{"id":"c4559719-540e-42b4-a5df-590c4a0b0630","arxiv_id":"2411.13703","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"A common optic was built that simultaneously delivers the LPBF laser and images the melt pool at 50 µm MWIR resolution, and its video signatures correlate with micro-CT porosity down to 4.3 µm.","lead":"This paper introduces aperture division multiplexing, a lens design that lets one optic both focus a 3D printing laser and image the melt pool in infrared. On a test metal printer, video signatures were matched to micro-CT scans and flagged pores as small as 4.3 micrometers, though thresholds were tuned on the same data.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Detection rates are in-sample: thresholds and dilation counts chosen on the same artifact, and two-step dilation to ~250 µm makes '4.3 µm detection' a proximity effect that needs held-out validation.","rationale":"The reader's weakest assumption identifies the same load-bearing concern: the detection probabilities are fitted in-sample. My analysis refines this by quantifying why the in-sample fit is particularly problematic: the dilation protocol expands alarms to ~250 µm, so the reported detection of 4.3–50 µm pores does not demonstrate pore-scale localization. The ADM optical contribution is independently supported by measured MTF, encircled-energy spot size, and tolerance analysis, so the issue is squarely with the statistical claim, not the hardware. A held-out split or second build with fixed a priori criteria would directly settle whether the 77–80% detection rate generalizes. The reader recommended CONDITIONAL; this concern reinforces that verdict rather than moving it to accept or reject.","tokens_in":29061,"tokens_out":3855,"duration_ms":38985,"concrete_test":"Split the artifact's layers into a training half and a test half. Fix all thresholds (4000 counts, 8 ms, 64 ms, 0.1 ms^-1, 5000 counts, 6000 counts) and the dilation count on the training half alone, then evaluate detection and false-alarm rates on the held-out half. If held-out detection of resolved voids drops below ~60% or false-alarm rates exceed ~20%, the reported 77–80% rates are in-sample overfits. Additionally, report the median centroid distance between each detected pore and the nearest alarm before dilation; if this distance is much larger than 50 µm, the 'detection' of fine pores is a proximity effect of dilation, not spatial localization.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim—77–80% of resolved voids detected with 3–14% false alarms—rests entirely on alarm criteria and dilation counts selected by inspecting this single print. Section 3.3 states the 4000-count threshold was 'chosen from a casual inspection of the video data,' and Section 3.5 applies dilation post hoc to improve detection. Because the same data are used both to choose these parameters and to compute detection statistics, the reported rates are optimism-biased upper bounds on expected performance. The pore-alarm matching protocol then counts a pore as detected if it falls inside an alarm after two or three dilations of 50-µm signature voxels; two dilations expand an alarm to a 250-µm-scale neighborhood. For a 4.3–50 µm pore, this means 'detected' often means 'located within a large dilated blob,' not that the pore is resolved or localized. The paper does not report confidence intervals, cross-validation, or the centroid distance between pores and pre-dilation alarms. The ADM optic itself is credibly characterized (77 µm laser spot, MTF, tolerance analysis), so the concern is not about the hardware but about whether the statistical evidence supports the qualification promise. Until fixed thresholds and dilation counts are evaluated on a held-out build, or at least with cross-validation within this build, the 77–80% numbers do not establish predictive performance for unseen LPBF parts.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript presents aperture division multiplexing (ADM) as an optical architecture for laser powder bed fusion (LPBF), in which one lens provides two optical paths: one for focusing the 1.07 µm process laser and one for high-resolution mid-wave infrared monitoring. The authors design, build, and characterize an ADM lens, reporting a measured D86 laser spot of 77 µm and a measured slanted-edge MTF consistent with the optical model. They demonstrate the system on a custom LPBF testbed by printing a 5×5×6 mm 316L artifact, obtaining 1250 Hz MWIR video, and registering the video-derived process signatures to micro-CT ground truth with 4.3 µm voxels. From this single build, they report pore detection probabilities for time-above-threshold, maximum radiance, meltpool size, and cooling-rate signatures, claiming correlation with pores as small as 4.3 µm and detection rates around 77–80% after dilation at false-alarm rates of 3–14%.","tokens_in":29354,"tokens_out":6311,"duration_ms":50886,"significance":"The ADM optical concept is a meaningful and credible contribution to LPBF process monitoring. It directly addresses the known difficulties of on-axis monitoring through f-theta lenses, and the hardware claim is independently validated: the measured 77 µm laser spot closely matches the design encircled energy, the slanted-edge MTF is consistent with the as-built model, and the tolerance analysis supports manufacturability. The paper provides a technically detailed and reproducible optical design. The pore-detection statistics, however, are not yet predictive evidence for unseen builds: the thresholds and dilation counts are selected on the same artifact used to compute the detection probabilities, and the '4.3 µm pore' claim is a proximity statement relative to dilated alarm regions rather than resolved localization. The work therefore stands as a strong proof-of-concept whose central quantitative claim needs reframing or held-out validation before it can support the qualification promise in the abstract.","major_comments":[{"comment":"The detection probabilities in Figs. 6 and 7 are in-sample estimates because the alarm thresholds (4000 counts, 8 ms, 64 ms, 0.1 ms^-1, 5000 counts, 6000 counts) and the dilation counts were selected from the same artifact used to compute the statistics. Section 3.3 states directly that the 4000-count threshold 'was chosen from a casual inspection of the video data,' and Section 3.5 applies dilation after the fact to improve detection. With the same data used both to set these parameters and to evaluate them, the reported 77–80% detection with 3–14% false alarms are optimism-biased upper bounds for a new build. The manuscript should either evaluate fixed criteria on a held-out build or use a cross-validation scheme within this build, report the parameter selection protocol, and provide confidence intervals for the detection probabilities.","section":"§3.3, §3.5, Figs. 6–7"},{"comment":"The dilation procedure makes the '4.3 µm pore detection' claim misleading as stated. With 50 µm process-signature voxels, one dilation expands an alarm by one voxel in each direction and two dilations create a roughly 250 µm neighborhood around the original alarm; three dilations are still larger. A pore of diameter 4.3–50 µm is therefore counted as detected when it lies anywhere inside this dilated region, not when the pore is resolved or localized. The paper should report the correspondence before dilation (e.g., overlap or centroid distance between pores and undilated alarms) and should phrase the claim in terms of correlation at the scale of the alarm neighborhood unless such localization data are provided.","section":"§3.5, Figs. 6–7"},{"comment":"The generalization to LPBF qualification is not yet supported by the reported single-build experiment. The authors themselves note that balling is the predominant porosity source, that the process parameters are near the edge of the process window, and that one region of the artifact is 98.7% dense; only one material, one geometry, one parameter set, and one print are tested. This does not weaken the ADM optical demonstration, but the abstract's 'promise for qualification of LPBF component fatigue performance' should be scaled back to a proof-of-concept correlation in a balling-dominated artifact until keyhole and lack-of-fusion porosity regimes and additional builds are evaluated.","section":"§3.5 (limitations paragraph), §4"}],"minor_comments":[{"comment":"The notation '1.2− 2.4+ µm' is unclear; it should be '1.2–2.4 µm' or the intended spectral band should be stated explicitly.","section":"§2.2.1"},{"comment":"The text refers to MTF curves as appearing in Fig. 3b, but Fig. 3b is captioned as spot diagrams; the MTF discussion appears to correspond to Fig. 4b. Please correct the cross-reference.","section":"§3.1"},{"comment":"'has operands operands for' is a typo in §S2.1, and 'preformed' in §S2.4 should be 'performed'.","section":"§S2.1, §S2.4"},{"comment":"The definition of 'false alarm' would be clearer if the denominator were stated (alarms total versus alarm voxels total) and if the relationship between the quoted 6% and 14% false-alarm figures and the curves in Figs. 6–7 were made explicit.","section":"§3.5, Figs. 6–7"},{"comment":"The text 'Rayleigh-Tailor meltpool instabilities' should read 'Rayleigh-Taylor'.","section":"References, near [25]"}],"recommendation":"major_revision","confidential_remarks":"The paper's hardware contribution is solid and independently validated, and the writing is generally clear. My recommendation of major_revision rests on the gap between the in-sample pore-detection statistics and the predictive qualification claim; I believe this is fixable by adding held-out validation or by substantially softening the claim. No concerns about novelty disclosure."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, the hardware is the real story: aperture division multiplexing is a genuinely new way to get on-axis, high-resolution process views in LPBF, and the paper supports it with a measured 77 µm D86 laser spot, slanted-edge MTF, and a tolerance analysis. That part deserves to be taken seriously. Second, the pore-detection numbers in Figs. 6 and 7 are in-sample fits, and the abstract's \"4.3 µm\" claim outruns them. The 4000-count threshold is admitted to be chosen from casual inspection, dilation is applied post hoc, and no confidence intervals or cross-validation appear anywhere. With two or three dilations, an alarm is roughly a 250 µm blob; counting a 4.3 µm pore as detected if it falls inside such a blob is a proximity effect, not resolution. The paper is honest about some limitations (balling-dominated porosity, near-edge parameters) but does not flag this statistical circularity.\n\nWhat the paper does well, besides the lens, is the careful mutual-information alignment of video and micro-CT, and the open acknowledgment of the artifact's shortcomings. The writing is clear and the literature coverage is fair. This is not a flawed-from-the-start paper; it is a strong instrumentation contribution with a validation section that oversells itself.\n\nThe fix is straightforward in principle: fix thresholds and dilation counts a priori, or at least cross-validate within the build; report error bars; and report centroid distances between pores and pre-dilation alarms. Then the detection claim becomes something you can rely on. As it stands, the 77-80% detection with 3-14% false alarms should be read as optimistic upper bounds.\n\nWho is this for? People working on LPBF process monitoring and optical design for manufacturing will get value from the ADM architecture. The qualification claim needs another pass. A serious referee should be able to get this into shape with major revision; I would send it to review rather than desk-reject, with the statistical validation as the focus.","headline":"The ADM optic is a real and well-characterized contribution; the pore-detection rates are in-sample fits that outrun the data.","tokens_in":29906,"tokens_out":3101,"would_cite":true,"duration_ms":27059,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A single lens focuses the process laser and images melt-pool pores as small as 4.3 microns.","keywords":["laser powder bed fusion","aperture division multiplexing","process monitoring","porosity detection","mid-wave infrared","micro-CT","fatigue qualification","in-situ sensing"],"falsifier":"Print a separate component with a different geometry or process parameters, apply the exact thresholds and dilation counts reported here (e.g., 4000 counts, 8 ms and 64 ms, $0.1\\ \\mathrm{ms}^{-1}$, 5000 and 6000 counts), and compare the resulting alarms to micro-CT: if detection probability collapses to near chance or false alarms dominate, the in-sample statistics do not generalize.","tokens_in":28839,"feed_emoji":"🔬","tokens_out":8453,"duration_ms":79378,"temperature":0.7,"pith_summary":"The paper aims to show that aperture division multiplexing (ADM) — splitting one lens into separate optical paths for the process laser and for an infrared camera — can give laser powder bed fusion (LPBF) an unobstructed, on-axis view of the melt pool at 50-micron resolution. Using a production-representative testbed, the authors record high-speed mid-wave infrared video during a stainless-steel build and compare four process signatures against micro-CT ground truth. They report that voids as small as 4.3 microns, the size range implicated in fatigue-crack nucleation, are detected by low time-above-threshold or low cooling-rate alarms after modest spatial dilation, with false-alarm rates of a few percent to 14 percent. If this holds, in-process optical monitoring could qualify LPBF components for fatigue-limited service without relying on downstream computed tomography.","feed_headline":"One lens focuses the laser and sees 4.3-micron pores","feed_subtitle":"Aperture-division optics give laser powder bed fusion a live infrared view of the defects that limit fatigue life.","key_machinery":"The central object is the aperture division multiplexing (ADM) lens: a single optical assembly in which the clear aperture is split into a laser-delivery path and an imaging path, each decentered by 17.5 millimeters, so one optic simultaneously focuses the process beam and forms a diffraction-limited mid-wave infrared image of the build plane. A lens form with four positive calcium-fluoride elements and two negative fused-silica elements controls chromatic aberration across the 1.2–2.4-micron sensing band, and an f/2.3 cold-stop camera with an indium-antimonide detector records the scene. The argument is carried by this shared-aperture layout plus a signature-extraction pipeline that reduces each pixel's time series to alarms, then registers those alarms to micro-CT via mutual-information alignment and scores detection probability as a function of void size.","core_discovery":"The central claim is that aperture division multiplexing erases the usual trade-off between delivering a 1.07-micron process laser and imaging the melt pool: a single lens, with two decentered optical paths sharing calcium-fluoride and fused-silica elements, focuses the laser to a 77-micron spot while a mid-wave infrared camera looks through the same optic at 50-micron geometric resolution. On a custom LPBF testbed printing 316 stainless steel, the camera ran at 1250 frames per second with 0.7-microsecond exposures, and per-pixel process signatures — time above threshold, maximum radiance, meltpool area, and cooling rate — were registered to micro-CT voxels of 4.3 microns. Pore-by-pore statistics show that after two or three dilations of the alarm volume, roughly 77 to 80 percent of resolved voids are caught by low time-above-threshold or low cooling-rate alarms, rising to 98.9 percent for the high time-above-threshold signature at a 14 percent false-alarm cost. The paper concludes that this detection scale reaches below the minimum pore sizes reported to nucleate fatigue cracks in common LPBF alloys, establishing the promise of ADM for qualification of component fatigue performance.","pith_inferences":["The reported detection statistics are computed on the same build that was used to choose alarm thresholds, so the decisive follow-up is a blind test on an unseen component with thresholds fixed in advance.","Because the 4.3-micron floor is the micro-CT voxel size, the optical system may resolve even smaller pores; confirming this would require a higher-resolution ground truth such as synchrotron tomography.","The same shared-aperture idea could extend to synchronized multi-laser LPBF, where overlapping f-theta scan fields are normally limited to a narrow stripe; ADM-style split apertures might allow coordinated beams across the full build area.","Cooling-rate signatures, if they track local thermal history, may later be correlatable to residual stress and microstructure, though that correlation is not shown in this paper."],"forward_implications":["ADM provides coaxial optical access to the melt pool without the barrel distortion and chromatic aberration that f-theta lenses impose on monitoring wavelengths.","The demonstrated detection of voids at 4.3 microns reaches below the 34–52-micron pore sizes reported to nucleate fatigue cracks in Ti-6Al-4V, so alarm maps from this method could tighten fatigue-life bounds for LPBF parts.","Because the four process signatures respond to different phenomena (e.g., slow cooling versus sustained overheating), combining them is expected to outperform any single alarm channel.","The authors' identified upgrades — smaller detector pixels, a longer camera focal length, and alternative infrared materials — point toward roughly 10-micron feature resolution and frame rates approaching the 1.4-MHz exposure limit."],"supporting_citations":[{"why":"Reports 92 percent detection of pores with effective diameter near 160 microns, providing the prior best-case sensitivity that ADM must beat.","marker":"[43]"},{"why":"Shows LWIR microbolometer detection of flaws 100 microns and larger, establishing the baseline resolution of stationary off-axis thermal monitoring.","marker":"[44]"},{"why":"Demonstrates correlation of large (100-micron) porosities to a CT baseline with a hybrid camera-photodiode instrument, the closest prior coaxial result.","marker":"[45]"},{"why":"Reports 82 percent detection of lack-of-fusion defects at 100 microns but only one third of keyhole pores, quantifying the gap ADM targets.","marker":"[46]"},{"why":"Establishes reliable detection of 70-micron pores with two-color pyrometry, a prior limit on the finest optically detected voids.","marker":"[47]"},{"why":"Supplies the micro-CT resolution heuristics (voxel about one third of the smallest pore, part size limit) that motivate the need for an alternative to CT-based qualification.","marker":"[48]"},{"why":"Provides the rigorous statistical framework for predicting porosity probability from thermal features, which the present pore-by-pore analysis extends.","marker":"[66]"},{"why":"Documents meltpool temperatures near 4000 K and cooling rates near 40 K per microsecond, defining the dynamic-range and speed requirements the ADM camera must meet.","marker":"[75]"},{"why":"Supplies the mutual-information registration method used to align micro-CT voxels to the process-signature data space.","marker":"[105]"}],"fun_headline_variants":["One lens: laser focus and 4.3-micron pore vision","Aperture division multiplexing sees 4.3-micron pores live","Single optic monitors LPBF pores down to 4.3 microns","In-process vision catches 4.3-micron fatigue pores"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The detection probabilities treat alarm thresholds and dilation counts chosen by looking at this same test print as fixed criteria that will work on unseen builds; if those thresholds were tuned to this artifact, the reported rates do not yet prove predictive performance.","fun_headline_variants_meta":{"raw":{"variants":["One lens: laser focus and 4.3-micron pore vision","Aperture division multiplexing sees 4.3-micron pores live","Single optic monitors LPBF pores down to 4.3 microns","In-process vision catches 4.3-micron fatigue pores"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000197,"raw_usage":{"total_tokens":1394,"prompt_tokens":1006,"completion_tokens":388,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":622,"completion_tokens_details":{"reasoning_tokens":308}},"tokens_in":622,"tokens_out":388,"duration_ms":4445,"temperature":1.0,"reasoning_tokens":308,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T15:58:45.704401+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Print a separate component with a different geometry or process parameters, apply the exact thresholds and dilation counts reported here (e.g., 4000 counts, 8 ms and 64 ms, $0.1\\ \\mathrm{ms}^{-1}$, 5000 and 6000 counts), and compare the resulting alarms to micro-CT: if detection probability collapses to near chance or false alarms dominate, the in-sample statistics do not generalize.","supporting_citations":[{"cited_title":"Coeck, M","cited_arxiv_id":null,"evidence_quote":"Reports 92 percent detection of pores with effective diameter near 160 microns, providing the prior best-case sensitivity that ADM must beat."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the rigorous statistical framework for predicting porosity probability from thermal features, which the present pore-by-pore analysis extends."}],"review_version":1}