Custom Battery Pack Design: 9 Engineering Decisions That Control Performance & Service Life

A custom battery pack can meet its nominal voltage and Ah target and still be wrong for the machine it powers. In industrial custom battery pack design, the decisive inputs are usually hidden behind the headline specification: real current profile, pulse duration, temperature distribution, cell consistency, BMS response logic, mechanical failure path, and the data available when the pack eventually needs diagnosis.

This article stays inside that engineering problem. It does not reopen the generic custom-versus-standard battery comparison. If that decision is still unresolved, our discussion of custom lithium battery pack design for OEMs addresses that earlier decision point. Here, the job is to turn equipment requirements into testable design decisions before prototype assumptions become production constraints.

Custom Battery Pack Design Starts With the Load Profile, Not Voltage and Ah

The first useful specification for custom battery pack design for industrial equipment is the machine's duty cycle. A request such as "48 V, 200 Ah" defines two labels; it does not define how the battery will be stressed.

An AGV may cruise at moderate power but produce repeated acceleration peaks. A forklift can combine traction demand with hydraulic lifting. A mobile robot may alternate long low-load periods with short motion pulses, while some traction systems return regenerative current to the pack. These differences change cell loading, voltage sag, thermal rise, conductor sizing, BMS protection timing, and sometimes the contactor architecture.

For design review, the requirement set should establish the operating voltage window, required energy/runtime, continuous current, peak current and duration, recharge window, ambient temperature, installation volume, vibration or shock exposure, communication interface, and any regenerative or reverse-current condition.

A high-voltage machine also shows why nominal current is insufficient. Connecting a battery directly to equipment with substantial DC-link capacitance can create high inrush current and contactor stress, so a pre-charge sequence may need to be part of the system architecture. Texas Instruments documents this relationship between DC-link capacitance, inrush current and pre-charge design. (Texas Instruments)

For OEM teams still defining these inputs, our custom battery pack design for industrial equipment resource covers the information worth locking before an RFQ reaches battery engineering.

Translate the Duty Cycle Into Electrical Architecture

Once the load profile is known, custom battery pack design for OEMs can translate that requirement into series count, parallel capacity, current paths and switching hardware.

Series count establishes the usable voltage window. Parallel capacity contributes energy and current capability. The important part is not the arithmetic itself. It is whether the resulting architecture still works at high SOC, low SOC, peak load, elevated resistance and the intended temperature range.

The basic electrical relationship is covered separately in our custom battery pack design and manufacturing architecture discussion. The engineering issue here is what happens after those cells are physically connected.

Parallel cells do not divide current perfectly forever. Differences in temperature, internal resistance, state of charge, connection resistance and aging can redistribute current inside a parallel group. Current-path symmetry, cell consistency and thermal behavior therefore need to be reviewed as one pack-level problem rather than as three unrelated checklist items.

Ask for peak-load voltage traces and thermal data from the intended configuration, not only a calculation showing that nominal cell current multiplied by parallel count exceeds the machine rating. This is often the point where a theoretically adequate configuration starts to show its real margin under the equipment's duty cycle.

Cell Matching: Capacity Is Only One Variable

Cell matching should control the variables that drive divergence over time: capacity, internal resistance or impedance, measurement condition, SOC, temperature and production traceability. A resistance value measured at one SOC and temperature should not be compared casually with a value measured under another condition.

Capacity-only matching is not a sufficiently strong specification for an industrial pack expected to deliver repeatable performance over long service periods.

Polinovel has published internal LFP pack testing that illustrates the scale of this issue. Across configurations tested at 25°C, 80% DOD and 0.5C charge rate, packs assembled with ≤3% cell-to-cell tolerance reached roughly 1.5–2 times the cycle count before 80% SOH compared with packs assembled with ≤12% tolerance.

Those results should not be copied into an arbitrary custom LiFePO4 battery pack design as universal incoming-QC thresholds. The published summary does not establish one cell model, sample size and production distribution as representative of every LFP platform. The correct acceptance window still has to be derived from the selected cell specification, supplier distribution and controlled measurement method.

A supplier saying "we use Grade-A cells" does not answer the matching question. Ask what is measured at incoming inspection, at what SOC and temperature, what tolerance is applied, whether capacity and DCIR are both controlled, and whether the production lot remains traceable after assembly.

Cell matching should reduce long-term divergence, not merely make beginning-of-life readings look uniform. A tolerance number without SOC, temperature, test current and cell-model context is not a meaningful supplier-comparison metric.

Design for Temperature Uniformity, Not Just a Safe Average Temperature

For custom battery pack thermal management, maximum temperature is only half of the problem. The other half is temperature distribution across cells and parallel paths.

 

A 2026 Energy study reported an 8.2°C intra-pack temperature gradient associated, under its test conditions, with about 35% energy loss after 200 cycles. (ScienceDirect)

 

A 2026 eTransportation study compared parallel-connected cells over 1,200 fast-charge cycles, including a module with a 10°C temperature gradient. The warmer cell initially carried more current and later showed greater capacity fade and resistance growth. (ScienceDirect)

 

A 2026 Journal of Power Sources study using a 101.8 Ah large-format cell and a controlled 10°C longitudinal gradient reported local current-density differences of approximately 15–20%, with lithium plating appearing earlier than under uniform-temperature conditions. (ScienceDirect)

 

These are experimental results, not universal battery-pack acceptance limits. Their value is that they demonstrate the same engineering mechanism from different angles: thermal heterogeneity can change current distribution and accelerate unequal aging.

 

There is no useful single "good pack temperature" without the operating scenario.

 

A compact stationary pack running near steady state should be evaluated differently from an AGV that alternates acceleration, opportunity charging and idle periods. A high-power industrial vehicle adds busbar, contactor and enclosure heat to cell heat. The validation target therefore needs both absolute temperature and spatial distribution under the actual load profile.

 

Polinovel's published cold-chain field data provides another useful boundary condition. Across cold-storage deployments, field measurements have shown roughly 80–90% of rated discharge capacity retained at −20°C. That figure does not prove that a pack has good thermal uniformity. It demonstrates why room-temperature bench data cannot replace measurements in the intended environment.

 

Request a thermal map under representative load, identify the hottest cell or group and the hottest-to-coldest spread, then verify whether sensor placement can actually see the region driving the peak. A single BMS temperature channel located in an easy-to-wire position can report a safe number while another part of the pack operates under a materially different thermal condition.

 

Temperature uniformity matters because temperature differences can become current and aging differences. The project-specific question is which operating state creates the worst thermal map, and that cannot be inferred from nominal pack rating alone.

BMS Design Is a Behavior Specification, Not a Feature Checklist

A useful custom battery pack BMS design specification defines behavior: threshold, delay, hysteresis, recovery, balancing logic, sensor placement, contactor control, pre-charge, communication and fault history.

Listing OV/UV/OCP/OTP/CAN only confirms that functions exist. It does not establish how the pack behaves when the equipment reaches a boundary.

Over-current protection is a good example. A legitimate motor-start pulse can exceed continuous current substantially. If the supplier gives only an OCP threshold without a delay curve, the OEM cannot determine whether the BMS will trip during normal operation or remain permissive during an unsafe event.

The same applies to temperature protection. A complete requirement should establish which sensor triggers the action, whether charge and discharge use different limits, what delay or hysteresis applies, how recovery occurs and what event remains available for diagnosis.

Communication also has to serve the machine. CAN or RS485 data may include pack current, cell-group voltages, temperatures, SOC, SOH, active faults, historical events and contactor state. The exact mapping still needs to be agreed for the target controller rather than assumed from the protocol name alone.

Diagnostic information that was never designed into the BMS cannot be reconstructed reliably after a field failure. This is the point where protection architecture starts to determine maintainability as well as safety.

Mechanical Design Must Control Both Everyday Stress and Worst-Case Failure

For custom battery pack enclosure design, the first engineering question is which mechanical, thermal and electrical failure modes the enclosure must control over the product life.

Everyday requirements include cell restraint, vibration resistance, busbar support, connector retention, insulation, sealing, thermal paths and mounting loads.

Material choice should follow the failure mode. Mica-based barriers are useful where thin electrical insulation and high-temperature resistance are priorities between cells or modules. Ceramic-fiber insulation is more appropriate where thermal isolation during an abnormal heat event matters more than compactness. Thermally conductive interface or potting materials fit a different problem: moving normal operating heat toward a controlled thermal path, although full potting can complicate serviceability and rework. Flame-retardant structural polymers may be appropriate for housings or internal supports when impact, insulation and weight matter, but they should not be treated as substitutes for a validated propagation barrier.

The final material stack and thickness still have to be validated against the selected cell format, propagation strategy, service model and enclosure geometry. The important correction is that "application dependent" does not mean "no engineering preference exists." It means the preferred material follows the dominant failure mode.

Worst-case design begins with a less comfortable assumption: one cell may fail despite correct manufacturing and normal protection.

NASA's X-57 battery program demonstrated the positive side of that design philosophy. During a test in which a cell thermal event was deliberately triggered, NASA reported that the system contained the event to the trigger cell rather than allowing propagation to neighboring cells. (NASA)

The Boeing 787 APU battery investigation shows the opposite side. The NTSB concluded that an internal short circuit in one cell led to thermal runaway and propagation into adjacent cells, and its investigation criticized the design's ability to mitigate the most severe consequences of an internal short. (NTSB)

The design question is therefore not only:

"How do we prevent a cell failure?"

It is also:

"If a cell has already entered failure, does the pack architecture limit propagation and preserve a controlled failure path?"

That question drives spacing, barriers, vent paths, sensing, insulation, enclosure strength and validation. It is also why a normal charge/discharge test says very little about worst-case mechanical safety.

Design for Maintainability Before the First Prototype

A battery that is difficult to diagnose will usually also be expensive to support, even if its original bill of materials was optimized aggressively.

The useful custom battery pack design maintenance data should be chosen from the service model backward rather than by adding every possible telemetry channel.

For an AGV or automated fleet, event logging and remotely readable cell-group voltage, temperature, current, SOC and fault history should usually come before elaborate physical service access. The fleet operator needs to identify patterns across many packs without opening each enclosure.

For equipment serviced by trained technicians, accessible diagnostic communication and inspection of connectors, fuses or other designated service components may deserve higher priority.

For a sealed low-service product, physical accessibility may be intentionally limited. In that case persistent fault history and remote diagnosis become more important because opening the pack is not the expected troubleshooting path.

Polinovel has published a field example that shows why this matters. In a 2023 AGV support case involving a previous supplier's LFP packs, the BMS used 80 mA balancing current. After roughly six months, cell-voltage spread had drifted to 63 mV, and the pack triggered a protection shutdown that caused approximately four hours of production-line downtime.

The important part of that case is not that 63 mV is a universal failure threshold. It is not. The lesson is that a field symptom such as "the battery shuts down" is difficult to root-cause unless the system preserves cell-group data, protection events and operating context.

Before prototype release, write down the five field questions you expect a technician or fleet manager to ask after a failure. Then check whether the BMS and mechanical design preserve the evidence required to answer them. If one cannot be answered from available data or the service interface, that is a design gap rather than a future after-sales problem.

Maintainability must be designed before deployment. Which diagnostic signals deserve priority depends on whether the product is fleet-managed, technician-serviced or effectively sealed for life.

Validate the Pack With the Real Load Before Mass Production

A prototype that powers the machine successfully has demonstrated basic compatibility. It has not demonstrated a validated custom battery pack design.

Prototype characterization, DVT, pilot/PVT and production EOL testing serve different purposes. The naming can vary by OEM, but each critical requirement should eventually map to an acceptance method.

For custom battery pack design validation, that typically means checking usable capacity, voltage sag, thermal behavior, peak-current response, BMS protection behavior, communication, charging, environmental conditions and the failure modes relevant to the application.

Transport qualification and application safety also need to remain separate.

UN 38.3 belongs to the United Nations transport-test framework for lithium cells and batteries. It addresses transport-related test requirements rather than replacing application-specific safety qualification. (UNECE)

IEC 62619:2022 addresses safety requirements for industrial secondary lithium cells and batteries and includes industrial motive applications such as forklifts, golf carts and AGVs within its stated application scope. For an industrial vehicle project intended for European or North American deployment, application-safety requirements and transport requirements should therefore be scoped separately at project start rather than treating UN 38.3 as the complete compliance plan.

Certification answers whether the battery meets a defined standard or test requirement. Engineering validation answers whether this battery, in this equipment, under this load and environment, behaves as intended.

What to verify before mass production: the DVT/PVT report should show a direct line from each locked requirement to its test condition, result and acceptance criterion. "Tested successfully" without those three pieces is not enough evidence for design release.

What Field Data Should Trigger Maintenance or Engineering Review?

Condition-based diagnosis is more useful than applying one universal threshold to every custom battery pack design maintenance data set.

The comparison condition matters. Runtime collected in a cold warehouse cannot be compared blindly with runtime under a warmer, lighter duty cycle and attributed entirely to aging.

The same applies to cell-group ΔV or temperature spread. An isolated peak during a transient event and a progressively worsening trend at comparable SOC/load conditions are different diagnostic signals.

Custom Battery Pack Design Review Checklist Before You Release the Design

A custom Li-ion battery pack design checklist should do more than ask whether every subsystem has been discussed. It should identify answers that are too vague to release into tooling or mass production.

If an OEM engineering team asks how to design a custom battery pack, this checklist is the practical answer: every headline requirement must eventually become an architecture decision, a measurable acceptance criterion, and evidence that survives into production.

For projects that already have a duty cycle, operating voltage, peak load, installation envelope, environment and communication requirement defined, Polinovel can review those inputs against the electrical, BMS, thermal, mechanical, validation and manufacturing implications of a customized industrial battery solution.

The goal of custom battery pack design is not to add complexity. It is to expose the variables that can invalidate performance, safety or maintainability assumptions while the design can still be changed.

A pack designed this way is not defined only by nominal voltage and capacity. It is defined by whether its electrical architecture, thermal behavior, BMS logic, mechanical protection and diagnostic evidence remain consistent with the machine it was built to power.

FAQ