What a Battery Temperature Reading Misses About Cooling

A battery temperature reading describes one location at one moment. It does not, by itself, reveal how much heat the battery generates, whether another region is hotter, or how much electricity the cooling system consumes. Those questions require different measurements. Understanding the difference helps explain why a liquid-cooled battery can have a cool measurement point and still develop a substantial temperature gradient.

A 2024 study by Aron Saxon and colleagues at the National Renewable Energy Laboratory (NREL) makes this distinction concrete. It combines cell and module calorimetry, a vehicle-pack cooling experiment, and a model checked against a small prototype. Its most useful lesson for stationary storage is a chain of cause and effect: operating conditions determine where heat is generated; construction determines how that heat travels; cooling sets the boundary through which it leaves; and sensor placement determines which part of the story becomes visible. [1]

Heat generation, temperature and cooling electricity are different quantities

Heat generation is a rate, measured in watts. Temperature, measured in degrees Celsius, is the thermal state that results as heat accumulates or leaves. Cooling electricity is another power flow: the electrical input used by equipment such as pumps, fans or refrigeration. A temperature chart cannot substitute for a heat-generation measurement, and heat rejected by a cooling system cannot be treated as its electricity consumption.

A useful simplified energy balance is Cₜₕ × dT/dt = Q̇generated − Q̇removed. Here Cₜₕ is thermal capacitance in joules per kelvin, dT/dt is the temperature change in kelvin per second, and the two heat rates are in watts. This is a lumped explanation for a chosen part of the system, not a model of every point inside a cell. It shows why a flat temperature trace need not mean little heat is being generated: removal may simply be matching generation.

Heat also needs a temperature difference to move through a finite thermal resistance. In a simplified steady path, Q̇ = ΔT/Rₜₕ, with Rₜₕ in kelvin per watt. Cooling the end of that path does not instantly cool its far end. The path includes the cell materials, interfaces, contact to the cooling surface and the cooling assembly. This is the engineering reason to examine both heat-removal capacity and temperature distribution; increasing one does not automatically establish the other.

Why a cold plate can be cold while another part of the cell heats up

The clearest experiment in the NREL paper used a 5 kWh plug-in-hybrid vehicle pack containing approximately 100 pouch cells of 15 Ah each. The pack had an active cooling system with a cold plate along one edge of the cells. Before testing, the researchers held the pack at 40 °C until it reached thermal equilibrium. They then switched on cooling and applied the specified vehicle power profile. This was a vehicle-pack experiment, with a different cell format, scale and duty from a stationary LFP cabinet. (Saxon et al., pp. 8–9, Figure 5). [1]

Each cell was instrumented at three positions: near the cold plate, midway along the cell, and farthest from the plate. During the run, the highest measured temperature reached approximately 55 °C. The minimum-temperature trace fell below the initial 40 °C ambient during part of the test, while the measured temperature spread reached about 12 °C at times. These are features of the time series; the peak temperature and the largest spread should not be combined as if all extrema occurred simultaneously. (Saxon et al., pp. 8–9, Figure 5). [1]

The authors explain the separation through the heat path. The plate removed heat from the nearby edge faster than heat could travel from the hotter, more distant region. The cooling system was therefore removing heat while a gradient was developing. The experiment challenges the tempting assumption that a low reading near a cooling surface demonstrates uniform cooling throughout the cell.

It also changes how a temperature difference should be interpreted. The reported spread came from thermocouples at specified positions. It was not a direct measurement of every point inside the electrochemical layers. A sensor can be accurate at its own location while still leaving other regions unobserved. Calling a surface reading a “cell temperature” without its location can conceal that distinction.

For a stationary storage project, the transferable insight is the need to understand the path between heat source, sensor and coolant. The experiment does not supply a universal acceptable temperature spread, a stationary-battery life penalty, or a reason to adopt one cooling geometry for every cell. Those depend on cell construction, current, duration and the complete thermal design.

An isothermal test does not keep every part of a battery at one temperature

The paper measured heat using an isothermal calorimeter. A battery sample sat in a chamber submerged in a temperature-controlled liquid bath, with heat-flux sensors used to measure heat dissipated from the sample. Integrating that heat-flow signal over time gave thermal energy. This method addresses a different question from attaching a thermocouple to a cell: it measures the heat passing through the test boundary. (Saxon et al., p. 2, Methods). [1]

“Isothermal” describes the controlled measurement system. The researchers report system control of ±0.001 °C and baseline heat stability of ±5 mW, but explicitly state that the battery sample can still increase in temperature. Instrument stability is not evidence that the tested cell has no gradients. The distinction matters whenever a test report pairs very precise laboratory specifications with conclusions about a battery’s own temperature.

There is also a timing limitation. Heat takes time to travel from the sample to the calorimeter sensors. The authors warn that the measured heat-flow trace may not reproduce the time-resolved heat flow at the cell surface, especially under high power. A delayed thermal signal should not be aligned with an electrical pulse and interpreted as a perfectly instantaneous response. The measurement window and the handling of thermal lag are part of the result, rather than details that disappear behind a single average number. (Saxon et al., p. 2). [1]

A calorimeter and a surface-temperature sensor can give different-looking traces without either measurement being wrong. The calorimeter observes heat crossing its measurement boundary after a transport delay; the thermocouple records the temperature at its location. During a transient, those signals describe different parts of the same process, rather than two interchangeable measures of battery heating. (Saxon et al., p. 2). [1]

The same current does not always produce the same heat

The familiar relation Q̇ohmic = I²R explains why connection losses and resistive cell losses become more significant as current rises. If resistance were fixed, doubling current would quadruple that component of heating. In a battery, however, effective resistance changes with temperature, state and the timescale of the load. The square-law relation is a useful mechanism, not permission to scale an entire thermal test by current alone. (Saxon et al., pp. 4–6, Equations 3–4). [1]

The NREL study shows the temperature dependence using a 12 Ah LMO/LTO pouch cell tested at 30 °C, 0 °C and −15 °C. At a stated current, colder conditions changed both heat generation and available discharge capacity. The authors connect the additional resistance to reduced ionic transport in the electrolyte. This example uses a different chemistry from the LFP cells in HUA-X stationary cabinets; its temperatures and heat rates are not an LFP operating map. Its value is demonstrating why heat depends on operating state as well as the current setting. (Saxon et al., pp. 5–6, Figure 2). [1]

The paper also separates irreversible losses from reversible, or entropic, heat associated with electrochemical state changes. That reversible contribution can change sign with state and current direction. Charging and discharging therefore need not have identical thermal signatures, even at equal current magnitude. Treating all observed heating as one constant resistor discards part of the mechanism. (Saxon et al., p. 4, Equation 3). [1]

This creates an important tradeoff: a warmer cell may have lower transport resistance during a particular test, but that does not make higher operating temperature a general design goal. Short-term electrical performance and long-term degradation are different outcomes. A cooling strategy has to manage the actual duty while respecting the applicable operating and life conditions; it cannot be chosen from the lowest resistance point alone.

Why single-cell heat multiplied by cell count can miss the pack load

A second experiment compared a stand-alone 4.5 Ah cylindrical LFP/graphite cell with the same cell type assembled into a high-power hybrid-vehicle module. The researchers expressed module heat on a per-cell basis so that it could be compared with the isolated cell. At the approximately 8C current relevant to that application, the module produced about 20% more heat per cell. The authors attributed the additional heat to the module interconnects. (Saxon et al., pp. 7–8, Figure 4). [1]

That comparison isolates a system-design issue: packaging adds electrically resistive paths that are absent from a single-cell heat measurement. A cell count can scale cell heat, but it cannot introduce the missing connection heat. It also cannot tell a designer whether heat is concentrated at a connector, distributed across a cell surface, or separated from the cooling path by a poor thermal interface.

The 20% result belongs to that high-power module and its approximately 8C application. It is not a surcharge to add to every BESS cooling calculation. HUA-X PowerFac-522, for example, lists 0.25C and 0.5C charge/discharge options; a high-power vehicle experiment does not establish its loss fraction. The assembly adds current-carrying connections as well as cells, so its heat load depends on both cell behaviour and the resistance of those connections under the operating current. [1] [2]

A further boundary appears at cabinet level. PCS losses, battery heat and the cooling equipment’s own electricity are distinct terms in a system energy account, and their physical locations matter. The question “how much heat must this loop remove?” is therefore different from “how much energy does the complete system deliver at its AC connection?” The round-trip efficiency guide explains that second measurement boundary.

A counterexample: less heat after cycling is not proof of slower ageing

The study’s formation experiment is especially useful because it resists a simple “older means hotter” narrative. The researchers tested 4.4 Ah cylindrical LFP/graphite cells after the first formation cycle and repeated measurements over 30 subsequent cycles. Figure 3 describes a reduction in the approximately 1C heat rate from about 0.15 to 0.1 W/Ah after this early conditioning period. These are heat rates normalized by cell capacity, not cabinet efficiency percentages. (Saxon et al., pp. 6–7, Figure 3). [1]

The authors interpret the early change in terms of electrode-surface limitations, electrolyte wetting and the evolution of surface impedance. It is a formation-stage observation, not a long-duration lifetime trial. Reading the word “aged” in a figure legend without the methods could turn a short conditioning result into the opposite claim: that prolonged ageing improves thermal performance.

The broader lesson is that a change in a heat trace needs a history. It may reflect temperature, current, state of charge, early conditioning, or later degradation. A lower temperature could also reflect a different cooling condition. Without separating those causes, a cooler reading cannot establish improved state of health. The battery life article develops the distinction between test history, capacity retention and service-life conclusions.

What a validated model can add—and what its validation actually covers

Measurements do not have to capture every internal point for a thermal model to be useful. In the final example, the researchers characterized a 5 Ah NMC cylindrical cell, developed a second-order resistor–capacitor electrical model, and coupled it to a three-dimensional thermal representation. They then compared predictions with a six-cell prototype under a US06 vehicle driving-cycle load. (Saxon et al., pp. 9–10, Figure 6). [1]

For that prototype and test, the reported maximum discrepancy between measured and predicted surface temperatures was less than 2.5 °C. The model also estimated differences between central and corner cells, helping identify where the design might need a lower load or a better heat-transfer path. The spatial result adds design insight beyond a single temperature trace, but the internal distribution remains a model prediction. The surface comparison validates the stated test response; it does not directly measure every predicted internal temperature. (Saxon et al., pp. 9–10). [1]

A cell model describes the cell’s response; the housing, contacts and cooling boundaries determine additional paths through which an assembly exchanges heat. Identical cell parameters can therefore produce different temperature distributions in different pack geometries. In the six-cell prototype, the comparison with surface measurements supported the model’s representation of heat transfer through that particular housing. Its accuracy remains tied to the tested geometry, load and boundary conditions. (Saxon et al., pp. 9–10, Figure 6). [1]

Liquid cooling, power derating and operating duty

PowerFac-261 is an LFP, liquid-cooled system with 125 kW rated AC power and 261 kWh nominal battery energy. Its stated operating temperature range is −20 to 55 °C, with output derating above 45 °C. That range describes operating conditions; it does not describe the temperature at every point inside a cell. Above the derating threshold, the rated AC power alone is insufficient to determine available output—the applicable derating curve defines the reduction. The PowerFac-261 technical brief lists the model ratings and operating-temperature condition. [3]

PowerFac-522 also uses liquid-cooled LFP architecture, with 522 kWh nominal energy, 125 kW or 250 kW power options, and stated 0.25C/0.5C charge/discharge options. The same nominal energy can be paired with different power ratings, so stored energy alone does not describe the load history or the resulting thermal response. These electrical ratings also do not specify coolant flow or cooling-system electricity; those are separate thermal and auxiliary-power quantities. [2]

A short power event can finish while heat is still accumulating in the cells. A sustained load continues to generate heat, so its temperature history depends on removal through the thermal path as well as the starting temperature. Even where two duties have the same peak power, different durations can produce different temperature responses. At a steady temperature, heat removal balances generation; a flat trace is therefore compatible with continuing heat production. Model-specific ratings and operating conditions can be requested through the PowerFac-261 document pack or the PowerFac-522 configuration documents.

The research changes the question from “is this battery cool?” to “where is heat produced, how does it leave, and what does this measurement actually observe?” A credible answer connects those three parts. That connection is what turns a temperature reading into useful evidence for the design and operation of a liquid-cooled BESS.