{"id":"556edfe4-7fb2-4825-9787-a407bc40bf52","arxiv_id":"2502.08088","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Using spectrally resolved thermal images, the authors retrieve both temperature and spectral emissivity during laser powder bed fusion, achieving ±28 K accuracy against melting-point benchmarks and correlating temperature signatures with porosity.","lead":"This paper shows that a remote-sensing technique called temperature emissivity separation can measure meltpool temperatures in laser powder bed fusion metal 3D printing to within about 28 K, without assuming a fixed surface emissivity. The authors validate it by melting metals with known melting points and link the resulting temperature histories to micron-scale pores detected by CT scans.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The two-temperature TES assumption that emissivity is invariant between paired frames is load-bearing for the layerwise temperature and porosity results, and the point-melting validation does not exercise the same emissivity drift.","rationale":"The reader's weakest-assumption analysis identifies exactly the point I consider most load-bearing: the two-temperature TES method assumes spectral emissivity is unchanged between paired measurements, while the LPBF environment and the paper's own Figure 1 show that emissivity can vary strongly with temperature and material state. This assumption is not merely a calibration detail; it underlies the layerwise temperatures used for cooling-rate trends and for the porosity detection that forms part of the central claim. The point-melting validation is strong evidence that TES can recover melting points accurately, but it does not transfer automatically to the dynamic cooling regime because the melting plateau supplies a nearly constant-temperature anchor that suppresses the emissivity-drift error. The paper's comparison among TES methods is qualitative and stops short of re-running the downstream analysis, so the concern is unresolved. I therefore agree with the reader's CONDITIONAL verdict and recommend no change to it. A re-analysis using the alternative TES variants on the same raw spectra would settle whether the assumption actually changes the reported porosity and cooling-rate results.","tokens_in":22017,"tokens_out":4405,"duration_ms":47748,"concrete_test":"Recompute the layerwise process signatures (integral, cooling rate, maximum temperature) and the Figure 7 porosity-detection curves from the stored raw spectra using the two-location and greybody TES variants instead of the two-time method, keeping the same alarm thresholds and ±1 layer vertical allowance. If the detection rates or false-alarm rates change materially relative to the two-time results, the emissivity-invariance assumption is load-bearing and the porosity claim must be restated as conditional on that assumption.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The layerwise temperature, cooling-rate, and porosity-detection results all rely on the two-temperature TES method, whose core assumption is stated in Section 2.4 as 'assuming that emissivity did not appreciably change between the two measurements, i.e., it is only a weak function of temperature.' In LPBF, however, the two measurements are taken while material is rapidly cooling and can cross solid, liquid, and oxide states, each with different radiative character, as the paper itself notes in the introduction. Figure 1, reproduced from the authors' own literature review, shows that spectral emissivity of stainless steel is not a weak function of temperature in the relevant wavelength range. If emissivity changes between the two spectra, the fitted temperatures T1 and T2 are biased, and the derived cooling rates, time-at-temperature integrals, and porosity alarms inherit that bias. The point-melting validation provides partial support, but melting plateaus are quasi-isothermal and therefore do not reproduce the large-temperature-interval emissivity drift present during rapid cooling from above 2000 K to below 1000 K. The comparison of TES variants in Figure S2 is qualitative and is not carried through to the process-signature or porosity analysis, so the impact of this assumption on the central claims remains unquantified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper describes a temperature-emissivity separation (TES) approach for laser powder bed fusion (LPBF) thermography. A custom imaging spectrometer, integrated with an LPBF testbed, acquires spectrally resolved radiance in 30 spectral bands over about 1.2 to 2.4 µm. Three TES algorithms—the two-temperature method, a two-location method, and a greybody method—are compared. The two-temperature method is used to retrieve layerwise temperature series during printing of a 316L test artifact. Point-melting experiments on five pure metals and three alloys serve as an external validation, with a reported maximum error of 28 K over a 933–1940 K range. From the layerwise data the authors extract three process signatures (time-at-temperature integral, cooling rate, and maximum temperature), and correlate them with micro-CT porosity, reporting detection of 71% of pores larger than 20 µm at a 5% false-alarm rate with a ±1 layer allowance. The porosity study is explicitly described as exploratory.","tokens_in":22164,"tokens_out":8376,"duration_ms":70369,"significance":"The external melting-point benchmarks are the paper's strongest feature: retrieved melting/liquidus temperatures are compared with tabulated values, so the central temperature-accuracy claim is not circular. If the ±28 K fidelity holds over the full retrieval range, this would be a substantial improvement over constant-emissivity and two-color approaches in LPBF, and the demonstration of in situ detection of pores below 20 µm would be industrially relevant. The manuscript is also candid about the exploratory character of the porosity analysis and about known unknowns such as the channel sensitivity volume. The main reservations concern the lack of a quantitative bound on the two-temperature emissivity-invariance assumption in the rapid-cooling regime, and the apparent in-sample selection of porosity thresholds; both need clarification before the layerwise temperature and porosity results can be fully relied upon.","major_comments":[{"comment":"The two-temperature TES method is load-bearing for all layerwise results, but the manuscript does not state how the two time frames are selected for each retrieval. Watson's method assumes emissivity is unchanged between two measurements taken at different temperatures; in LPBF, the material can pass through solid, liquid, and oxide states with strongly temperature-dependent emissivity, as Fig. 1 itself shows. The point-melting validation in §3.1 only tests quasi-isothermal melting plateaus, so it does not bound the bias of cooling-curve temperatures used for the process signatures in Fig. 6 and the porosity analysis in Fig. 7. Please specify the frame-pairing rule, report the typical temperature difference between paired frames, and quantify the sensitivity of the retrieved temperatures and of the three process signatures to plausible emissivity drift (for example, by comparing against a two-location or greybody retrieval on the same layerwise data, with error propagation rather than visual comparison as in Fig. S2).","section":"§2.4, §3.2, Fig. 5"},{"comment":"The reported pore-detection probabilities are in-sample estimates because the threshold values and the ±1/±2/±3 layer deviation allowances are evidently chosen using the same micro-CT labels that define detection success. The text gives no threshold values or a priori selection rule, and the statement that \"setting the threshold for roughly equal rates of pore detection\" is used for the cooling-rate panel indicates post hoc tuning. Without a validation split, bootstrap, or pre-registered thresholds, the headline figure of 71% detection of pores >20 µm at a 5% false-alarm rate is optimistic. Please state how each threshold was chosen, and either fix thresholds from a training subset or evaluate the detection curves with cross-validation.","section":"§3.4, Fig. 7"},{"comment":"The instrument calibration is fit to a blackbody over 1023–1423 K and then extrapolated to the full 933–1940 K measurement range, but neither the functional form of the correction function nor the fit residuals are reported. The external melting-point benchmarks provide some end-to-end validation, but they sample only specific transition temperatures; the layerwise traces in Fig. 5 use intermediate temperatures and rapid transients that are not independently benchmarked. Please provide the calibration function, residuals, and an uncertainty estimate, and propagate this uncertainty into the retrieved temperature and process-signature data.","section":"§S1 (radiance calibration)"}],"minor_comments":[{"comment":"The 316 stainless steel liquidus values are reversed between the text and the table; the text says the measured inflection is 1723 K against a nominal 1739 K, while the table lists Reference 1723 K and Measured 1739 K. Please reconcile.","section":"Table 1 and §3.1"},{"comment":"The conclusion states \"temperature range of 633 to 1940 K\"; Table 1 shows the lowest benchmark is aluminum at 933 K, so \"633\" appears to be a typo.","section":"§5"},{"comment":"There are several typos: \"Plank's Law\" should be \"Planck's law\", \"Wein Approximation\" should be \"Wien approximation\", and \"spacebourne\" should be \"spaceborne\".","section":"Abstract and §1"},{"comment":"The misspelling \"Sicpy.optimize\" should be \"SciPy.optimize\"; elsewhere in the paper \"SciPy\" is spelled correctly, so please unify.","section":"§2.4"},{"comment":"The paragraph beginning \"It should be noted that this is effectively a worst-case assumption\" conflates the conservative volume-equivalent-pore convention with the optimistic ±layer deviation allowance; these two effects should be separated and explained clearly.","section":"§3.4"},{"comment":"The phrase \"where the laser powder is reduced to 90 W\" should read \"where the laser power is reduced to 90 W\".","section":"§3.3"},{"comment":"Alloy measurements are reported without uncertainty; the text explains this by the single-point nature of liquidus, but a repeated-measures or fitting-uncertainty estimate would strengthen the comparison.","section":"Table 1"}],"recommendation":"major_revision","confidential_remarks":"To the editor: the paper's thermal validation is credible because it is anchored to external melting-point data, and the porosity study is framed as exploratory, which is appropriate. The revisions I request—documenting the two-temperature frame-pairing rule, providing a sensitivity analysis of emissivity drift, and moving the porosity threshold selection out of the test set—are feasible within the manuscript's scope and should materially strengthen it. I also encourage the authors to release the calibration data and analysis code, as the method's details are otherwise hard to reproduce. The 316SS text/table discrepancy and the \"633 K\" typo should be fixed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the point-melting validation is the real contribution. They show three TES algorithms retrieve temperatures within 28 K of known melting and liquidus points across a 1000 K range, and that is an honest, checkable result. The layerwise cooling-rate and porosity-detection results are more provisional than the abstract suggests. If you read one thing, read Table 1 and the calibration section.\n\nWhat is new: this is the first application of TES to LPBF with a purpose-built imaging spectrometer, benchmarked externally. The algorithms themselves are reused from remote sensing—Watson's two-temperature method, the two-location method, and the greybody method—but the engineering adaptation is nontrivial and clearly described. The instrument description is thorough enough to reproduce, which matters.\n\nWhere it is soft: first, the two-temperature method carries the layerwise analysis, and it assumes emissivity does not change between the two time frames. That is load-bearing because the two frames span large temperature drops from above 2000 K to below 1000 K, crossing solid, liquid, and oxide states. The paper's own Fig. 1 shows spectral emissivity of stainless steel is not a weak function of temperature in this range. The point-melting validation does not exercise this because melting plateaus are quasi-isothermal. So the absolute temperatures in Fig. 5 and the cooling rates in Fig. 6 may carry a bias that is not quantified. Second, the porosity detection numbers—71% of pores larger than 20 µm at a 5% false alarm rate with a ±1 layer allowance—are in-sample. The thresholds and the vertical allowance are tuned on the same artifact used for evaluation, so those metrics are optimistic. The paper does call this \"provisional\" and gives a fair worst-case discussion, but the abstract and conclusion state it more firmly than the analysis supports. Third, the blackbody calibration is fit over 1023–1423 K and extrapolated to 933–1940 K; that is a real extrapolation, and the alloy liquidus points have no error bars. Fourth, no data or code were shipped, so independent checks are limited.\n\nAm I skeptical of the central claim? No. The melting-point benchmark is external, and ±28 K accuracy over that range is plausible. The soft spots are in the downstream uses of the temperature retrieval, not the retrieval itself.\n\nRecommendation: send it to review. The validation deserves referee time, and the emissivity-drift and porosity concerns are addressable with an independent test set and uncertainty propagation. I would want the authors to either ship data and code or provide a second artifact as a holdout.","headline":"Useful engineering validation of TES for LPBF meltpool temperature, with a solid ±28 K point-melting benchmark; the layerwise and porosity claims are weaker than the abstract implies.","tokens_in":22773,"tokens_out":1937,"would_cite":true,"duration_ms":18777,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Temperature emissivity separation brings LPBF temperature readings to ±28 K accuracy, and time-at-temperature signatures flag 71% of pores above 20 µm at a 5% false-alarm rate.","keywords":["temperature emissivity separation","laser powder bed fusion","spectral emissivity","imaging spectrometer","meltpool temperature","porosity detection","process monitoring","cooling rate"],"falsifier":"A decisive check would be to instrument an LPBF build with fine thermocouples or a known-phase reference (such as the melting plateau of a pure metal) while using the two-temperature method, and to compare retrieved temperatures against the reference specifically during intervals when emissivity is expected to change rapidly; if the bias exceeds the claimed ±28 K in those intervals, the constant-emissivity assumption fails under real process conditions.","tokens_in":21708,"feed_emoji":"🔥","tokens_out":2241,"duration_ms":470410,"temperature":0.7,"pith_summary":"This paper argues that the main obstacle to accurate optical temperature measurement in laser powder bed fusion (LPBF) is unknown spectral emissivity, and that temperature-emissivity separation (TES) can remove that obstacle by solving for temperature and emissivity together from spectrally resolved radiance. Using a custom imaging spectrometer on an LPBF testbed, the authors show that TES retrieves melting and liquidus temperatures of pure metals and alloys to within ±28 K across a 1000 K range, a precision they contrast with typical wideband and two-color pyrometry. In a printed test artifact, TES-derived temperature fields reveal cooling-rate evolution as heat accumulates, and a time-at-temperature process signature detects 71% of pores larger than 20 µm at a 5% false-alarm rate when a ±1 layer vertical allowance is used. If the approach holds, it would give LPBF a path from raw radiance to trustworthy temperature fields, enabling deterministic process tuning and in situ defect detection during printing.","feed_headline":"One method reads meltpool temperature to ±28 K and flags pores down to 20 µm","feed_subtitle":"Temperature emissivity separation turns radiance into trustworthy temperature fields for laser powder bed fusion quality control.","key_machinery":"The central mechanism is temperature-emissivity separation (TES), a set of algorithms that jointly estimate temperature and spectral emissivity from spectrally resolved radiance, applied to an imaging spectrometer that records ~30 spectral bands from each of 19 spatial channels viewing the meltpool. The key identity is the underdetermined system\n$L_n = \\varepsilon_n B_n(T)$; the two-temperature method resolves it by measuring the same location at two times and assuming $\\varepsilon_n$ is unchanged between them, the two-location method makes the same assumption between two neighboring channels, and the greybody method sets two spectral bands to equal emissivity. The instrument couples a custom LPBF testbed to the spectrometer through an aperture-division multiplexing lens, and temperature-emissivity fitting is performed by least-squares optimization. This machinery converts raw radiance into temperature fields whose fidelity is then used to derive process signatures such as maximum temperature, time-at-temperature integral, and cooling rate, which are correlated to porosity measured by micro-CT.","core_discovery":"The paper claims that a temperature field in laser powder bed fusion can be recovered without assuming a known or constant emissivity by applying temperature-emissivity separation (TES) to spectrally resolved radiance measurements. The measured radiance in each spectral band is modeled as\n$L_n = \\varepsilon_n B_n(T)$, where $B_n(T)$ is Planck's law and $\\varepsilon_n$ is the spectral emissivity; with $n$ bands this is $n$ equations for $n+1$ unknowns, so the authors break the underdetermination by assuming invariance along one axis of the spatial-spectral-temporal dataset. They implement three such methods: the two-temperature method, which assumes emissivity is unchanged between two times; the two-location method, which assumes emissivity is constant between two spatial channels; and the greybody method, which sets two spectral emissivities equal. In point-melting experiments on titanium, iron, nickel, copper, aluminum, Ti-6Al-4V, 316 stainless steel, and Inconel 718, the recovered melting and liquidus temperatures match reference values with a maximum error of 28 K over a 633–1940 K range. During printing of a 316 stainless steel artifact, the two-temperature method yields layerwise temperature traces showing a decrease in cooling rate as the build progresses, and thresholding the time-at-temperature integral produces detection of 71% of pores larger than 20 µm at a 5% false-alarm rate with a ±1 layer vertical allowance.","pith_inferences":["The retrieved spectral emissivity itself is a physically meaningful process signal: because emissivity differs between powder, liquid, and oxide states, its retrieved value could flag the material state at the meltpool without waiting for a thermal signature to cross a threshold.","The porosity-detection results likely understate the instrument's sensitivity, since the paper counts any pore in an alarm cylinder as one volume-equivalent pore and grants no lateral allowance; validating the lateral volume of sensitivity could improve reported detection rates.","A testable extension would be to determine the minimum number of spectral bands needed for robust TES, since the paper notes the detector could theoretically support more than 10,000 spatial channels, suggesting a path to wide-field, high-speed monitoring.","Because the two-temperature method assumes emissivity is constant between the two observation times, adapting TES to rapid solidification events may require shorter time baselines or a model that lets emissivity evolve between the two frames."],"forward_implications":["LPBF temperature measurements can be made without assuming a constant or textbook emissivity, so radiometric readings no longer carry a large unknown systematic bias.","Time-at-temperature and cooling-rate signatures can be computed from trusted temperature fields, allowing defect-prone fusion conditions to be identified by layer and by spatial channel.","Deterministic process tuning becomes possible: scan parameters can be adjusted between parts based on measured temperature history rather than on indirect radiance levels.","The same TES approach could support closed-loop control, where measured temperature deviations feed back into laser power or scan speed during a build."],"supporting_citations":[{"why":"Provides the two-temperature TES method, the core algorithm used for the layerwise temperature retrieval and porosity signatures.","marker":"[40]"},{"why":"Provides the greybody emissivity method, one of the three TES algorithms compared in the validation experiments.","marker":"[61]"},{"why":"Supplies the benchmark study of two-color pyrometry accuracy that the paper's ±28 K TES result is contrasted against.","marker":"[36]"},{"why":"Gives the prior pyrometry-to-porosity detection result (70 µm pores) that the paper's 20 µm detection capability extends.","marker":"[39]"},{"why":"Provides the spectral emissivity data of stainless steel used to show the wavelength and temperature dependence that motivates TES.","marker":"[26]"},{"why":"Describes the aperture-division multiplexing optical monitoring approach that the imaging spectrometer integration builds on.","marker":"[21]"},{"why":"Supplies the optimization routines and peak-finding tools used to fit temperature-emissivity and compute cooling-rate signatures.","marker":"[62]"}],"fun_headline_variants":["TES gives LPBF meltpool temperatures to ±28 K and spots 20 µm pores","Meltpool thermometry to ±28 K and 20 µm pore detection via TES","Accurate LPBF temperature maps: ±28 K, 20 µm pores found","Temperature emissivity separation enables ±28 K accuracy and 20 µm defect detection","One TES method nails LPBF meltpool temperature to ±28 K and sees 20 µm pores"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The layerwise temperature data rely on the assumption that the spectral emissivity of the observed material does not change between the two time frames used in the two-temperature fit, even though the material crosses solid, liquid, and oxide states during cooling.","fun_headline_variants_meta":{"raw":{"variants":["TES gives LPBF meltpool temperatures to ±28 K and spots 20 µm pores","Meltpool thermometry to ±28 K and 20 µm pore detection via TES","Accurate LPBF temperature maps: ±28 K, 20 µm pores found","Temperature emissivity separation enables ±28 K accuracy and 20 µm defect detection","One TES method nails LPBF meltpool temperature to ±28 K and sees 20 µm pores"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000736,"raw_usage":{"total_tokens":3396,"prompt_tokens":1156,"completion_tokens":2240,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":772,"completion_tokens_details":{"reasoning_tokens":2130}},"tokens_in":772,"tokens_out":2240,"duration_ms":32864,"temperature":1.0,"reasoning_tokens":2130,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T10:29:51.667829+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A decisive check would be to instrument an LPBF build with fine thermocouples or a known-phase reference (such as the melting plateau of a pure metal) while using the two-temperature method, and to compare retrieved temperatures against the reference specifically during intervals when emissivity is expected to change rapidly; if the bias exceeds the claimed ±28 K in those intervals, the constant-emissivity assumption fails under real process conditions.","supporting_citations":[{"cited_title":"Watson, Two-temperature method for measuring emissivity, Remote Sensing of Environment 42 (2) (1992) 117 – 121","cited_arxiv_id":null,"evidence_quote":"Provides the two-temperature TES method, the core algorithm used for the layerwise temperature retrieval and porosity signatures."},{"cited_title":"Barducci, I","cited_arxiv_id":null,"evidence_quote":"Provides the greybody emissivity method, one of the three TES algorithms compared in the validation experiments."},{"cited_title":"Monier, F","cited_arxiv_id":null,"evidence_quote":"Supplies the benchmark study of two-color pyrometry accuracy that the paper's ±28 K TES result is contrasted against."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the prior pyrometry-to-porosity detection result (70 µm pores) that the paper's 20 µm detection capability extends."},{"cited_title":"Touloukian, D","cited_arxiv_id":null,"evidence_quote":"Provides the spectral emissivity data of stainless steel used to show the wavelength and temperature dependence that motivates TES."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Describes the aperture-division multiplexing optical monitoring approach that the imaging spectrometer integration builds on."}],"review_version":1}