Reduction in fossil fuel consumption is a key avenue towards reducing rising worldwide carbon emissions [1
]. Electric vehicles (EVs) can potentially not only reduce fossil fuel consumption, but also satisfy customer demands [2
]. The percentage of EVs in global light vehicle production is expected to rise from <2% in 2016 to about 6% by end of 2020, and to at least 22% by 2030 [3
]. In an EV, the cost of the battery pack can be as much as 25% of the total cost. To become cost comparable with internal combustion engines, the battery pack cost will need to decrease from around $
200–300 per kWh presently, to around $
100–150 per kWh by 2025, which is predicted to happen [1
Lithium-ion batteries have become the most popular energy storage technology used in EVs, due to their high gravimetric and volumetric energy densities, excellent cycling performance and long calendar life [5
]. Building on the success of land based EVs, aerospace electrification is also emerging. Norway and Sweden have already announced plans to use electric airplanes for all short-haul flights in their airspace [6
]. From recent prototypes, it is clear that the electric aircraft will use lithium-ion or similar comparable future battery technology.
Countries like Norway and Sweden, and large parts of continental Europe and North America, are characterised by seasonally extreme cold weather down to −20 °C [8
]. In these conditions, a lithium-ion battery’s capacity, maximum output power and charge acceptance capability are reduced. At lower ambient temperatures (≤10 °C), due to slower chemical kinetics, charge transfer at the electrode-electrolyte interphase slows down, leading to higher overpotentials [10
]. Furthermore, lower ionic conductivity in the electrolyte and slower diffusion, particularly in the electrode contribute to higher voltage loss at lower temperatures [11
]. These manifest as reductions in driving range and performance of EVs at low ambient temperatures [12
]. This performance reduction is worsened by increased energy/power demands due to higher cabin heating requirements in cold weather conditions [13
Low ambient temperatures also lead to large reductions in real time range/performance estimation of EVs [15
]. For example, the real-world usable range of the Nissan Leaf is 298.7 km at −2 ℃ [12
]. This is a substantial reduction from the 385 km range claimed in advertisements. These factors exacerbate predisposed range anxiety and have forced users to consider alternatives, limiting the customer acceptance of EVs [19
]. Inaccuracies in range estimation, rarely have led to catastrophic failure in road transport EVs; however, in electric aircrafts they would. For example, following an aborted landing, inaccurate range estimation could be the difference between an aircraft undertaking a safe holding pattern or an unsafe rough landing. Therefore, with burgeoning aircraft electrification, accurate range estimation has never been more important.
Results from galvanostatic battery experiments, at different ambient temperatures and C-rates, are employed to directly model the usable charge/discharge capacity of lithium-ion cells under different operating scenarios [20
]. In these model parameterisation experiments, it is assumed that the test conditions are isothermal. However, at low ambient temperatures, due to higher cell internal resistances (R
, cells heat up more [21
]. This effect is compounded if the operating current is high. Grandjean et al. recently argued that self-heating is a performance enhancement feature for lithium-ion cells operating at low ambient temperatures [22
]. They conducted galvanostatic discharge cycle experiments at −10 °C ambient temperature. Towards the end of discharge at 5C, the cell surface temperature was about 30 °C; a nearly 40 °C temperature rise due to self-heating. On the other hand, if the cells were not allowed to heat up, the cell voltage would have decreased at a faster rate and likely resulted in a lower capacity discharged. This implies that while the starting cell temperature may be close to the ambient temperature, at the end of a discharge cycle the cell temperature can be considerably higher than the ambient temperature, particularly for high load currents.
Furthermore, there is a vast literature on lithium-ion cell models, where researchers have claimed that model parameters have been obtained from battery experiments conducted under isothermal conditions [15
]. On closer examination of the literature, claimed isothermal conditions are often found to be non-isothermal. For example, to parameterise their equivalent circuit model, Jaguemont et al. obtained discharge capacities for their cells at −20, −10, 0 and 25 °C ambient temperatures [23
]. For 1C discharge at −20 °C, the final cell temperature was −5.6 °C and for 1C discharge at −10 °C, the final cell temperature was 0.9 °C. This means that the discharge characteristics were representative of a test temperature much higher than either the −20 or −10 °C starting ambient temperature. In fact, the final cell temperature for the test at −20 °C is 4.4 °C higher than the starting ambient temperature of the next parameterisation point, −10 °C, which leads to redundancy.
While beneficial for a drive cycle, self-heating is an issue for a parameterisation experiment where repeatability is key. Clearly, the tests performed in literature are not isothermal, so attention should be paid to the test setup to improve this shortcoming.
The effects are particularly important for low temperature parameterisation where the effects of self-heating are much larger than at room temperature. This is a prominent limitation because by allowing the cells’ operating temperature to diverge from the ambient (or test) temperature, the effect of temperature and current on the accessible capacity can become convoluted. This is because, if the capacity performance is being assessed at a particular temperature, the cell itself should be anchored to that temperature for the duration of the parameterisation test. This is because cell overpotential is a function of both operating current and operating temperature [27
]. To draw out the effect of one, the other should be fixed for a particular experiment.
Conventionally, battery cycling is conducted in commercially available thermal test chambers. A thermal chamber allows control of operating temperature and humidity conditions during a test. In particular, temperature control is achieved by forced convection of air within the chamber. As widely reported in the literature, for experiments conducted in air-based thermal chambers, as a lithium-ion cell self-heats under load, its operating temperature increases [22
]. This means that for ‘air’ cooling, the cell operating (surface) temperature is not fixed to the starting ambient temperature and changes significantly during a parameterisation cycle. Limited investigation has shown that battery parameters are more in line with those from temperatures higher than the starting ambient temperature [28
]. In their study, Ardani et al. employed Peltier elements to control cell surface temperature during pulse discharge. However, this case did not represent the more extreme case of a continuous discharge at an ambient temperature (for example, 1C discharge).
In the literature, active thermal management of lithium-ion cells (such as those employing oil baths) has largely been investigated from the perspective of the performance characteristics of the heating/cooling system [29
]. Bolsinger et al. investigated the effect of different cooling configurations for cylindrical cells and argued that tab (or terminal) cooling is better than surface cooling. Similarly, other researchers have compared the pros and cons of choosing between air cooling and liquid cooling [29
]. The comparisons were made on the basis of cost, complexity, efficiency, direct versus indirect cooling, and so on. Air has poor thermal properties compared to other materials (such as mineral/silicone oils) (Table 1
), and so the cooling capability may be insufficient to counter the rapid heat generation seen for high load currents and/or low ambient temperatures. This means that for a similar cooling area and coolant mass, the cooling capability of air is substantially lower than that of a direct liquid cooling system [33
]. Compared to conductive cooling with Peltier elements [28
], an immersed oil bath allows simultaneous cooling of all cell surfaces (including the tabs) and is independent of cell format or size.
In operation, cells naturally heat up, however, the aim of this paper is to improve parameterisation data for range estimation. Therefore, the contribution of this paper is to show that traditional (air) thermal control methods potentially lead to overestimation of remaining driving range by inaccurately assessing usable capacity/energy characteristics. This paper offers a methodology to conduct close-to-isothermal experiments by better tracking accessible capacity as a function of temperature and current. This allows better correlation of accessible battery capacity/energy with remaining driving range. This is done by parameterising the cells within an immersed oil bath with direct liquid cooling to allow isothermal control. Results will be compared with the air thermal control method. A validation of the distinction between ‘air’ and ‘oil’ based parameterisation data will be shown by tracking SOC evolution by employing a typical 1st order battery electrical equivalent circuit model, parameterised with two different datasets.
The paper is structured as follows. In Section 2
, the experimental and modelling methods employed for this study are described, including the experimental setup and test matrix chosen. Then in Section 3
, the results are presented and discussed. The effect of choosing oil cooling compared to air cooling will be shown. Then, combined with a typical first order equivalent circuit model (ECM) with hysteresis parameterised between −20 and 25 °C, SOC evolution and remaining driving range estimation for both air and oil approaches are compared for a legislative drive cycle (2 US06 cycles, back-to-back). Finally, in Section 4
, the key learnings from this study are summarised.
3. Results and Discussion
presents the capacity test results for the air cooling parameterisation condition. Capacity measured at a particular discharge rate was decreasing with temperature, and cell surface temperature was higher at lower temperature, as expected [23
]. At 25 °C, both capacity and total energy discharged decreased with discharge rate. In contrast, it was increasing at lower temperatures, e.g., at −20 °C capacity increased by 0.9 % and 8.8 % at 1C and 3C, compared to 0.25C. This is directly linked to the temperature raise of the cell. For example, at −20°C, as discharge current increases from 0.25C to 3C, the temperature rise changes from 4.7 °C to 30.7 °C with respect to the ambient temperature. This implies that the capacity/energy performance cannot just be ascribed to the starting ambient temperature. Although cell capacity takes into account the change in lithium concentration in the electrode’s active material, it does not consider the energy used to change the concentration. The energy delivered is the energy that is transferred to an external load when the cell is being discharged. Thus, depending upon the cell voltage, the energy delivered can be different for a particular capacity (Ah) discharged. As shown in Table 4
, this is the case at 0 °C ambient temperature, where discharge capacity increases as discharge current increases from 1C to 3C, but the discharge energy decreases. This implies that the effect of self-heating is not enough to affect the voltage as much as it affects the cell overpotential.
3.1. Comparison between Air and Oil Based Thermal Control System for Low Ambient Temperatures and High Load Currents
presents the capacity test results for the oil cooling method. Conducting experiments using an oil-based thermal control system makes a substantial difference to the discharge capacity of the cell compared to air control in a thermal chamber. For example, Figure 2
shows 1C discharge curves at −20 °C for both air- and oil-cooled methods. The air cooling method produces 31.7% higher capacity (39.2% higher energy) than when employing active thermal management (oil cooled). This implies that self-heating is substantially affecting cell discharge behaviour at −20 °C ambient temperature. This is reflected in the voltage profiles as well, which is the cause for higher percentage of total energy. For the same discharge current, the voltage for the oil cooling case decreases much faster than for the air cooling case. The counterintuitive increase in cell voltage observed for the air cooling case, which was also reported in previous research [46
], is not seen for the oil cooling case, as it is likely related to the temperature increase of the cell.
As also shown in Figure 2
, for the same test, the final surface temperature for the ‘air cooling’ case is −3.9 °C, considerably higher than the −15.3 °C measured for the ‘oil cooling’ case. Note that, even for the oil cooling case, there is a temperature rise seen during the discharge cycle, indicating it is also not completely adiabatic. This is due to the limited cooling capability of the oil-based immersion rig; this is limited by the performance of the Lauda unit at −20 ℃ and below. This is possibly complicated by the decreased heat capacity, increased viscosity and decreased viscosity of Kryo-95 in such conditions; although the oil is rated for use till −95 ℃. Therefore, the difference in capacity seen will be higher when the surface temperature is controlled exactly at the ambient temperature. The shape of temperature change, i.e., a sharp increase at the beginning and slower increase for remaining discharge duration likely due to the change of resistance with temperature and SOC.
Resistance was measured using 1C discharge pulses of 10 s at 20%, 50% and 80% SOC, at −20, −10, 0 and 25 °C. The pure Ohmic part (Ro
) of the resistance is not expected to remain constant over the SOC range tested. Figure 3
presents the Ro
values parameterised at different temperatures averaged across the SOC set points. For a −20 °C ambient temperature, the Ro
is 21.7 mΩ. Whereas at 25 °C, the DCR is 1.8 mΩ. DCR increases with decreasing ambient temperature because of higher overpotentials induced due to charge transfer, lower ionic conductivity and sluggish diffusion in the electrode active material [23
For both oil and air cases, the electrical characteristics for the constant current discharges are similar until self-heating takes effect and becomes dominant. At the end of discharge, when the load current was removed, it was seen that, for the air test, the Ohmic resistance was 7.8 mΩ (assuming cubic fit for resistance at −3.9 °C). On the other hand, for the oil test, the corresponding value was 16.3 mΩ (fit at −15.3 °C). In fact, the difference between the two thermal control methods, and thus the impact of self-heating, is reflected in the rising cell voltage seen for discharge using air cooling (red circle in Figure 2
). As temperature increases, Ro
drops, producing less heat, slowing the temperature rise rate.
In Figure 4
, it is shown that the discrepancy between choosing air based and oil based thermal control system for parameterisation tends to become more prominent as ambient temperature decreases and/or discharge C-rate increases. For the air cooling case, higher discharge C-rate and lower ambient temperature means greater self-heating and higher cell temperatures. However, for the oil cooling cases, the surface temperature rise seen was suppressed to a considerably smaller value. This was reflected in the capacity characteristics for the oil cooling cases compared to the air cooling cases. For example, for 3C discharge at −10 °C, the capacity discharged for the air cooling case was higher by 9.5% than that discharged for 0.25C. However, for the oil cooling case, the capacity discharged for 3C was lower than that for 0.25C by 22.3%.
Forced air convection or ‘air cooling’ is employed as part of traditional parameterisation methods. However, this method overestimates capacity available at low ambient temperatures due to rapid degree of self-heating making corresponding models unreliable. This means that depending on the thermal control system, any conclusions regarding the effect of self-heating can be different. Therefore, when reviewing models, the temperature rise for the capacity test is important because the lower the temperature rise, the more reliable the data. Thus, it is recommended that active thermal management or ‘oil cooling’ is employed as part of standardised parameterisation methods.
To validate the effect of overpotential for both cases (‘air cooling’ and ‘oil cooling’ parameterisation), after discharge at the low ambient temperature was terminated, the cells were brought up to 25 °C ambient temperature and then discharged with a nominal 0.10C constant discharge current. This is shown in Figure 5
and is termed as the residual capacity [48
]. To investigate the effect of low ambient temperature induced cell overpotential, the OCV-SOC discharge relationship of the cells was also included in the comparison. While the residual capacity for the ‘air cooling’ case was 9.8 Ah, the corresponding value for the ‘oil cooling’ case was 18.0 Ah. This aggregates to a total capacity discharged of 41.7 Ah and 41.8 Ah, respectively. Furthermore, the total capacity discharged for the OCV case was similar (42 Ah).
In the literature, it is argued that self-heating helps improve discharge performance of cells at low ambient temperatures. While self-heating is beneficial in terms of real-world performance, it adversely affects the cell characterisation data for battery modelling, where the performance of the cell is required to be maintained at the desired parameterisation temperature. For example, to obtain capacity values for different C-rates at an ambient temperature (for example −20 °C), it is imperative that the cell operating temperature remains close to the ambient temperature throughout the experiment, i.e., isothermal test conditions.
For the air cooling cases, for example for 1C discharge, the final cell temperature was −4.9 °C. This means that the capacity value obtained would be for an effective temperature between −20 and −4.9 °C rather than at the chosen ambient temperature. In this section, it has been shown that by keeping the test conditions more isothermal (using oil cooling), the capacity values obtained are considerably lower than those seen in the air cooling cases. This issue has been shown to become more prominent as ambient temperature decreased and/or discharge C-rate increased. For example, at 25 °C ambient temperature and 0.25C discharge current, the ‘air cooling’ capacity was only higher by 0.5% than the ‘oil cooling’ capacity (Figure 6
). However, at −20 °C and for 3C discharge, the ‘air cooling’ capacity was 61.7% higher than the corresponding ‘oil cooling capacity’. Finally, due to convolution of electrical and thermal phenomena, for 3C discharge, as ambient temperature decreases from 25 to −20 °C, the capacity reduction should be about 66% (oil cooling case), rather than just 12.1% (air cooling case).
Overall, there is a significant difference between characterising lithium-ion cells for performance investigation and for model parameterisation. In real-world applications at low ambient temperature, it is better for cells to heat up faster to achieve better performance. However, for model parameterisation involving current and temperature as separate model inputs, it is better that the effects are separated and that the temperature of the cell is maintained at a constant level. The modelling implications are demonstrated in the following section.
3.2. Validation Using SOC Estimation: Effect of Legislative Drive Cycles
To compare the effect of choosing different thermal control methods (air based and oil based), the cells were cycled for two legislative drive cycles. The starting SOC chosen was 65%. The ambient temperatures chosen were −15, −5, and 5 °C. Fully charged cells were first discharged to 65% SOC using 0.5C operating current at 25 °C ambient temperature. Then, the cells were cooled down to the test temperature and rested for at least 4 h before employing the drive cycle. Note, the drive cycle experiments were conducted in a standard thermal chamber with forced air convection. This is in contrast to the SOC parameterisation experiments that were conducted for both ‘air cooling’ and ‘oil cooling’ cases. The power, current, voltage and surface temperature profiles are given in Figure 7
. For translating the difference between air and oil approaches at a parameterisation level to RDR estimation level, it was assumed that the inputs were experimental operating current and measured surface temperature. That is, for estimating SOC for 2xUS06 at 65% starting SOC and −15 °C starting ambient temperature, the inputs considered in the model were current and surface temperature (as per Figure 7
b, d). For the 2xUS06 cycling for −15 °C starting ambient temperature, the experimental capacity and energy discharged were 11.3 Ah and 33.9 Wh, respectively. From the model, the end SOC estimated for the air case was 33.3% and for the oil case it was 24.9% (Figure 8
). The SOC difference between the two approaches is 8.4%. Assuming a 100 mile remaining driving range for 100% SOC, the remaining driving range would therefore be 8.4 miles lower for the oil approach compared to air approach. The mean power for the drive cycle was 140 W. To validate the SOC estimated, after the cells were cycled for 2xUS06 cycles, the cells were immediately further discharged at 140 W until their power delivery reduced by 50% (i.e., to 70 W under 2.7 V cut-off voltage). Assuming a 42 Ah capacity from the rest voltage in Figure 5
, the approximate real SOC was 22.3%. Therefore, employing an ‘air’ parameterised dataset will lead to inaccurate assessment of usable capacity and cause models to overestimate remaining driving range compared to those parameterised using ‘oil cooling’, using data from the cell closer to the aim parameterisation temperature.
The significance of ensuring isothermal operating conditions may have been overlooked, as most battery testing is conducted at higher operating conditions where the degree of self-heating is lower. This is due to the cell internal resistances being lower (Figure 3
). The reduced impact of ‘isothermal operation’ is illustrated in Figure 9
a. It is shown that for the 2xUS06, the end SOC values parameterised for air and oil cases converge at higher ambient temperatures. Correspondingly, the surface temperature rise is also reduced as ambient temperature increases (Figure 9
b). The difference in the end SOC values for air compared to oil decreases from 8.4% (at −15 °C ambient temperature) to 0.2% at 25 °C ambient temperature. The corresponding values for surface temperature rise are 10.2 °C and 1.1°C, respectively. The cell core temperature will be higher due to poor through-plane thermal conductivity of lithium-ion cells [49
], which can be estimated through an appropriate thermal model. If temperature is measured at the core, a better thermal control can be effected [50
]. This would be ideal, allowing better separation of thermal and electrical phenomena by maintaining exact isothermal conditions during the experiment.
Overall, the results have demonstrated that maintaining isothermal test conditions during model parameterisation experiments have a direct effect on battery state evolution. However, the latest testing standards, such as the IEC 62660-1 and UNECE 38.3, only specify the ambient temperature. This is because most standards are defined to emulate testing for real-world performance of lithium-ion cells. However, for modelling purposes, particularly to parameterise models, there do not exist any testing standards. Most researchers and OEMs rely on engineering best practices to obtain repeatable data [20
]. Within this, the effect of thermal boundary condition during a charge/discharge process has been ignored. It is recommended that, future testing standards should include adherence to isothermal test conditions and a maximum tolerance level, so as to improve the quality of model parameterisation data and in turn battery states relying on them.
Immersion oil cooling using dielectric fluids is a promising technique to reduce thermal gradients within a real-world battery pack due to the large thermal mass of the cooling system [32
]. In future work, effect of this thermal management system can be investigated at module level, particularly to investigate thermal performance for motorsport applications and cycling life for passenger applications. Furthermore, similar products such as boiling fluids, have shown additional potential as a thermal runaway deterrent if the cells are immersed in it [51
]. Based on this, further work can be conducted to investigate if expensive thermal chambers, rated at European Council for Automotive R&D (EUCAR) hazard level 6, can be replaced with cheaper immersion oil baths to conduct the same experiments, while maintaining similar safety standards and allowing for better thermal control.