{"id":"85619ba6-94d7-4445-b410-e3075bb77f03","arxiv_id":"1908.03431","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A regularized Fourier-plus-polynomial regression reconstructs 2D engine temperature fields from sparse rakes, and the integrated area averages differ from the standard sector-weighted averages by 0.5 to 2 K.","lead":"Engine temperature data from a few sparse thermocouple rakes are used to fit a smooth two-dimensional temperature map, built from two circumferential sine-cosine waves and a radial quadratic curve. The authors say the resulting area-average temperature is closer to the true average than the standard sector-weighted method, with differences of 0.5 to 2 K on five similar engines.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Internal contradiction in the only controlled validation: the assumed profile's true average is 526.85 K, the fitted model's area average is reported as 525.85 K, yet Sec. 4.3 calls them 'equivalent,' so the central averaging claim is not demonstrated as written.","rationale":"I read the paper as attempting to provide a practical sparse-measurement reconstruction and area-averaging method for engine thermocouple data, with the headline conclusion that a low-order harmonic model can sometimes give the true area average even without capturing all spatial harmonics. The strongest evidence for this claim is the assumed-profile study in Sec. 4, since the engine cross-validation only tests predictive error at held-out rake locations, not the accuracy of the area average itself. My review found that this controlled demonstration is internally inconsistent: the reported model area average of 525.85 K is neither greater than the sector-weighted value of 526.20 K nor equivalent to the profile average of 526.85 K. Because the central claim depends on this demonstration, the discrepancy is load-bearing. I do not think this requires rejection, since the discrepancy might be a typographical or arithmetic error that is straightforward to correct, and the regression formulation itself is standard and reproducible in principle. The reader's conditional verdict is therefore appropriate, but the paper should be revised to correct or clarify Sec. 4.3 and to provide an uncertainty-aware comparison of model averages against true averages in the controlled study. The reader's stated weakest assumption (the small model class and aliasing risk) is related but distinct; I agree with it as a secondary concern, but the immediate arithmetic contradiction in the paper's only controlled validation is the more pressing issue.","tokens_in":12328,"tokens_out":5963,"duration_ms":61976,"concrete_test":"Recompute the Sec. 4 assumed-profile averages from first principles: (i) integrate the full four-harmonic field over the annulus to get the true area average; (ii) solve Eq. (14) with omega=(1,4) at the four rake sets, form T(r,theta) via Eq. (9), and evaluate Eq. (21); (iii) compare with the reported 526.85 K, 526.20 K, and 525.85 K. If the true and model averages differ by 1 K as printed, the central claim's only demonstration is contradicted. As a stronger check, repeat for random phase and amplitude draws of harmonics (1,4,19,49) and measure the distribution of model average minus true average, to see whether the bias is typically within the engine-level 0.5 to 2 K band.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim in the Conclusions is that the fitted two-harmonic model 'need not necessarily capture all the spatial harmonics, but in some cases it can still deliver the true area average.' Since Eq. (21) shows the area average depends only on the constant Fourier term after radial integration, the claim is plausible: aliased higher harmonics need not bias the mean. The paper's own controlled test is therefore the load-bearing evidence. In Sec. 4, the assumed profile is stated to have an average temperature of 526.85 K; the sector-weighted average is 526.20 K; and the fitted (1,4) model is reported to give 525.85 K. The text then says this value is 'greater than the area weighted value' and 'equivalent to the profile average temperature value.' Both statements are arithmetically false: 525.85 is less than 526.20, which is less than 526.85. Either the reported numbers are wrong, or the only demonstration of the central claim fails by 1 K. The subsequent engine comparisons (0.5 to 2 K differences against sector-weighted averages) do not rescue the claim, because sector-weighted averages are not true area averages; without ground truth there is no way to know whether the fitted field's mean is unbiased. The model-class and aliasing concern in Sec. 3.3 compounds this: if true energy at frequencies above 10 aliases into the fitted constant term, the area-average error could exceed the 0.5 to 2 K band, and the leave-rake-out cross-validation cannot detect it because the candidate set excludes those frequencies. At minimum, Sec. 4.3 must be corrected and re-verified before the central claim can be accepted.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a regularized multivariate linear regression model for reconstructing the 2D temperature field at a fixed axial plane in a turbofan engine from a small number of circumferentially placed thermocouple rakes. The model uses a Fourier basis in the circumferential direction and a quadratic polynomial in the radial direction, with Tikhonov regularization and a brute-force search over two-harmonic frequency pairs. Using 70 extracts from Engines A–D, the authors identify four candidate frequency pairs, and a leave-two-rakes-out cross-validation on Engine E is used to argue that the pair ω=(1,4) has the lowest predictive error. An assumed analytical temperature profile with harmonics (1,4,19,49) is used to test whether a two-harmonic model can recover the true area average; the paper reports an area average of 525.85 K from the fitted model and claims it is equivalent to the true profile average of 526.85 K. The central conclusion is that the fitted model need not capture all spatial harmonics but can still deliver the true area average.","tokens_in":12528,"tokens_out":4954,"duration_ms":56869,"significance":"If the central claim were established, the paper would provide a practical and computationally cheap method for reconstructing sparse engine temperature fields and computing area averages from them, which is relevant to turbomachinery instrumentation practice. The paper has genuine strengths: the formulation in Sec. 3 is explicit and reproducible in structure; Eq. (21) gives a clean analytical expression showing that the area average depends only on the constant Fourier coefficient after radial integration; and the leave-rakes-out experiment on Engine E is a sensible holdout device. However, the headline claim is not currently supported as written. The single controlled validation contains an internal numerical contradiction, and the model-class restriction to two harmonics with max frequency ≤10 means that aliasing of unmodeled modes could bias the very constant term that determines the area average. The significance is therefore conditional: the method is promising, but the evidence for its unbiased averaging property needs correction and strengthening.","major_comments":[{"comment":"The controlled validation of the central area-averaging claim contains an internal arithmetic contradiction. The assumed profile is stated to have an average temperature of 526.85 K (Sec. 4), the sector-weighted average is reported as 526.20 K, and the fitted (1,4) model area average is reported as 525.85 K (Sec. 4.3). The text then says that 525.85 K is 'greater than the area weighted value' and 'equivalent to the profile average temperature value.' Both statements are false as written: 525.85 < 526.20 < 526.85. Since this controlled example is the only place where the fitted model's area average is compared to a known true average, the paper does not demonstrate the conclusion that a two-harmonic model 'can still deliver the true area average.' The numbers must be corrected, or, if the 1 K difference is considered acceptable, the paper must state an explicit tolerance and justify it in the context of engine temperature measurement uncertainty.","section":"Sec. 4, Sec. 4.3, Conclusions"},{"comment":"The model-class restriction is load-bearing for the averaging claim but is not tested against it. The search is restricted to two harmonics with max{ω}≤10, and the radial dependence is a quadratic polynomial. Because Eq. (21) shows that the area average depends only on the estimated constant Fourier coefficient, any aliasing of true higher harmonics (e.g., ω=19 and 49 in the assumed profile) into that constant term directly biases the area average. The paper's only controlled case does show a 1 K discrepancy between 525.85 K and the true 526.85 K, which is consistent with such bias; the Engine E cross-validation cannot detect this because the candidate set never contains the higher frequencies. To support the conclusion, the authors should either prove a condition under which the fitted constant term is unbiased despite aliasing, or demonstrate numerically on an ensemble of profiles with varied higher-harmonic content that the area-average error remains within a stated tolerance.","section":"Sec. 3.3, Sec. 4, Eq. (21)"},{"comment":"The cross-validation on Engine E is partly a model-selection step, not an independent test of the final selected model. The four candidate frequency pairs were pre-selected from Engines A–D, and then Engine E is used to 'further prune down' the harmonics; the claim that ω=(1,4) minimizes ε_test is the output of that pruning. Consequently the reported test errors are not unbiased estimates of the generalization error of the final model, and the comparison in Fig. 7 does not by itself establish that (1,4) is the best pair for unseen data. The authors should either use a nested procedure (select on A–D, evaluate once on E) or explicitly frame Fig. 7 as descriptive model comparison and temper the claim accordingly.","section":"Sec. 3.5, Fig. 7"}],"minor_comments":[{"comment":"Table 2, Case I is described as being based on the Engine A arrangement, but the listed rake positions (54°, 90°, 162°, 234°, 306°, 342°) match Engines B, C, and D in Table 1, not Engine A, which has 270° in place of 306°. Please correct either the table entry or the description.","section":"Sec. 4, Table 2"},{"comment":"Equation (21) contains a typographical double equals sign ('Tavge = = 1/(π(...))'); this should be cleaned up.","section":"Sec. 4.3, Eq. (21)"},{"comment":"The radial polynomial degree p is not explicitly specified; the text refers to a 'quadratic polynomial' but Eq. (6) leaves p general. Please state p=3 explicitly and define M, the number of radial probes per rake, in the notation of Sec. 3.2.","section":"Sec. 3.2"},{"comment":"In Algorithm 1, the relationship between the constraint β in problem (12) and the fixed set of λ values (0.0001, 0.001, 0.1, 10) is not stated; the while loop appears to select among these four values, but the stopping criterion and the choice of β are not tied to the optimization formulation. A brief clarification would improve reproducibility.","section":"Sec. 3.3"},{"comment":"Reference [9] contains a typographical artifact in the author list ('Seshadri, P., , Duncan, A.'); the missing author initial or comma should be fixed.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The arithmetic inconsistency in Sec. 4.3 is likely a typo, but it sits exactly at the load-bearing point of the paper, so the authors must be required to fix it and re-examine the conclusions. In addition, the validation strategy conflates model selection with model assessment; a referee should ask for a clearer separation. The proprietary nature of the data is understandable, but the paper would be strengthened by releasing the assumed-profile test case or a synthetic benchmark so readers can independently check the averaging claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth your time if you work with sparse turbomachinery thermocouple data. It assembles standard tools—ridge regression, Fourier basis, radial polynomial, L-curve selection—into a 2D reconstruction-and-averaging workflow, and it validates the harmonic choice with held-out rakes on a fifth engine. That held-out test is the strongest part: across 28 rake splits, (1,4) does minimize prediction error, and the framework gives a concrete way to turn a few rakes into a spatial field and a sector-aware average.\n\nThe soft spots are real but mostly addressable. The most serious is the controlled case study in Sec. 4. The assumed profile's true average is stated as 526.85 K, the sector-weighted average as 526.20 K, and the fitted (1,4) model as 525.85 K. The text then says the model average is \"greater than the area weighted value\" and \"equivalent to the profile average temperature value.\" Both statements are arithmetically false: 525.85 is less than both 526.20 and 526.85. Since this is the only place where the \"true area average\" claim is checked against ground truth, the central claim is not demonstrated as written. This could be a typo or round-off issue, but it is load-bearing and must be corrected and re-verified before the claim can stand.\n\nTwo other issues are worth noting. First, the four candidate harmonic pairs were preselected from Engines A-D and then pruned on Engine E; the test on E is legitimate, but the selection step is not accounted for, so the reported errors are somewhat optimistic. Nested cross-validation would fix this. Second, the model class restricts max frequency to 10 and the radial polynomial to quadratic. If a real engine has significant energy at higher circumferential or radial modes, those modes can alias into the fitted coefficients, including the constant term that determines the area average; leave-rake-out CV cannot detect this because the candidate set never includes those frequencies. The paper's own discussion of aliasing does not address this particular risk.\n\nThe math is standard but used correctly, the citation pattern is fine, and the paper is honest about its two-part structure. The methodology itself is not fatally flawed; the Sec. 4.3 discrepancy is an arithmetic error, not an incoherent argument. I would send this to peer review, but with the discrepancy flagged as a mandatory revision, and with a request for uncertainty quantification on the reported averages and a clear statement about the model-class limitation. For now, I would not cite the 0.5–2 K claim until the contradiction is resolved. Read it for the formulation and the cross-validation design; treat the quantitative claims with caution.","headline":"Useful engineering recipe with a genuine held-out test, but the controlled validation contains an arithmetic contradiction that blocks the central claim as written.","tokens_in":13241,"tokens_out":2090,"would_cite":false,"duration_ms":23833,"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":"A two-harmonic temperature model can recover true area averages from sparse engine measurements.","keywords":["engine temperature measurements","spatial field reconstruction","area averaging","Fourier harmonics","Tikhonov regularization","cross-validation","sparse thermocouple data","turbomachinery"],"falsifier":"Take a real or simulated temperature field with significant energy at circumferential frequencies above 10 and sharp radial gradients, sample it at the rake positions in Table 1, fit the two-harmonic quadratic model, and compare the model area average with a high-resolution numerical area average of the true field; a discrepancy well above 2 K would refute the claim that low-order fits can still deliver the true average.","tokens_in":11964,"feed_emoji":"🌡️","tokens_out":6582,"duration_ms":72071,"temperature":0.7,"pith_summary":"The paper claims that from a handful of thermocouple rakes at one axial plane, a regularized linear model with just two circumferential Fourier harmonics and a quadratic radial polynomial can reconstruct the two-dimensional temperature field well enough to compute a trustworthy area average. The claim is tested on 82 data-sets from five similar engines, with held-out rake cross-validation on the eight-rake engine showing that the harmonic pair \\(\\omega=(1,4)\\) gives the lowest prediction error. The central point is that the model need not capture every spatial harmonic: because the area average integrates out all nonzero circumferential modes, a low-order fit can still deliver the true average if its constant circumferential term is right. If correct, this gives a practical, parameter-light route from sparse engine test data to spatial temperature maps and area averages.","feed_headline":"Six rakes, two harmonics: true temperature average recovered","feed_subtitle":"Fitted 2D fields come within 0.5–2 K of sector-weighted averages across five engine data-sets.","key_machinery":"The carrying object is the parametric field model \\(T(r,\\$\\theta$)=v^T(r)UX^Ta(\\$\\theta$)\\), where \\(a(\\$\\theta$)\\) contains the Fourier basis for the chosen circumferential harmonics, \\(X\\) holds the harmonic coefficients at each measured radial tap, and \\(v(r)\\) is the quadratic radial polynomial basis. Tikhonov regularization with the L-curve criterion keeps the fitted coefficients small and prevents unphysical oscillation between rakes. The mechanism that makes the averaging claim work is that the circumferential integral in the area-average formula kills every nonzero Fourier harmonic, so the area average depends only on the mean, constant term of the fitted field; if the low-order fit recovers that constant term, the average is correct even when individual harmonics are aliased or wrong.","core_discovery":"The authors' central claim is that a regularized linear model with two circumferential Fourier harmonics and a quadratic radial polynomial, fitted to a few thermocouple rakes at one axial plane, can reconstruct the two-dimensional temperature field well enough to compute a meaningful area average. Across 82 extracts from five engines of the same family, the harmonic pair \\(\\omega=(1,4)\\) consistently gives the lowest held-out prediction error when two of eight rakes on Engine E are reserved for testing. On an analytically generated profile containing four harmonics, the same pair recovers the dominant spatial pattern without capturing the higher harmonics, and its area average matches the true profile average. On the engine data, area averages computed from the fitted fields differ from sector-weighted averages by 0.5 to 2 K. The authors conclude that the model need not capture every spatial harmonic; in some cases it can still deliver the true area average.","pith_inferences":["If the constant-term mechanism holds generally, the same approach should transfer to any circumferentially averaged quantity, such as pressure or species concentration, whenever the averaging operator annihilates the omitted harmonics; a direct test would be to build synthetic fields with energy above \\(\\omega=10\\) and check how large the average error becomes.","The 0.5 to 2 K spread likely tracks how well the quadratic radial basis captures the true radial profile and how favorably the rakes sample the constant term; adding a third radial degree or optimizing rake placement could shrink the spread, but the paper does not test this.","The frequency-selection recipe could be inverted into a sensor-placement rule: choose rake positions that make the Fourier matrix well-conditioned for \\(\\omega=(1,4)\\), since that pair already wins held-out tests; the paper does not pursue this optimization."],"forward_implications":["The paired frequency-selection and regularization approach gives engine-test engineers a parameter-light way to produce full 2D temperature maps from the six or eight rakes already installed.","Area averages from the fitted field can serve as an alternative to sector-weighted averaging, with a quantified agreement of 0.5 to 2 K on five similar engines.","Because the model intentionally fits only low circumferential harmonics, the resulting maps isolate engine modes from blade-to-blade modes, which is exactly the information rigs and CFD cannot supply.","The assumed-profile study shows that even when higher harmonics are present, a six-rake fit with \\(\\omega=(1,4)\\) plus a minimum-norm correction can approximate the true circumferential temperature profile at representative span positions."],"supporting_citations":[{"why":"Supplies the decomposition of temperature into engine modes versus blade-to-blade modes that motivates fitting only low circumferential harmonics.","marker":"[1]"},{"why":"Defines the purpose-based averaging framework that the paper uses to argue why area averaging from a fitted field matters.","marker":"[6]"},{"why":"Provides the sector-weighted area averaging baseline against which the fitted-field averages are compared.","marker":"[10]"},{"why":"Supplies the prior 1D Fourier least-squares approach that the paper's 2D model extends.","marker":"[11]"},{"why":"Provides the training/testing and cross-validation methodology used to select harmonics and certify the model.","marker":"[12]"},{"why":"Supplies Tikhonov regularization and the L-curve criterion used to stabilize the harmonic fit.","marker":"[15]"},{"why":"Explains aliasing, motivating the restriction to low harmonics and the need for regularization.","marker":"[17]"}],"fun_headline_variants":["Few rakes, right harmonics: temperature field resolved","Sparse probes, precise average: new regression method","Two harmonics nail engine temperature average","From 8 rakes to full field: <2K error","Selecting optimal harmonics from sparse data yields correct average"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method assumes the true temperature field is dominated by two low-frequency circumferential patterns, with frequency at most 10, and a smooth radial variation; if a real engine has substantial faster or sharper variation, those patterns will be folded into the fitted low-order coefficients and the field and its average will be wrong.","fun_headline_variants_meta":{"raw":{"variants":["Few rakes, right harmonics: temperature field resolved","Sparse probes, precise average: new regression method","Two harmonics nail engine temperature average","From 8 rakes to full field: <2K error","Selecting optimal harmonics from sparse data yields correct average"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001048,"raw_usage":{"total_tokens":4416,"prompt_tokens":973,"completion_tokens":3443,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":589,"completion_tokens_details":{"reasoning_tokens":3368}},"tokens_in":589,"tokens_out":3443,"duration_ms":28020,"temperature":1.0,"reasoning_tokens":3368,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:30:59.421513+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a real or simulated temperature field with significant energy at circumferential frequencies above 10 and sharp radial gradients, sample it at the rake positions in Table 1, fit the two-harmonic quadratic model, and compare the model area average with a high-resolution numerical area average of the true field; a discrepancy well above 2 K would refute the claim that low-order fits can still deliver the true average.","supporting_citations":[{"cited_title":"Analysis of rotor-stator-interaction and blade- to-blade measurements in a two stage axial flow compressor","cited_arxiv_id":null,"evidence_quote":"Supplies the decomposition of temperature into engine modes versus blade-to-blade modes that motivates fitting only low circumferential harmonics."},{"cited_title":"Averagingnonuniformflowforapurpose","cited_arxiv_id":null,"evidence_quote":"Defines the purpose-based averaging framework that the paper uses to argue why area averaging from a fitted field matters."},{"cited_title":"W., 1980","cited_arxiv_id":null,"evidence_quote":"Provides the sector-weighted area averaging baseline against which the fitted-field averages are compared."},{"cited_title":"Reducing instrumentation errors caused by circumferential flow field variation in multi-stage axial compressors","cited_arxiv_id":null,"evidence_quote":"Supplies the prior 1D Fourier least-squares approach that the paper's 2D model extends."},{"cited_title":"The Elements of Statistical Learning,2nded","cited_arxiv_id":null,"evidence_quote":"Provides the training/testing and cross-validation methodology used to select harmonics and certify the model."},{"cited_title":"C., 2010.Discrete Inverse Problems: Insight and Algorithms","cited_arxiv_id":null,"evidence_quote":"Supplies Tikhonov regularization and the L-curve criterion used to stabilize the harmonic fit."},{"cited_title":"SIAM, Philadelphia, PA","cited_arxiv_id":null,"evidence_quote":"Explains aliasing, motivating the restriction to low harmonics and the need for regularization."}],"review_version":1}