrusions, and hole positions with ~10 μm reproducibility after correction and shows >80% agreement with independent manual stainless-steel-rod screening (Table 1). The completed system classifies cubes into 48 orientation-preserving groups by hole position, achieving a 3.1% rejection rate on ~6500 cubes while keeping within-group hole scatter below the 0.5 mm fiber-insertion geometric tolerance. The work is motivated by the need for scalable, quantitative QC for future large neutrino detectors.

Significance. If the results hold, the paper supplies a concrete, quantitative alternative to the labor-intensive, subjective rod-screening method used for SuperFGD, with demonstrated throughput (~6–15 s/cube), non-contact metrology, and a path to full automation via dual robotic arms. The 48-group strategy that converts hole-position scatter into usable bins is a practical engineering contribution for detectors larger than SuperFGD (e.g., DUNE near detectors). Strengths include quantified reproducibility (Fig. 9), an independent validation set (Sec. 4.2), and public demonstration videos. The absence of an end-to-end fiber-insertion test on robot-sorted batches is a limitation of scope rather than an internal contradiction, but it leaves the central usability claim partially unclosed.

major comments (2)

Sec. 5.1–5.2 and Fig. 14: The claim that 48-way grouping yields fiber-insertable rows rests solely on geometric clearance (1.5 mm hole vs 1.0 mm fiber) and SuperFGD rod-screening heritage. Within-group scatter is stated to be <500 μm, yet no rods or fibers are threaded through multi-cube rows drawn from a single category. A modest insertion test (or explicit Monte-Carlo accumulation of residual misalignments along a 192-cube row) is needed to close this load-bearing assumption.
Sec. 3.8: Classification thresholds (|x,y_hole−2.92 mm|, |L−10.215 mm|, P_bump) are chosen empirically to match the manual rod pass/fail distributions. While the measurement chain itself is independent, the paper should state more clearly that the cuts are tuned to the SuperFGD rod method rather than derived from first-principles fiber-insertion tolerance, and should quantify sensitivity of the 3.1% reject rate to modest threshold shifts.

minor comments (5)

Fig. 9 caption and text: clarify whether the 10 μm figure is the mean of the per-cube standard deviations or the RMS of all repeated measurements; both are useful.
Sec. 2.3: the 0.28 mm platform positional error from the 0.05° stepper is quoted but never folded into the final uncertainty budget; a short sentence would help.
Sec. 4.1: the caliper comparison (σ = 39.8 μm) is informative; note that the soft reflective layer may systematically bias contact measurements, so the discrepancy is expected rather than a system flaw.
References [15] and [16] point to Zenodo videos; ensure DOIs remain stable and that the videos are permanently archived.
Minor typographical issues: “Semiautomatic” vs “semi-automatic” consistency; occasional missing spaces after periods in the arXiv text.

Circularity Check

1 steps flagged

No significant circularity: measurement and grouping claims are independent of the SuperFGD rod method; only classification thresholds were empirically tuned to match manual labels.

specific steps

fitted input called prediction [Sec. 3.8 (Cube Classification) and Sec. 4.2 (Test Run with Manually Screened Cubes)]
"These classification criteria were determined empirically from the parameter distributions of cubes manually screened as acceptable or defective using stainless-steel rods. Thresholds were chosen to classify clearly separated regions as "acceptable" or "defective," while the overlapping region was conservatively assigned to the "reinspection required" category. ... As summarized in Table 1, the system showed over 80% agreement with the manual classification results for both acceptable and defective cubes."

The numerical cuts (|x_hole-2.92 mm|>0.21 mm, P_bump>0.7, etc.) are fitted so that image-based labels reproduce the manual rod pass/fail distributions; the subsequent >80% agreement on a held-out manual set is therefore partly a consistency check of that calibration rather than a fully independent prediction. The circularity is mild and ordinary for a QC classifier; the measurement chain itself (contour/Hough/fit + linear lighting correction) remains independent of the rod method.

full rationale

This is an instrumentation/methods paper, not a first-principles derivation. The load-bearing claims (10 μm reproducibility after lighting/alignment correction, >80% agreement with an independent manually labeled set, 3.1% reject rate under 48-way hole-position grouping with within-group scatter <500 μm) are measured outputs of the camera/robot pipeline, not quantities forced by definition or by a self-citation chain. The only mild circularity is ordinary threshold calibration: Sec. 3.8 states that the |x_hole-2.92 mm| etc. cuts were chosen empirically from the parameter distributions of cubes already labeled acceptable/defective by stainless-steel rods, and Sec. 4.2 then scores the prototype against a separate manually labeled sample. That is standard supervised cut-setting, not a prediction that reduces to its inputs by construction. The geometric fiber-insertion argument (1.5 mm hole vs 1.0 mm fiber, SuperFGD heritage) is an external engineering assumption, not an internal circular step. No uniqueness theorem, ansatz smuggled via self-citation, or renaming of a known result is present. Score 1 reflects only the minor empirical-threshold dependence; the central experimental results stand independently.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central performance claims rest on standard imaging math, domain tolerances from SuperFGD fiber geometry, and several empirical thresholds and linear correction coefficients fitted on calibration cubes. No new physical entities are postulated; free parameters are the QC cut values and per-configuration lighting/alignment slopes.

free parameters (5)

Hole-position defective cut |x,y_hole - 2.92 mm|
Empirical threshold (0.21 mm defective, 0.19 mm reinspection) chosen from distributions of manually rod-screened cubes (Sec. 3.8), not derived from first principles.
Cube-size reinspection cut |L - 10.215 mm| > 0.115 mm
Nominal size and tolerance set empirically from manual-screened cube distributions (Sec. 3.8).
Protrusion index thresholds P_bump > 0.7 / 0.5
Brightness-fraction cuts for classifying surface protrusions vs dust, set by hand from labeled examples (Sec. 3.5, 3.8).
Per-configuration linear correction slope and intercept
192 camera×platform×orientation configs fitted against a chosen reference on 16 cubes (Sec. 3.7); coefficients are free parameters of the measurement model.
48-group bin boundaries on X/Y hole positions
Chosen to distribute cubes approximately evenly while keeping within-bin span <500 μm (Sec. 5.2, Fig. 14).

axioms (5)

domain assumption Acceptable per-cube hole deviation for SuperFGD-scale stacks is well below 0.5 mm/√N (≈36 μm for N=192) under independent Gaussian accumulation.
Stated in Introduction as the design driver for QC; used to motivate precision targets.
domain assumption Manual stainless-steel rod screening is a valid practical reference for pass/fail labels despite subjectivity and neighbor dependence.
Prototype validation (Sec. 4.2, Table 1) treats rod outcomes as ground truth for agreement rates.
domain assumption Within-group hole variation <500 μm implies fibers can be inserted through full rows of same-group cubes.
Sec. 5.2 asserts feasibility from geometric clearance without reporting insertion tests on sorted batches.
standard math OpenCV contour, Hough line/circle, and least-squares circle fit on 16 contour samples yield unbiased hole centers after the linear lighting/alignment correction.
Standard computer-vision pipeline (Secs. 3.2–3.7); correctness assumed after empirical residual reduction to 10 μm.
ad hoc to paper Passive gravity seating on textured PLA platforms plus 0.05° stepper accuracy is sufficient for imaging without active fixturing.
Mechanical design choice (Sec. 2.3); supported by observed roll-failure rate <0.5% but not independently proven for long-term wear.

pith-pipeline@v1.1.0-grok45 · 16156 in / 3448 out tokens · 37415 ms · 2026-07-13T21:47:03.587988+00:00 · methodology

0 comments

Cite this review

Pith. "Pith review of Semiautomatic dimensional screening of plastic scintillator cubes using image analysis and robotics." pith.science (2026). https://pith.science/paper/FYQ73SNL

@misc{pith2026260326732,
  author       = {Pith},
  title        = {Pith review of: Semiautomatic dimensional screening of plastic scintillator cubes using image analysis and robotics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FYQ73SNL}},
  note         = {Machine review of arXiv:2603.26732}
}

Share X Bluesky LinkedIn Reddit HN

read the original abstract

Large-scale particle physics detectors often contain millions of repeated components, making precise and efficient quality control essential. We have developed a semiautomatic system for dimensional screening of 1 cm$^3$ plastic scintillator cubes for their potential use in future neutrino detectors. The system employs a motorized rotating stage, six high-resolution cameras, and image analysis software to measure cube size, surface protrusions, and the positions of holes for wavelength-shifting fibers used in optical readout. Based on these measurements, each cube is automatically classified as either acceptable or defective. We constructed and validated a prototype system, achieving a measurement precision of 10 $\mu$m and over 80% consistency with manual screening. To enable classification of cubes into 48 groups based on hole positions while preserving their orientation, we introduced a 6-axis robotic arm. The completed system achieved a rejection rate of 3.1%. Our approach contributes to scalable, precise, and efficient quality control for future large-scale particle physics detectors.

discussion (0)

Reference graph

Works this paper leans on

16 extracted references · 2 canonical work pages

[1]

ATLAS Collaboration, JINST3, S08003 (2008)

2008
[2]

CMS Collaboration, JINST3, S08004 (2008)

2008
[3]

Abe et al.(Hyper-Kamiokande Proto-Collaboration), arXiv:1805.04163 (2018) – 16 –

K. Abe et al.(Hyper-Kamiokande Proto-Collaboration), arXiv:1805.04163 (2018) – 16 –

Pith/arXiv arXiv 2018
[4]

Abe et al.(T2K Collaboration), Nucl

K. Abe et al.(T2K Collaboration), Nucl. Instrum. Meth. A659106 (2011)

2011
[5]

Abe et al., arXiv:1901.03750 (2019)

K. Abe et al., arXiv:1901.03750 (2019)

arXiv 1901
[6]

Blondel et al., JINST13P02006 (2018)

A. Blondel et al., JINST13P02006 (2018)

2018
[7]

Kikawa et al., Nucl

T. Kikawa et al., Nucl. Instrum. Meth. A1080, 170616 (2025)

2025
[8]

Abe et al., arXiv:2603.14921 (2026)

S. Abe et al., arXiv:2603.14921 (2026)

arXiv 2026
[9]

Dergacheva et al., Nucl

A. Dergacheva et al., Nucl. Instrum. Meth. A104111 (2022)

2022
[10]

Giganti et al., CERN-SPSC-2020-008, SPSC-SR-267 (2020)

C. Giganti et al., CERN-SPSC-2020-008, SPSC-SR-267 (2020)

2020
[11]

Abi et al.(DUNE Collaboration), arXiv:2002.02967 (2020)

B. Abi et al.(DUNE Collaboration), arXiv:2002.02967 (2020)

arXiv 2002
[12]

G. Bradski. Dr. Dobb’s Journal of Software Tools (2000)

2000
[13]

OpenCV, Open Source Computer Vision Library, https://opencv.org
[14]

Ballard, Pattern Recognition 13, 2 (1981)

D.H. Ballard, Pattern Recognition 13, 2 (1981)

1981
[15]

Operation of the prototype inspection system for plastic scintillator cubes

T. Kikawa, "Operation of the prototype inspection system for plastic scintillator cubes" Zenodo, https://doi.org/10.5281/zenodo.18885041 (2026)

work page doi:10.5281/zenodo.18885041 2026
[16]

Operation of the completed inspection system incorporating a robotic arm

T. Kikawa, "Operation of the completed inspection system incorporating a robotic arm" Zenodo, https://doi.org/10.5281/zenodo.19108914 (2026) – 17 –

work page doi:10.5281/zenodo.19108914 2026

[1] [1]

ATLAS Collaboration, JINST3, S08003 (2008)

2008

[2] [2]

CMS Collaboration, JINST3, S08004 (2008)

2008

[3] [3]

Abe et al.(Hyper-Kamiokande Proto-Collaboration), arXiv:1805.04163 (2018) – 16 –

K. Abe et al.(Hyper-Kamiokande Proto-Collaboration), arXiv:1805.04163 (2018) – 16 –

Pith/arXiv arXiv 2018

[4] [4]

Abe et al.(T2K Collaboration), Nucl

K. Abe et al.(T2K Collaboration), Nucl. Instrum. Meth. A659106 (2011)

2011

[5] [5]

Abe et al., arXiv:1901.03750 (2019)

K. Abe et al., arXiv:1901.03750 (2019)

arXiv 1901

[6] [6]

Blondel et al., JINST13P02006 (2018)

A. Blondel et al., JINST13P02006 (2018)

2018

[7] [7]

Kikawa et al., Nucl

T. Kikawa et al., Nucl. Instrum. Meth. A1080, 170616 (2025)

2025

[8] [8]

Abe et al., arXiv:2603.14921 (2026)

S. Abe et al., arXiv:2603.14921 (2026)

arXiv 2026

[9] [9]

Dergacheva et al., Nucl

A. Dergacheva et al., Nucl. Instrum. Meth. A104111 (2022)

2022

[10] [10]

Giganti et al., CERN-SPSC-2020-008, SPSC-SR-267 (2020)

C. Giganti et al., CERN-SPSC-2020-008, SPSC-SR-267 (2020)

2020

[11] [11]

Abi et al.(DUNE Collaboration), arXiv:2002.02967 (2020)

B. Abi et al.(DUNE Collaboration), arXiv:2002.02967 (2020)

arXiv 2002

[12] [12]

G. Bradski. Dr. Dobb’s Journal of Software Tools (2000)

2000

[13] [13]

OpenCV, Open Source Computer Vision Library, https://opencv.org

[14] [14]

Ballard, Pattern Recognition 13, 2 (1981)

D.H. Ballard, Pattern Recognition 13, 2 (1981)

1981

[15] [15]

Operation of the prototype inspection system for plastic scintillator cubes

T. Kikawa, "Operation of the prototype inspection system for plastic scintillator cubes" Zenodo, https://doi.org/10.5281/zenodo.18885041 (2026)

work page doi:10.5281/zenodo.18885041 2026

[16] [16]

Operation of the completed inspection system incorporating a robotic arm

T. Kikawa, "Operation of the completed inspection system incorporating a robotic arm" Zenodo, https://doi.org/10.5281/zenodo.19108914 (2026) – 17 –

work page doi:10.5281/zenodo.19108914 2026

Semiautomatic dimensional screening of plastic scintillator cubes using image analysis and robotics

Core claim §

What carries the argument

Load-bearing premise §

What would settle it §

discussion (0)

Core claim

Load-bearing premise

What would settle it