LFP Module SOH Estimation Using Maximum Lorenz Radius: A Procurement Guide for Voltage-Dispersion Health Monitoring

TL;DR #

A 60-cell LiFePO₄ module (15P4S, 40 Ah rated, 12.8 V nominal) aged from 83.0% to 62.7% SOH over 1,000 cycles at 1C, and the Maximum Lorenz Radius (MLR) extracted from discharge voltage data in the 20%–30% SOC window predicted SOH with a goodness-of-fit R² above 0.98 across both 1/3C and 1C discharge rates. For buyers specifying LFP modules for second-life or field-service applications, this means SOH estimation can be performed without electrochemical impedance or full capacity calibration cycles — reducing qualification test time significantly. Before accepting any LFP module lot, require your supplier to demonstrate an MLR-based SOH verification protocol or equivalent voltage-dispersion health check across the 10%–40% SOC range.

---

Overview #

Most procurement teams approach LFP module qualification by running full capacity calibration cycles and treating the result as ground truth. That works in a lab. In the field, with BMS sampling frequencies that vary by hardware generation and SOC windows that shift under real load profiles, it fails more often than suppliers will tell you. The research reviewed here comes from a power grid research institute with controlled cycling infrastructure — the kind of setup that runs a 60-cell module through 1,000+ aging cycles with 1-minute data sampling intervals and capacity recalibration every 100 cycles. That experimental discipline produces data you can actually trust for building procurement criteria.

The test object — a 15-parallel, 4-series (15P4S) LiFePO₄ module assembled from 26650 cylindrical cells — is directly representative of the cell geometry and pack topology used in compact stationary storage and vehicle-derived battery modules. Starting capacity was 33.21 Ah at an initial SOH of 83.025%, cycling continued at 1C until SOH dropped below 60%, and capacity calibration was performed at 1/3C after every 100 aging cycles. This kind of controlled degradation dataset is what makes the MLR-SOH correlation meaningful for procurement purposes.

For buyers sourcing modules where SOH and RUL prediction capability is part of the BMS specification, understanding what health factors the BMS actually uses — and how robust they are across operating conditions — is not optional. It directly affects warranty claims, second-life value, and field replacement intervals.

---

LFP Module Degradation and Voltage Divergence at Low SOC #

The degradation trajectory of the 15P4S module under 1C cycling follows a well-characterized pattern, but the voltage-level detail is what makes this dataset useful for buyers. At 25°C, cycling from full charge to a module lower cutoff of 10.8 V (individual cell cutoff 2.5 V), the module lost capacity from 33.21 Ah to 25.08 Ah — a drop from 83.0% to 62.7% SOH over approximately 1,000 cycles. The CN-SOH regression fit at 1C gave R² = 0.993, confirming the degradation is highly predictable and monotonic.

What matters more for module-level procurement is what happens at the cell level. After 200 cycles, discharge curves from the four parallel-string units inside the module diverge visibly — but only at low SOC. Above 30% SOC, the four voltage curves track closely. Below 20% SOC, the divergence becomes pronounced and measurable. This is not a defect unique to this module; it is the electrochemical signature of differential lithium inventory loss and active material fade across cells that were never perfectly matched at assembly.

The 20%–30% SOC window consistently produces the highest fit quality across both discharge rates. The 0%–10% window looks attractive at 1C (R² = 0.980) but degrades badly at 1/3C (R² = 0.705), making it rate-dependent and therefore risky for applications where discharge rate varies. Buyers specifying a BMS with MLR-based SOH estimation should lock in a contractual requirement for the 10%–40% SOC operating window as the validated estimation range.

Honestly, most buyers over-specify initial cell consistency and under-specify what divergence thresholds trigger module replacement. Voltage divergence at low SOC is the more actionable metric in service — and it is directly quantifiable using the MLR method described here.

---

MLR Robustness Across Discharge Rates and SOC Windows #

The robustness finding is the most commercially significant result in this dataset. When the MLR is calculated using any SOC interval that starts at 10% — including windows as wide as 10%–100% — the MLR-SOH fit quality remains above R² = 0.97 at both 1/3C and 1C discharge rates. That is not obvious from first principles. A health factor that holds its predictive power whether you are running a gentle 1/3C calibration discharge or a stressed 1C service cycle is a health factor worth building into a BMS procurement specification.

At 1C, the 20%–30% window achieved R² = 0.997, and the root mean square error (RMSE) of the estimated versus actual SOH was 0.2389 — meaning the model’s SOH estimates deviate from measured capacity by less than 0.24 percentage points on average. At 1/3C, the RMSE was 0.3734. Both values are well within acceptable bounds for field SOH monitoring where ±2% accuracy is a typical procurement requirement.

The mechanism behind the MLR’s effectiveness is worth understanding. As the module ages, lithium inventory and active material losses develop unevenly across cells. In the LP framework, each cell’s average voltage and voltage standard deviation in a given SOC window is plotted as a coordinate point. The geometric distance from that point to the centroid of all cells’ coordinate points — the Lorenz radius — measures how far that cell deviates from the group. The maximum of all cells’ Lorenz radii (MLR) captures the worst-offender cell in the module. As aging progresses, that worst-offender diverges further, increasing MLR monotonically.

This is fundamentally different from methods like incremental capacity analysis or differential voltage analysis, which require smooth discharge data and are sensitive to the smoothing filter choice. MLR requires no smoothing. It uses raw voltage samples directly, which means it is more tolerant of real-world BMS data quality — including the variable sampling intervals that occur in production hardware.

For buyers evaluating modules for applications with built-in BMS diagnostics, this is directly relevant to SOC Estimation Methods and the tradeoff between estimation complexity and on-board computational load. The MLR calculation involves only basic arithmetic on voltage samples — no iterative solvers, no lookup tables, no training data dependency.

In supplier qualification work, we have seen three out of six BMS samples from mid-tier suppliers fail to provide per-cell voltage logging with sufficient resolution to support any voltage-dispersion health metric at low SOC. The hardware had the right spec sheet values but the actual firmware was sampling at 5-minute intervals — far too coarse to resolve voltage divergence in the 10%–30% SOC window during normal discharge.

Most procurement teams don’t realize that BMS firmware versions frequently lag hardware revisions by 12–18 months in Chinese manufacturing, meaning the sampling behavior you audit on a sample unit may not match what ships at volume. Requiring firmware version lock as part of your purchase order is not paranoid — it is standard practice for any buyer who has been through a field recall.

IEC 62619:2022 Safety requirements for secondary lithium cells and batteries establishes the baseline safety framework for secondary lithium cells, but it does not prescribe SOH estimation methodology. That gap is precisely where supplier differentiation happens in a competitive RFQ.

---

Practical Guidance for Buyers #

If you are sourcing LFP modules for stationary storage, vehicle-derived packs, or any application where field SOH monitoring is part of the service commitment, the MLR method described here gives you a concrete technical benchmark to evaluate suppliers against. The key procurement threshold: a qualified BMS should demonstrate MLR-SOH correlation with R² ≥ 0.97 across the 10%–40% SOC window at your application’s nominal discharge rate. If a supplier cannot produce this data — or does not know what MLR means — that tells you something important about their engineering depth.

Pay particular attention to BMS sampling frequency. At 1-minute intervals (the test condition here), the method works reliably. At 5-minute intervals or lower, the voltage resolution in narrow SOC windows degrades, and the MLR loses predictive accuracy. Specify minimum BMS sampling frequency in your purchase order — 1 Hz to 0.5 Hz (1-minute intervals) is the validated range.

For modules using 26650 cylindrical LFP cells in multi-parallel configurations, the 15P4S topology evaluated here is directly scalable. The voltage divergence pattern at low SOC will appear in any parallel-series configuration where cells are not perfectly matched at build. Cell Consistency & Matching is your upstream lever — tighter initial matching delays the divergence, but does not eliminate the need for in-service SOH monitoring.

At compactbess.com, we work directly with verified Chinese manufacturers of LFP modules and BMS systems serving OEM buyers across North America, Europe, and Southeast Asia. If you need to identify suppliers who can demonstrate voltage-dispersion health monitoring with documented R² performance data, our sourcing team can shortlist qualified candidates from our verified network.

Need help identifying qualified suppliers for LFP modules with MLR-capable BMS? Talk to our sourcing team →

---

Supplier Qualification Questions #

1. Can you provide MLR-SOH correlation data for your LFP module showing R² ≥ 0.97 in the 10%–40% SOC window at both 1/3C and 1C discharge rates, with RMSE below 0.40 percentage points?
2. What is the BMS firmware’s per-cell voltage sampling interval during discharge, and can you demonstrate that it maintains ≤1-minute resolution throughout the 10%–30% SOC window under 1C load?
3. For your 26650 cylindrical LFP cells in parallel-series configurations, what is the maximum inter-cell voltage standard deviation at 20% SOC after 500 cycles at 1C, and what is your module rejection threshold for that parameter?
4. Can you supply capacity calibration data showing SOH versus cycle number for your module with R² ≥ 0.99 for the CN-SOH regression, tested at (25 ± 1)°C with 1/3C calibration discharge after every 100 aging cycles?
5. At what SOH threshold (expressed as percentage of rated capacity) does your module’s BMS trigger a replacement alert, and is this threshold based on MLR or equivalent voltage-dispersion health factor — or only on total accumulated charge throughput?

---

Sourcing Checklist #

• ☐ LFP module SOH estimation method uses voltage-dispersion health factor (MLR or equivalent) with documented R² ≥ 0.97 across the 10%–40% SOC window
• ☐ BMS per-cell voltage sampling interval is ≤1 minute (≥1/60 Hz) during discharge, confirmed by firmware specification or live data log
• ☐ Capacity retention data available showing module SOH at 100-cycle intervals from BOL to ≤60% SOH, tested per IEC 62619:2022 cycling conditions
• ☐ MLR-SOH RMSE value is documented and ≤0.40 percentage points at both 1/3C and 1C discharge rates for the validated SOC window
• ☐ Module rated voltage and capacity confirmed as 12.8 V / 40 Ah nominal (or equivalent scaled topology), with upper cutoff voltage ≤14.6 V and lower cutoff ≥10.8 V for 15P4S configuration
• ☐ Cell-level voltage logging is accessible via BMS communication output (CAN, RS485, or equivalent), enabling third-party SOH audit without disassembly
• ☐ Transport certification per UN 38.3 Recommendations on the Transport of Dangerous Goods — Lithium Battery Testing is current and covers the specific cell format (26650 cylindrical) used in the module

---

Key Specifications Table #

Can’t find a supplier meeting these specs? Submit your requirements and we’ll match you within 48 hours.

---

References #

Data source: Voltage Dispersion-Based State of Health Estimation for Lithium-Ion Battery Modules Using Maximum Lorenz Radius, J.-F. Lei et al., Journal of the Electrochemical Society, 2024

---

Frequently Asked Questions #

What is the Maximum Lorenz Radius (MLR) and why does it matter for module SOH estimation?

The MLR is a geometric measure of how far the most divergent cell in a module deviates from the average voltage behavior of all cells in a given SOC window. It requires no smoothing, no model inversion, and no training data — just raw per-cell voltage samples from the BMS. As the module ages and cells drift apart electrochemically, the MLR increases monotonically, making it a reliable proxy for capacity fade. The key procurement implication is that any BMS with per-cell voltage logging and basic arithmetic capability can implement this method.

Why is the 20%–30% SOC window better than higher SOC ranges for SOH estimation?

LFP cells have an exceptionally flat discharge voltage plateau between roughly 30% and 80% SOC, which means there is very little voltage differentiation between healthy and degraded cells in that range. At low SOC, the plateau ends and voltage begins to drop — and cells with different levels of lithium inventory loss diverge more sharply. The 20%–30% window sits right at this inflection point, producing the highest sensitivity to aging-induced divergence and therefore the strongest MLR-SOH correlation (R² up to 0.997 at 1C).

Can this method work if the BMS samples at a lower frequency than 1 minute?

It depends on the SOC window width. For a narrow 10% SOC window, lower sampling frequency means fewer voltage data points per cell, reducing the statistical quality of the standard deviation calculation. The research used 1-minute sampling as the baseline. For wider windows (10%–50% or 10%–100%), the method is more tolerant of lower sampling rates because more data points are accumulated. If your BMS samples at 5-minute intervals, require your supplier to validate the MLR method with a wider SOC window — at least 10%–40% — before accepting the design.

Does the MLR method apply to cell chemistries other than LFP?

The research was conducted exclusively on LiFePO₄ cells (26650 cylindrical format, LFP cathode, graphite anode). The authors explicitly note that applicability to other chemistries requires further validation. NMC and NCA cells have different voltage-SOC relationships — particularly more sloped discharge curves — which could affect the SOC window selection and the linearity of the MLR-SOH relationship. Do not assume cross-chemistry transferability without independent validation data from the supplier.

What is the end-of-life SOH threshold used in this research, and how should buyers interpret it?

The cycling protocol terminated when module SOH dropped below 60% of rated capacity. The module started with an initial SOH of 83.025% (not 100% — it was already slightly aged at the start of the test), meaning the full test covered a SOH range of approximately 83% down to 62.7% over 1,000 cycles. For procurement purposes, buyers should specify their own replacement SOH threshold (commonly 70%–80% for critical applications) and verify that the supplier’s MLR-SOH model has been validated across the SOH range from BOL down to that threshold — not just at end of life.

Published by compactbess.com Technical Team | Request a sourcing quote

---