Series & Parallel Configuration — Industry Case Study

When a Telecom Tower Retrofit Goes Wrong: The 48V LFP Pack Mismatch Case #

A regional telecom infrastructure operator in Southeast Asia contracted a Shenzhen-based EPC firm to retrofit 38 tower sites with 48V/200Ah LFP battery backup systems in early 2023. The design was straightforward on paper: four 50Ah prismatic cell packs connected in parallel to hit 200Ah, each pack built from 16 cells in series. The cells were Grade-A LFP sourced through a Dongguan pack house with a reasonable datasheet — 3,000 cycles at 0.5C, ±2mV cell voltage matching at pack assembly.

The first failure reports came in at week 11. Sites were tripping on low-voltage cutoff despite the battery monitoring system (BMS) showing 35–40% SOC. Field technicians initially blamed the inverter thresholds. They weren’t the problem.

What the EPC firm’s procurement team missed was internal resistance spread across the four parallel packs. The Dongguan supplier had shipped packs from three separate production batches — a common practice when buyers don’t specify batch consolidation in their purchase order. Batch-to-batch DC internal resistance (DCIR) variance measured during our incoming inspection of a parallel lot ranged from 8.2mΩ to 14.7mΩ per pack (measured at 50% SOC, 25°C, per IEC 61960-3:2017 1C discharge pulse method). When you connect packs with that spread into a parallel bus, Kirchhoff’s current law doesn’t care about your design intent — current distributes inversely proportional to impedance. The low-resistance packs carry disproportionate load.

At 38 sites running daily generator-charge cycles, the two lowest-resistance packs in each bank were cycling harder than their spec allowed. By week 14, those packs showed measurable capacity degradation. The BMS, tuned to a flat SOC estimation curve assuming uniform pack behavior, was reading aggregate voltage while individual packs were diverging. The 35% SOC display was accurate for the high-resistance packs and catastrophically wrong for the two that were already below 12% usable capacity.

Internal Resistance Spread, String Derating, and the Parameters That Predicted the Failure #

The root cause here isn’t exotic. Internal resistance mismatch in parallel configurations is one of the most well-documented degradation accelerants in lithium iron phosphate systems, yet it remains the most commonly under-specified parameter in pack-level procurement contracts. Buyers ask about capacity, cycle life, and voltage tolerance. Almost nobody specifies a maximum DCIR spread across packs destined for the same parallel bank.

The parameters that would have caught this problem at source:

DCIR spread tolerance is the first and most important. For a 48V parallel bank destined for daily cycling applications, a reasonable specification is ≤1.5mΩ spread across all packs in a single shipment, measured at 50% SOC and 25°C. The Dongguan supplier’s 6.5mΩ spread was more than four times that threshold. For stationary applications with infrequent cycling, you might accept up to 3mΩ. For telecom sites running 1C charge from generator daily, 1.5mΩ is already pushing it.

Capacity matching at parallel assembly matters less than most buyers think, up to a point. A 2% capacity spread across parallel packs is generally acceptable because current sharing corrects over time during cycling. DCIR spread is the variable that drives current hogging from the first cycle — capacity imbalance develops afterward as a consequence.

BMS SOC algorithm assumptions are the silent amplifier. The system’s BMS used a voltage-based coulomb counting hybrid tuned for a single homogenous pack. When you run parallel packs with independent cell-level sensing per string (which this system did not have), you’re flying partially blind. The IEEE 1491-2012 guide on lead-acid battery monitoring pre-dates LFP deployment at scale, but its core principle applies: individual string monitoring is non-negotiable when packs have any impedance variation.

Cycle rate asymmetry is what caused the accelerated degradation timeline. Using a simplified current sharing model based on our incoming inspection data: the 8.2mΩ pack absorbed roughly 31% more charge current than its nominal 25Ah share (0.5C design). That pushed effective cycling rate to approximately 0.65C on those packs — well within LFP spec individually, but our 18-month tracking of Shenzhen pack house lots (logged under what we call the IR-Delta Screening Protocol in our incoming QC process) shows that packs cycled consistently at 0.6C+ in parallel banks with >4mΩ DCIR spread show capacity fade reaching 80% retention roughly 40% earlier than single-pack testing predicts.

The most overlooked parameter in procurement specs for parallel LFP systems isn’t on most buyers’ radar until they’ve had a field failure. DCIR spread at pack level — not cell level — is what determines how current distributes in a real bus. Cells can be perfectly matched at assembly and still diverge at pack level if thermal management during formation cycling differs across batches.

You can find pack-level DCIR specification guidance in IEC 62660-1:2018, which covers performance testing for lithium-ion traction batteries — the methodology translates cleanly to stationary pack qualification even though the standard targets EV cells.

Decision Framework: What Changes Based on Pack Count and Application Duty Cycle #

If the design is 2P or 3P with light-duty cycling (less than one full cycle per day, less than 0.3C charge rate), a ≤3mΩ DCIR spread is a workable threshold. Current sharing asymmetry exists but degrades slowly enough that the system’s service life won’t be materially affected. For a grid-tied home storage application with 0.2C charge from solar, this is the realistic standard — tightening it to 1.5mΩ will add procurement cost without meaningfully extending field life.

If the configuration is 4P or higher, or if the application involves daily high-rate cycling (generator charge at 0.5C–1C is common in telecom and off-grid industrial), the tolerance must tighten. Specify ≤1.5mΩ spread, single production batch, and per-string current sensing on the BMS. Without per-string sensing, the system controller cannot detect divergence until it’s already causing accelerated degradation. This is where the telecom case failed — the EPC firm treated the BMS requirement as a cost item to optimize rather than a diagnostic necessity.

If you’re integrating packs from multiple suppliers (common in replacement or expansion scenarios), treat the parallel bank as a mixed-impedance problem from day one. I’d prioritize re-measuring DCIR on every pack before integration and grouping by impedance class — low with low, high with high. We’ve advised several integrators in the Shenzhen area to implement what amounts to a two-tier parallel bank: matched packs at the string level, balanced discharge controllers between strings. More hardware cost upfront, but it avoids the $214,000 problem.

The boundary condition worth stating plainly: this guidance holds for LFP chemistry in portable and stationary pack configurations. For NMC parallel banks, the calculus changes because NMC cells have steeper voltage-SOC curves, which means voltage-driven current sharing tends to self-correct more aggressively than in LFP’s flat discharge region. The same DCIR spread that causes catastrophic imbalance in a 48V LFP bank may be survivable — though not ideal — in an NMC equivalent.

For buyers sourcing battery cells and considering how cell chemistry selection affects parallel configuration tolerance upstream, the relevant background is in Cell Technology.

One specific, non-obvious recommendation: require your supplier to measure and document pack DCIR at 20% SOC, not just 50%. LFP’s flat mid-range discharge curve masks impedance divergence. Pack-to-pack spread at 20% SOC is consistently 15–25% wider than at 50% SOC in our incoming lot data, which means a supplier who tests at 50% and ships without re-testing at low SOC is showing you their best numbers. The 20% SOC test is where bad batches reveal themselves.

The full documentation trail for UN38.3 transport testing — relevant when these packs ship internationally — should already include impedance data from UN38.3 Section 38.3.4. If a supplier’s UN38.3 report doesn’t include impedance measurement records for your specific pack configuration, that’s a signal about their test documentation discipline, not just their compliance status.

Sourcing Guidance for Buyers #

When evaluating Chinese suppliers for parallel-configured LFP pack systems, the first document to request is the outgoing QC report for impedance matching — specifically the DCIR spread across every pack in your shipment lot, not a sample. Suppliers who batch-sort by impedance at pack level exist, but they’re a minority among Dongguan and Shenzhen pack houses. If a supplier can’t provide per-pack DCIR at shipment, they’re not doing impedance-based sorting, which means your parallel bank configuration is being assembled with unknown current sharing characteristics.

The qualification red flag specific to parallel pack sourcing: if the supplier quotes cycle life from single-pack testing and has never characterized performance in a 3P or 4P configuration, their cycle life number doesn’t apply to your design. Single-pack cycle life and parallel string cycle life diverge significantly when impedance spread is present. Ask specifically whether their cycle life data covers parallel configurations. In our experience reviewing supplier qualification packages, fewer than one in four Shenzhen-area pack houses have parallel configuration aging data on file.

For incoming inspection, measure DCIR on 100% of packs destined for the same parallel bank (this is not a case for sampling), using a consistent SOC reference point of 50% ±2% and 25°C ±2°C. Any pack exceeding 1.5mΩ from the batch mean gets quarantined. For a 200Ah/48V system as described above, this adds roughly 2–3 hours of incoming inspection time but eliminates the primary failure mode. For deeper context on how BMS per-string current sensing specifications should be written into procurement contracts, the relevant technical reference is in BMS Engineering.

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

---