{"id":"eccc596f-74ee-46e6-b203-952967e23a0c","arxiv_id":"2505.06410","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A linear regression over an HPPC discharge pulse estimates the open-circuit voltage drop and corrects the internal resistance estimate, cutting simulation errors from up to 292% to at most 49% and lowering real-cell resistance estimates by up to 20 mOhm.","lead":"This paper shows that standard HPPC battery resistance measurements overestimate internal resistance because they ignore the battery's open-circuit voltage change during the current pulse, and it proposes a least-squares correction that estimates that voltage change from the same pulse data. The correction needs no battery-specific parameters, so it could be added cheaply to battery testers and management systems.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unmodeled RC polarization over the 30 s pulse can bias the proposed R0 estimate, and the real-cell results lack an independent resistance reference.","rationale":"The paper gives a clean analytic decomposition of HPPC overestimation and a least-squares correction that works in simulation when the battery matches the R-int model with constant OC V slope. The simulation numbers (1.6% error at SOC=1, 0.39% at SOC=0.5) support the algebra. However, the real-cell claim depends on the R-int model, which is known to be inadequate for 30 s pulses on Li-ion cells. The authors themselves note in Remark 3 that polarization may deteriorate performance, yet no experiment is designed to quantify this. The constant-slope failure is acknowledged and localized to low SOC, whereas RC dynamics affect all SOCs and bias the intercept from which R0 is recovered. The real-data section compares only two estimators of the same data, so the observed reduction cannot be certified as accuracy improvement without an independent reference. This strengthens the case for the reader's CONDITIONAL verdict: the method is a plausible correction, but the real-cell improvement is not established. No change to the verdict is needed; the conditions should explicitly include a model-validation experiment and an independent resistance benchmark.","tokens_in":14802,"tokens_out":6111,"duration_ms":66551,"concrete_test":"Generate synthetic HPPC data from a first-order Thevenin model with known R0, R1, tau, and a nonlinear OCV-SOC curve; sample at 1 Hz; apply the proposed LS estimator (Eq. 30) at SOCs between 50% and 90% where the OCV slope is nearly constant; if the R0 estimate is biased by more than 10% of true R0, the R-int assumption is the limiting factor. A complementary test on the real cells is to measure resistance with a 100 ms-1 s current pulse or EIS at the same SOC and compare against the proposed estimates.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that Eq. (30) yields accurate R0 requires the voltage during the HPPC pulse to follow an R-int model with constant OCV slope. Real Li-ion cells exhibit RC polarization: v(t) = E(s(t)) + I R0 + I R1(1-exp(-(t-t0)/tau)). Fitting Eq. (30) to such data folds the exponential term into the kappa regressor and shifts the LS intercept, so the estimated R0 is not the true ohmic resistance. Section V only tests the method on an R-int simulator, and Remark 3 concedes that polarization may deteriorate performance on real cells. In Section VI the proposed estimates are compared only with the biased HPPC estimate; there is no independent measurement (EIS, short-pulse, or manufacturer reference) to show the lowered values are closer to truth. The reported 5-20 mOhm reduction is therefore an internal consistency check, not a validation of accuracy.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper addresses internal resistance estimation in HPPC tests. The authors show that the conventional calculation R0 = |Δv/Idis| overestimates the true ohmic resistance because it neglects the open-circuit voltage change ΔE during the pulse. They propose a least-squares observation model (Eq. 30) that simultaneously estimates R0, the initial OCV, and an effective OCV slope κ from the sampled terminal voltage during the discharge pulse, assuming a constant OCV-SOC gradient. The corrected estimate is R0 = (Δv − κ̂C)/Idis. Simulation with a known 5 mΩ resistance shows the proposed method reduces estimation error from up to 291.7% to at most 49.3% across SOC, and the method is applied to four Molicel INR-21700-P42A cells, where it reduces the HPPC resistance estimates by 5–20 mΩ.","tokens_in":14989,"tokens_out":5086,"duration_ms":49083,"significance":"The paper's core derivation is sound, and the simulation benchmark against a known ground-truth resistance is a strength; it clearly demonstrates that the conventional HPPC resistance is inflated by the OCV drop, and the proposed correction is attractive because it requires no additional battery information such as SOC, capacity, or OCV parameters. If the accuracy on real cells were established, the method would be a low-cost improvement for BMS resistance estimation. However, the claimed accuracy is not yet established: the constant-gradient assumption fails substantially at low SOC, and the real-cell experiments lack an independent resistance reference, so the central quantitative claim is only partly supported.","major_comments":[{"comment":"The central assumption of a constant OCV-SOC gradient over the pulse is load-bearing and it breaks down at low SOC. The authors' own Table II shows the proposed estimate has 49.31% error at SOC 0.15, versus 0.39% at SOC 0.5, and the conclusion acknowledges that the gradient changes significantly between 10% and 30% SOC. Since the paper claims accurate resistance estimation, this SOC dependence should be quantified and the valid operating range of the method should be stated; otherwise the reported 49.3% error is not an accuracy guarantee but a bound that is too large for many BMS applications.","section":"Section IV, Eqs. (26)–(29), and Table II"},{"comment":"The real-cell results are not validated against any independent resistance reference. The proposed estimates are compared only to the biased HPPC ratio of Eq. (4), so the reported reduction of 5–20 mΩ cannot be distinguished from a systematic downward shift. The manufacturer datasheet quoted in Table III lists an internal resistance of 16 mΩ, but no comparison is made to this value. An independent measurement, such as EIS at a comparable frequency, a short sub-second pulse, or a manufacturer reference, is needed to establish that the corrected values are closer to the true resistance.","section":"Section VI, Fig. 11 and Table III"},{"comment":"The observation model assumes an R-int circuit with no RC polarization. Over a 30 s HPPC pulse, real cells exhibit an exponential polarization transient; this unmodeled term will be absorbed by the κ regressor and the intercept, biasing the estimated R0. The simulation in Section V uses only R-int data, so the magnitude of this bias is not quantified. The authors should test the estimator on data with realistic RC dynamics, or restrict the method to timescales where the R-int model is a better approximation and provide evidence for that restriction.","section":"Section V, Remark 3, and Eqs. (23)–(30)"}],"minor_comments":[{"comment":"The optimization in Eq. (32) is written as an arg max of the squared residual norm, which is unbounded; it should be arg min to match the least-squares objective and the use of 'lsqnonneg'.","section":"Section IV, Eq. (32)"},{"comment":"The 'gain' metric is defined as the difference between two percentage errors (e.g., 291.67 − 49.31 = 242.36 percentage points), but the abstract phrases this as a 'performance gain in the range of 30% to more than 250% in percentage estimation error', which is ambiguous and should be restated as a reduction in percentage-point error.","section":"Section V.C"},{"comment":"The initiation procedure states that the battery is discharged to Vmin, rested, charged to Vmax, and rested again, but the rest times are not consistently specified; clarifying that the first rest is one hour and the post-charge rest is one hour would improve reproducibility.","section":"Section II.B, Algorithm 1"},{"comment":"Figures 5(b)–(d) show both terminal voltage and OCV, but the captions do not clearly identify which curve corresponds to E(s(k)) and which to v(k); adding labels to the figure or caption would remove ambiguity.","section":"Section V, Fig. 5"},{"comment":"The observation model is attributed to the authors' own under-review paper [52]; although the model is written out in full here, citing a version with a preprint identifier or a self-contained derivation would help readers verify the provenance.","section":"Section IV and Reference [52]"}],"recommendation":"major_revision","confidential_remarks":"The paper falls within the journal's scope and the analytic derivation is a useful contribution. However, the experimental validation is currently an internal consistency check rather than an accuracy validation, and the low-SOC performance is acknowledged but not adequately handled. I would like to see the authors address the three major comments before reconsidering the paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the paper correctly identifies a real bias in HPPC resistance estimation and offers a simple linear correction that works in simulation when its constant-slope assumption holds. The real-cell demonstration is suggestive, not confirmatory, because there is no independent resistance reference and RC polarization is left out of the model.\n\nThe analytic decomposition (Eqs. 11-12) is clean and correct: R0 = (Δv - ΔE)/I, and the standard HPPC estimate is biased by ΔE/I. Their point that the bias grows at low SOC and high current is well taken. The vector observation model (Eq. 30) is not new—it comes from their own under-review paper [52]—but applying it to HPPC and quantifying the error reduction is a legitimate extension. The simulation benchmark against known R0 (5 mΩ, Combined+3 OCV) is the right way to test the method, and it shows the correction reduces error from up to 291% to at most 49%.\n\nThe soft spots are real. First, the load-bearing assumption (26)—constant OCV-SOC slope during the pulse—is violated at low SOC, and the authors' own Table II shows 49% residual error there. That is exactly the SOC range where the correction matters most. Second, the real-cell section lacks any independent reference. The proposed R0 values are only compared to the biased HPPC estimate, so the 5-20 mΩ reduction is an internal consistency check, not validation of accuracy. The datasheet value is AC impedance and not directly comparable. Third, the R-int model ignores RC polarization; over a 30 s pulse, the exponential relaxation is likely to fold into the κ regressor and shift the LS fit. Remark 3 acknowledges this, but it should be shown on real data, e.g., by comparing with a short-pulse resistance estimate. The abstract's claim that the method is 'universally applicable' is too strong, and Eq. (32) has a typo (max should be min). These are minor relative to the main gap.\n\nBottom line: this is an honest, useful engineering paper with a correct core derivation and a clearly stated limitation. It deserves a serious referee, but the real-data claims need independent validation—either EIS, short-pulse measurements, or manufacturer reference—before the correction is taken as established. If the authors add that, it's a solid contribution to the battery testing literature.","headline":"Solid, honest extension that fixes a real bias in HPPC resistance estimation, but the real-cell claims need an independent reference before they are taken as established.","tokens_in":15499,"tokens_out":2612,"would_cite":false,"duration_ms":25693,"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":"Standard HPPC internal-resistance estimates are biased upward by the voltage shift caused by the pulse itself, and a least-squares OCV correction removes most of that bias.","keywords":["battery internal resistance","HPPC test","open-circuit voltage","least-squares estimation","state of charge","equivalent circuit model","lithium-ion battery"],"falsifier":"Run the same HPPC pulse at SOC 0.15 and compare the corrected estimate (simulation predicts 7.47 mΩ at 22.5 A) against an EIS measurement or a 1 ms pulse resistance; if the true resistance is near 5 mΩ, the constant-slope assumption is the cause. More directly, split the 30 s pulse into early and late halves, fit κ separately to each, and check whether the two values differ by more than the estimation uncertainty — if they do, the model is misspecified in exactly the band the paper flags.","tokens_in":14618,"feed_emoji":"🔋","tokens_out":4735,"duration_ms":39816,"temperature":0.7,"pith_summary":"Standard HPPC internal-resistance measurement divides the voltage change by the discharge current, silently treating the battery's open-circuit voltage as frozen. The paper shows that for a 30-second, high-current pulse the OCV actually drops, so the quotient overstates resistance. It proposes a constrained least-squares observation model that fits terminal voltage as a linear function of current and cumulative ampere-seconds, yielding an estimate of the OCV drop with no need for SOC, capacity, or chemistry information. With that correction, simulated worst-case resistance error falls from 291.7% to 49.3% at low SOC and from over 100% to under 2% at full SOC; on four cylindrical cells the corrected estimates are up to 20 mΩ lower. The method's accuracy degrades where the OCV-SOC curve is strongly curved, between about 10% and 30% SOC.","feed_headline":"Battery resistance readings drop up to 20 mΩ after OCV fix","feed_subtitle":"Standard HPPC overstates internal resistance by up to 291%; a data-only OCV correction cuts the top error to 49%.","key_machinery":"The load-bearing object is the linear observation model z = Hx + n (Eq. 30), whose columns are the constant discharge current, a constant '1', and the cumulative coulombs C{a(tx,t0)} drawn up to each sample. It collapses the OCV-SOC relationship into a single scalar κ under the assumption that the OCV-SOC slope is constant across the pulse, so the initial OCV E(s(t0)), resistance R0, and κ can be recovered together by nonnegative least squares. The corrected resistance (Eq. 36) subtracts the estimated OCV drift ΔÊ = κ̂LS C{a(t1,t0)} from the measured voltage drop before dividing by current.","core_discovery":"The central claim is that the conventional HPPC resistance estimator, R0 = |Δv/Idis|, is biased upward by exactly ΔE/Idis, where ΔE is the open-circuit voltage change caused by the pulse, and that ΔE can be estimated from the same pulse data. The paper constructs a vector observation model in which each sampled terminal voltage during the pulse is expressed as Idis R0 + E(s(t0)) + κ times the cumulative coulombs drawn, with κ the ratio of OCV-SOC slope to capacity. Solving the nonnegative least-squares problem gives R0, the initial OCV, and κ simultaneously; the corrected resistance is (Δv - ΔÊ)/Idis. In controlled simulation with true R0 = 5 mΩ, the standard estimate ranged from 7.0 to 19.6 mΩ depending on SOC, while the corrected estimate stayed near 5 mΩ except at SOC 0.15 where it reached 7.47 mΩ.","pith_inferences":["Extending the model from one scalar κ to a piecewise-linear or quadratic OCV segment would likely close the residual 49% error at SOC 0.15; the data already collected contain the necessary samples.","The same bias argument applies to shorter or longer pulses: the correction term scales with pulse duration, so any standardized pulse with non-negligible coulomb throughput is affected.","A direct field test would be to compare corrected HPPC resistance against electrochemical impedance spectroscopy at the same SOC; the paper's claim implies the two should agree much more closely than uncorrected HPPC does."],"forward_implications":["Battery management systems that run HPPC-style pulses can reduce resistance bias without adding sensors or prior knowledge of SOC or capacity.","Reported internal resistance values from standard HPPC tests, and the state-of-power and state-of-health conclusions built on them, are systematically high, especially under high current and low SOC.","The correction makes resistance estimates nearly independent of the pulse's SOC operating point, except in the 10-30% SOC band where OCV curvature remains.","Because the method needs only voltage, current, and time samples from the existing pulse, it can be retrofitted to past HPPC datasets."],"supporting_citations":[{"why":"Supplies the on-off pulse observation model that the proposed resistance-and-OCV estimation approach is built on.","marker":"[52]"},{"why":"Defines the HPPC pulse profile and the standard resistance computation that the paper identifies as biased.","marker":"[46]"},{"why":"Justifies the standardized pulse timing and the treatment of relaxation effects that make the simplified model practical.","marker":"[42]"},{"why":"Provides the battery simulator used for the controlled performance comparison between old and new estimators.","marker":"[3]"},{"why":"Provides the Combined+3 OCV model used to generate simulated OCV and ground-truth voltage behavior.","marker":"[53]"},{"why":"Supplies the cell datasheet with the nominal internal-resistance baseline for interpreting the experimental estimates.","marker":"[55]"},{"why":"Describes the battery test cycler used to collect the experimental HPPC voltage and current data.","marker":"[54]"}],"fun_headline_variants":["HPPC resistance error cut by OCV-aware math","New method fixes battery resistance bias in HPPC","OCV correction shrinks battery resistance error up to 20 mΩ","Better battery resistance from same pulse data","Why standard HPPC overstates resistance—and the fix"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method assumes the slope of the OCV-versus-SOC curve is constant across each 30-second pulse; when the curve bends sharply, as between roughly 10% and 30% SOC, the fitted κ is a biased average and the corrected resistance remains overestimated.","fun_headline_variants_meta":{"raw":{"variants":["HPPC resistance error cut by OCV-aware math","New method fixes battery resistance bias in HPPC","OCV correction shrinks battery resistance error up to 20 mΩ","Better battery resistance from same pulse data","Why standard HPPC overstates resistance—and the fix"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00044,"raw_usage":{"total_tokens":2276,"prompt_tokens":1031,"completion_tokens":1245,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":647,"completion_tokens_details":{"reasoning_tokens":1165}},"tokens_in":647,"tokens_out":1245,"duration_ms":9559,"temperature":1.0,"reasoning_tokens":1165,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T22:43:54.252874+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same HPPC pulse at SOC 0.15 and compare the corrected estimate (simulation predicts 7.47 mΩ at 22.5 A) against an EIS measurement or a 1 ms pulse resistance; if the true resistance is near 5 mΩ, the constant-slope assumption is the cause. More directly, split the 30 s pulse into early and late halves, fit κ separately to each, and check whether the two values differ by more than the estimation uncertainty — if they do, the model is misspecified in exactly the band the paper flags.","supporting_citations":[{"cited_title":"Open circuit voltage and resistance estimation using on-off pulses,","cited_arxiv_id":null,"evidence_quote":"Supplies the on-off pulse observation model that the proposed resistance-and-OCV estimation approach is built on."},{"cited_title":"Battery test manual for electric vehicles,","cited_arxiv_id":null,"evidence_quote":"Defines the HPPC pulse profile and the standard resistance computation that the paper identifies as biased."},{"cited_title":"A study of the influence of measurement timescale on internal resistance characterisation methodologies for lithium-ion cells,","cited_arxiv_id":null,"evidence_quote":"Justifies the standardized pulse timing and the treatment of relaxation effects that make the simplified model practical."},{"cited_title":"Balasingam, Robust Battery Management System Design With MAT- LAB","cited_arxiv_id":null,"evidence_quote":"Provides the battery simulator used for the controlled performance comparison between old and new estimators."},{"cited_title":"Open circuit voltage characterization of lithium-ion batteries,","cited_arxiv_id":null,"evidence_quote":"Provides the Combined+3 OCV model used to generate simulated OCV and ground-truth voltage behavior."},{"cited_title":"Inr21700-p42a v4 datasheet","cited_arxiv_id":null,"evidence_quote":"Supplies the cell datasheet with the nominal internal-resistance baseline for interpreting the experimental estimates."},{"cited_title":"Battery test equipment","cited_arxiv_id":null,"evidence_quote":"Describes the battery test cycler used to collect the experimental HPPC voltage and current data."}],"review_version":1}