Energy Storage System Selection for Optimal Fuel Consumption of Aircraft Hybrid Electric Taxiing Systems

Aircraft taxiing is conventionally performed using the main engines’ inefficient idle thrust. Therefore, in line with greener aviation, the electrification of taxiing is the most viable option to reduce emissions, noise, and fossil fuel consumption during ground operations. This article studies the potential of hybridizing the conventional electric taxiing system, which is currently driven by the auxiliary power unit, with an electrical energy storage system, comprising commercial high-energy and high-power lithium-ion batteries, for the purpose of reducing fuel consumption. Hence, a power distribution optimization is formulated to minimize fuel consumption over a typical worst case taxi-out profile. Three different energy management strategies are presented for a narrow-body airplane. The optimization is performed for the selection of off-the-shelf batteries so that their impact on fuel savings can be evaluated in the early design stage. The study showed that a wide range of savings is achievable according to the selected strategy, the added weight allowance, and the battery characteristics. Considering a 180-kg added weight allowance and covering the three investigated strategies, up to 72% of taxiing fuel is saved.

weight allowance and covering the three investigated strategies, up to 72% of taxiing fuel is saved. Boolean coefficient for regeneration power.

I. INTRODUCTION
U NLESS the current scenarios change, the forecast increase in air travel will soon lead to higher levels of air pollutants. Indeed, it is expected that by 2040, the CO 2 and NO x emissions are going to increase by at least 21% and 16%, respectively [1]. Therefore, measures for reducing aircraft emissions have been adopted by both manufacturers and operators. Numerous actions have already been put into place to comply with these regulations. For example, the International Civil Aviation Organization has defined the environmental impact of taxiing as the combined effect of perceived noise on the ground, local air quality, and climate change. Logically, the airborne phase is the most demanding and represents the largest contributor to emissions and noise. Yet, ground operations also contribute, so they are also targeted for improvements.
A major role of the ground operations is taxiing. Taxiing is defined as the flight phase in which the movement of an aircraft occurs on the surface of an aerodrome under its own power, i.e., using jet engines, excluding takeoff and landing. This definition will be referred to as CT throughout the text [1]- [3]. During CT, an aircraft's engines are set to the inefficient idle thrust of 7%, with an average fuel consumption index of 0.1011 kg/s per engine [4]. Therefore, 364 kg of fuel is needed for 30 min of dual-engine CT. In [4] and [5], it is estimated that the taxiing phase can account for up to 6% of the total fuel consumption in the case of short-haul flights, and this is estimated to add up to an annual total fuel consumption worldwide of 5 million tons. In addition, the aircraft fuel cost is more than a third of airline companies' operating expenses [4], [6]. Therefore, the European Union (EU) introduced 2050 Flight Path Strategy with ambitious environmental goals to cut emissions of NO x by 90%, CO 2 by 70%, and noise levels to half of 2000 standard values [7], [8]. Besides, all aircraft movements while taxiing are required to be emission-free according to the 2050 target [9]. Hence, many solutions have been proposed which avoid using jet engines during this critical phase. In line with the more electric aircraft initiative, it is widely accepted in the aerospace community that electrification of taxiing procedure (i.e., ET solutions [10], [11]) is the most promising option for achieving emission-free ground movement. The ET solutions can be divided into on-board and external ones. Simplistically, the external ETS rely upon the replacement of diesel-based tugs with electrical or hybrid electric tractors, such as Mototok [12] and Taxibot [2], [13]. On the other hand, the on-board ETS require installation of electric motors and accompanying equipment (cables, control units, casings, etc.) inside the wheels of the landing gear that provide the traction power to move the aircraft on the ground. The on-board ETS solutions are a preferable choice compared with their external counterparts as they improve aircraft autonomy. Despite the fact that ETS represent a deadweight during flight, on-board ETS are proven to have significant block-fuel savings (fuel for the whole flight mission) and, therefore, are the focus of this article.
The most studied on-board ETS configuration is the one in which electric motors are supplied by the APU-driven electrical generator through an ATRU [14]. These solutions are being led by major aerospace manufacturers such as Safran and Airbus [2], [13] and, in this article, are referred to as CET. Although they have a lot of potential for fuel-efficiency improvements, they still rely upon the APU utilization for power supply, which consumes fuel, produces emissions, and generates noise. Nonetheless, studies have shown that block-fuel savings between 1% and 4%, and significant emission reductions can be achieved using APUs in CET [15]- [17]. The figures depend on the weight of the ETS, flight distance, and total taxiing time. On the other hand, with the goal of achieving carbon-free and noiseless aircraft ground movement, the ET drive must be powered from sources using noncarbon fuels, for instance, batteries, SCs, and fuel cells or a combination of them all [18]. Such ET solutions that use electrochemical energy as the sole powering source are going to be termed in this article AET solutions.
Having in mind that as of 2020 there is no commercially available on-board ETS, it is logical that CET solutions are the first step toward the fuel reduction through electrification, since their implementation would not require major changes of other aircraft systems and can be easily retrofitted to the current airplane models. Existing environmental control systems and jet-engine starting procedures require pneumatic power, which in the case of CET is provided by the APU. Contrastingly, the installation of AET systems would require changes to these systems and protocols, due to the fact that the APU is turned-off during taxiing procedures. Indeed, a noiseless and carbon-free AET can be introduced only once environmental control systems become electric, and the starting of the engines becomes bleedless, such as those found in the Boeing 787 Dreamliner.
Similar to the automotive industry, where hybrid electric vehicles are introduced as a bridge between fully electric and conventional gas vehicles, a hybrid ETS can bring taxiing electrification from the CET level closer to the AET level from the fuel consumption and emissions point of view. To this purpose, a second power source (i.e., ESS) is introduced with the aim of reducing the ETS dependence on the fossil-fuelpowered APU and, consequently, to minimize consumed fuel and ensuing pollution and noise during taxiing. The selected ESSs in this study are battery-based, as batteries have reached a certain maturity level and are already extensively used in various traction applications. Considering that ET is essentially a traction application, this means that the already implemented battery solutions in other vehicle types can be conveniently adapted to ET requirements. Also, it is considered that taxiing and engine warm-up phases are independent. Therefore, even though engine warm-up represents an impediment for more efficient implementation of ET, this aspect is not addressed.
In this article, the deployment of a hybrid ETS is explored considering an ESS based on batteries of different sizes, types, and characteristics. The investigation on fuel consumption reduction and total added weight is performed for a medium size aircraft following a demanding-both in duration and power required-taxiing profile as a worst case taxiing cycle scenario. Furthermore, as fuel consumption also depends on the adopted energy management strategy (EMS) [19], three different EMSs are presented in this study, which correspond to different levels of hybridization between CET and AET. Practically, the hybridized ETS resembles the powertrain of a series hybrid electric vehicle with the difference that the APU is used instead of a diesel/petrol engine, for which many techniques have been proposed that deal with fuel consumption minimization and they can be broadly categorized into controland optimization-based approaches [20], [21].
Control-or rule-based approaches require a decision tree (i.e., control law), which manages and distributes the power contributions of both ESS and APU [22], [23]. However, different control laws yield different outcomes for fuel consumption. Thus, one could argue that there is an improved control law which achieves increased fuel economy [24], [25]. In contrast, the optimization approaches aim to search through all the feasible design space of all possible APU and ESS combined power distributions such that the maximum fuel economy could be obtained, also known as global optimum [26]. Therefore, to avoid the utilization of control laws and their associated drawbacks [22], [24], [27], an optimization approach is adopted for calculating the maximum fuel economy in the early design stage, so that the system provides optimal performance for the worst case scenario.
As the primary goal of this study is to assess the battery impact on fuel consumption at the early design stage using an optimization process, an offline optimization method is selected for this study. The optimization is defined over a typical high-demanding taxiing profile known to be a worst case taxiing cycle (i.e., input to the optimization problem), and over a preselected pool of high-power (HP) and high-energy (HE) batteries suitable for more electric aircraft applications. Hence, the fuel minimization is performed for each battery, which results in a set of optimal solutions whose performance depends on the corresponding battery size (usable energy) and its specifications. This set of solutions is also assessed in terms of added weight (battery weight minus fuel saved) since weight reduction is one of the main drivers in aerospace engineering [28]. Therefore, the presented approach could be applied in the early stage design of a hybrid ETS and proposes a feasibility study on how the introduction of battery-based ESS could reduce fuel consumption during aircraft taxiing. The results of this study can be used as a battery selector tool given the set of constraints and figures of merit such as a weight allowance. Finally, it is shown how three different EMS reflect three different levels of hybridization between CET and AET from both the fuel consumption and the power distribution perspectives.
The rest of this article is organized as follows. In Section II, the model of CET is presented including the power plant model of the APU. Also, the drive cycle and EMSs are introduced in the same section. The optimization formulation is detailed in Section III, while the results are shown in Section IV for each EMS. Limitations and guidelines for future work are discussed in Section V. Finally, Section VI concludes this article.

II. ETS MODELING
The exploration of a hybrid ETS at the early design stage requires the representation of the system in terms of power and energy flows. The selected ETS architecture for the feasibility study is explained in detail in Section II, and the power and energy requirements for this particular system are presented in Section II-A. The benefits and drawbacks of this CET power system are assessed, with an emphasis on the fuel consumption, which is thoroughly explained in Section II-B analyzing emission rates as well. Finally, three levels of hybridization are defined in Section II-C.

A. Power and Energy Requirements
The MLG of an ETS for Boeing 737-400 aircraft (narrowbody aircraft) is studied. It utilizes TMs specifically designed for the ETS application [29]. The analyzed ETS consists of two high torque density gearless TMs, each mounted directly inside the wheel of the respective MLG (i.e., left "L" and right "R") [30]. The TM is an outer rotor double star three-phase permanent magnet synchronous motor powered from two separate two-level voltage source inverters (VSIs) connected to the same dc link [31], as depicted in Fig. 1. The maximum traction torque that this ETS can produce is 14 kNm with maximum continuous shaft output power at the wheels of 100 kW.
The energy and power requirements of this ET drive have been analyzed in [32] for a typical taxi-out speed cycle and Boeing 737-400 aircraft. This analysis consists in modeling the whole ETS powertrain, including aircraft longitudinal dynamics and road conditions. The drive is modeled using the forward approach, allowing for the implementation of the physical limitation of powertrain components and reproducing physical causality of the system [33]. A simplified two-dimensional longitudinal dynamics model provides the analytical framework to study the aircraft kinematic performance and the analytical method detailed in [32] is followed. The input to this method is a typical high-demand taxi-out speed cycle with a duration of 30 min (1800 s) with kinematic performance in line with those defined by the Electric Green Taxiing System [34]. The maximum achievable speed of the ET is 20 knots, with acceleration from standstill to maximum  speed to be achieved in 90 s. The main output of this model is the power required/regenerated at the dc link which is then used for defining the energy and power requirements, as plotted in Fig. 2. It can be seen that the maximum motoring power at the dc link P max,mot taxi is 144 kW, whereas maximum regeneration power P max,reg taxi is 95 kW. The total energy required for this taxiing cycle is 22.6 kWh.
It is evident that the desired ET kinematic performance cannot be achieved using an APU as the sole power source, as it has a continuous power rating P cont e of 90 kVA and peak rating P max e of 120 kVA [35]. Comparing Figs. 2 and 3, it can be seen that P cont e is exceeded during acceleration phases 10-20 knots and it is within the peak limit. In addition, the duration of those peaks is around 30 s, which is an acceptable overload time for an APU generator [35]. On the other hand, during acceleration from standstill to 20 knots, P max e rate is exceeded by 20%. Thus, the introduction of the ESS to the CET busbar (dc link) will not only improve the fuel economy, but also it will keep the APU generator within safety limits, while supporting the needed kinematic performance.

B. APU Fuel Consumption and Emissions in Taxiing
The APU fuel consumption model is difficult to obtain as a large number of thermodynamic properties of the engine are involved in energy production [36]. However, a simplified model in terms of the input-output relationship between fuel consumption and electrical power produced at the electrical generator (APU driven) is required. The intention is to analyze the impact of APU power variations on fuel consumption. This is equivalent to the power plant modeling technique extensively used in the power generation field [37] via successive tests at different loading levels, such that a fitting curve algorithm is used to fit the measured data. In this model, the APU fuel consumption rate can be expressed in terms of power production, and this relationship is appropriate for fuel estimation, having obtained the power requirements at the dc link. This approach has been practiced on an APU in [38] and this model has been adopted. Thus, the APU fuel consumption rate curve is shown in Fig. 4 and it can be characterized by the following equation: In (1), P e is in kW, and f (P e ) is in kg/s. Applying (1) to the taxiing power profile of Fig. 3 (excluding regeneration power), it is calculated that 44.3 kg of fuel is needed for CET. However, it is important to emphasize that when APU is in idle mode (i.e., P e = 0), it still burns fuel. If the APU is in idle during the whole taxi-out profile of 30 min, it burns 37.44 kg, which means that effectively for ET, only 6.84 kg of fuel is needed (effective taxiing fuel). Furthermore, as the amount of burned fuel directly affects the quantity of emitted pollutant gases, the emission savings are also evaluated in this   Table I, emission indices for narrow-body jet engines and APU are listed, whereas the total emissions for a 30-min taxiing cycle are in Table II. Considering that there is an idle fuel consumption on the APU regardless of the taxiing demand in CET, a larger amount of fuel could be saved if the APU is shut down for some time during taxiing. Besides, it is possible to set a condition where the battery is discharged as much as possible or request it to reach a certain energy value at the end of the taxiing period, so that it is ready for a new taxiing cycle. The former requires charging/swapping batteries during the ground turn-around time, while the latter requires a certain amount of fuel to recharge the battery during taxiing. Therefore, these considerations lead to the formulation of three strategies depending on the level of hybridization.

C. Energy Management Strategies
In the first strategy, the ESS and APU run in parallel and the ESS's energy at the end of taxiing is equal to that at the beginning. Next, in the second strategy, the ESS's energy is left unconstrained and fuel savings are expected to increase; yet, both the ESS and APU run in parallel for the entire taxiing period. These two strategies are commonly addressed in series hybrid electric vehicle researches with the exception that in this case, the APU could not be subjected to recurring on-off pattern [39], [40]. Instead, the APU is subject to idle mode. Therefore, a third strategy is introduced which allows the APU to be shut down at the start of the taxiing cycle for a certain percentage of time, during which ESS solely provides the taxiing demand. This strategy is the closest to the top level (AET) of taxiing electrification, whereas the first two strategies are closer to the bottom level (CET). In Section III, the optimization formulations for these strategies are detailed.

III. OPTIMIZATION FORMULATION
For the purpose of the optimization formulation, the hybrid ETS's architecture is represented as a dc power system whose power flow diagram is illustrated in Fig. 5. In this formulation, it is assumed that the components are highly efficient devices (as high as 96%), meaning that the system's losses are considered negligible. This is a fairly reasonable assumption as nowadays' electrical machines and power electronics actually achieve high efficiency values [41]- [43]. Since the battery-based ESS is unable to meet the load requirements, due to the battery's rating limitations, an add-on device is introduced. This device has bidirectional capabilities, and it could be envisioned as an additional battery or SC depending on the design criteria, or it could include a bidirectional converter to provide power exchanges with the rest of the aircraft's electrical power system [44].
The dc link of the ETS is represented as a single bus (ET bus), which receives the unidirectional power P e coming from the electrical generator (through an ATRU). All the other components are bidirectional. In this case, P S , P M , and P W are the power values of ESS, traction load, and add-on device, respectively. The regeneration power from the traction load can be injected to the ET bus to help charge the ESS via P S , and any remaining power can be directed to the add-on device. Also, the ESS can be discharged to provide traction power either jointly with the APU, or by itself whenever it can drive the taxiing load without APU assistance (P e = 0). Due to the inherent dynamic limitations of the APU's electrical generator and ESS, the add-on device can supply or receive power (P W ) when convenient, e.g., improving energy utilization through another battery, SC, or to the aircraft's power system network.

A. Main Formulation
The objective of the optimization formulation is to minimize the taxiing's fuel consumption F for a taxiing profile over a period of time [1, T ]. The outcome of the optimization is the optimal power distribution between the ESS (battery and add-on device) and the APU's electrical generator. The minimized fuel consumption is associated with the specific battery considered in the optimization, such that a hybrid ETS comprising this battery would perform the obtained fuel economy. The traction load profile P M is known in advance, that is, P M is an input. The required traction power P M must be supplied by the APU (P e ), the ESS's battery (P S ), and the ESS's add-on device (P W ) as observed in Fig. 5, so that P e , P S , and P W are unknown variables to be solved for. The taxiing period is divided into 1800 time steps corresponding to a 30-min taxiing (T = 1800); thus, each time step k has a duration of 1 s, i.e., k = 1 s. The APU engine's fuel consumption is assumed to be a function of the APU's power produced at the electrical generator terminals as shown in (1) and Fig. 4. Then, for every time step, the fuel consumption is defined as F(k) = k · f (P e (k)). The optimization solution consists of optimal values for fuel consumption F, battery power P S (k), add-on device power P W (k), APU generator power P e (k), and ESS's energy E(k) for each time step k. Therefore, the optimization objective is as follows: The constraints for this formulation are detailed below. The first constraint is the power balance of the ETS depicted in Fig. 5, i.e., for every time step, the demand P M (k) is supplied either by P e (k), P S (k), P W (k), or a combination of them. For the ETS, the power balance constraint can be expressed in the following equation, assuming the following sign convention: P S (k) is (−) for charging and (+) for discharging conditions; P M (k) is (+) when motoring and (−) when regenerating, and P W (k) is (−) when absorbing and (+) when injecting: The dynamics of the APU's electrical generator are represented in the following equations. Here, P max,ru e and P max,rd e are the APU's generator maximum ramp-up (positive rate of change) and maximum ramp-down (negative rate of change) powers Assuming that the battery's charging and discharging efficiencies are η ch and η dsch , respectively, the ESS's energy E is calculated in the following equations, and it is kept within maximum (E max ) and minimum limits (E min ) as shown in the following last equation: The upper and lower limits are selected to be 90% and 20% of total battery energy (E total ). Consequently, the depth of discharge is 70% and the battery's usable energy E bat is calculated as 0.7 E total . Finally, the last constraint regarding ESS's initial energy is reported in the following equation, where E 0 is the initial energy value and it is also constrained by (7): The rest of the constraints are related to the upper and lower boundaries for all the power variables. For the APU power, constraint (9) applies, where P max e is the maximum power that can be drawn from the APU's electrical generator. In (10), the battery's power is constrained to P max,ch S and P min,dsch S , which are the maximum charge and discharge powers, respectively, whereas (11) constrains the ESS's add-on power to a maximum value P max Given that the regeneration power can charge the battery, but the battery's charging is limited to P max,ch S , it is possible that some regeneration energy is absorbed by the add-on device. However, the add-on device could also inject power back to the ETS unnecessarily since P W (k) participates only in the power balance (3), and it is not accounted for in the optimization objective (2). In other words, P W (k) represents a degree of freedom for which there is no penalty in the fuel minimization objective of (2), leaving open the possibility of finding a suboptimal P W (k) and total fuel consumption, or even reaching an inadmissible solution. Therefore, an additional penalty term involving the add-on device power is required in (2), such that the add-on utilization is minimized. In this case, regeneration power is used as much as possible to charge the ESS (minimum amount of energy directed to add-on device), and unnecessary power from the ESS's add-on device is either avoided entirely or minimized if essentially required. Then, the problem's objective (2) becomes as follows: While the first term in (12) minimizes the total taxiing fuel, the second term is equivalent to consider maximization of battery's charging for the time steps where there is regeneration energy available.

B. Energy Management Strategies
At this point, the three energy management strategies that depend on the level of hybridization for taxiing electrification can be defined and detailed. 1) Sustained Condition: Both the APU's generator and ESS supply taxiing power for the period [1, T ]. ESS' battery has a sustained condition, i.e., the battery's energy at the end of the taxiing period is equal to that of the initial value E 0 , as in the following equation. Hence, ESS is charged and discharged during [1, T ] 2) Nonsustained Condition: Both the APU's electrical generator and ESS supply traction power during taxiing period [1, T ], but the ESS's battery does not have sustained condition, i.e., the battery's energy at the end of the taxiing period is unconstrained. In this case, the energy at the end of taxiing is not equal to that at the beginning; hence, the battery is predominantly subject to discharging.

3) APU Turned on at t A :
The taxiing period is divided into two segments. In the first segment, APU is shut down and ESS alone provides traction power until its energy is down to its minimum E min , which happens at time instant t A . For the second segment, from t A to T , the APU is turned on, and both APU's electrical generator and ESS supply taxiing power for the rest of taxiing period, i.e., t A ≤ k ≤ T . For strategies (i) and (ii), P W (k) is constrained to be unidirectional, i.e., only absorbing excessive regeneration power, because APU and ESS are capable of supplying traction power at every moment. Then, P W (k) is constrained to be 0 whenever traction load is not in regeneration mode. Let r t (k) be a Boolean coefficient which is true (r t (k) = 1) only on the intervals of time where there is regeneration (P M (k) < 0), and r t (k) = 0 elsewhere. Thus, r t is a vector of coefficients precalculated a priori. Then, (11) is rewritten as follows: Given (14), it is not possible to have regeneration and ESS discharge simultaneously as it is counterintuitive. Therefore, (10) becomes as follows: On the other hand, for the third strategy (iii), the APU is shut down during the first segment 1 ≤ k ≤ t A so that there is no fuel consumption at all, and then it is turned on for the rest of the time t A ≤ k ≤ T . It is assumed that the rest of aircraft loads are supplied by a non-APU nonfossil-fueled secondary generation system during 1 ≤ k ≤ t A (other more electric aircraft systems). Because the ESS has inherent power and energy limitations, the add-on device is the only component that can absorb or supply excessive and deficient energy requirements during the first period 1 ≤ k ≤ t A . Therefore, the solution for P W (k) provides valuable information to design and size the add-on device during 1 ≤ k ≤ t A . Besides, it is also possible that the ESS is sufficient to drive the traction load during all the period T (if t A = T ).
At this point, it is convenient to summarize the framework of this study as in Fig. 6. Initially, the different levels of hybridization of ET are introduced according to the fuel economy increase. Then, the three energy management strategies are mapped to the level of hybridization scale. Afterward, inputs and constraints of the optimization formulations are presented. In Section IV, the case study and results are presented.

IV. RESULTS
In this study, the feasibility analysis for the implementation of a hybrid ET for reduced fuel consumption is explored using the three energy management strategies detailed in the previous section. The results for each strategy are presented and discussed in terms of saved fuel. The fuel economy is compared with the CET fuel consumption, i.e., 44.3 kg, due to the reason that CET is the highest level of technology readiness in taxiing electrification currently found and there is no similar hybrid ET benchmark reported in the available literature. The analysis includes the tradeoff between the amount of saved fuel and the weight of the ESS, such that some options are recommended. A weight allowance m A = 180 kg is introduced as the suggested weight limit for the implementation of the hybridized ETS. This allowance corresponds to almost half of the fuel required for CT over a 30-min taxiing cycle. It is true that the battery's added weight will lead to an increased fuel consumption during flight; however, it has been shown that block-fuel savings could be achieved for an ETS's weight of up to 1000 kg [16]. Taking into consideration that commercial ETS's weight approximately 400 kg, the suggested battery weight should not lead to an increase in total block-fuel consumption. Therefore, the analysis excludes the airborne phases, as it is out of the scope of this article. In addition, the main target for this study is optimizing the ground fuel consumption and emissions. The initial energy of the ESS for the first two strategies is set at 75% of E max to allow some charging from regeneration power and the APU, while the third strategy starts at 95% of E max , where the ESS is operating in standalone mode during the first segment of the taxiing cycle. Also, the battery's charging and discharging efficiencies η ch and η dsch are assumed to be 1.0. This assumption allows an unbiased comparison on the fuel reduction potential of different types of batteries, without being compelled to perform further simplifications due to lack of detailed data to build accurate battery models for all brands, types, sizes, and power characteristics.
Forty-two cells which are available on the market have been selected from five different brands, specifically: Kokam HE, A123, Saft, Panasonic, and Kokam HP. These battery brands have been selected for their technical and economic benefits for traction applications, e.g., electric and hybridelectric vehicles [45]. The characteristics of each brand are presented in Table III including the number of cells, capacity range, the power to energy ratio (P/E) range, C charge rate (the rate at which battery is charged relatively to its capacity), and energy density range. The power to energy ratio (P/E) has the same value as maximum continuous discharge rate. Furthermore, only continuous discharge operation is considered because there is no consistency in datasheets regarding maximum pulse discharge neither in terms of pulse duration nor relaxation period required after the pulse. By considering this conservative approach, i.e., continuous operation only, the battery's lifetime will be longer which is important for taxiing applications. The selection covers both HE and HP cells as well as three different chemistries (lithium nickel manganese cobalt oxide NMC, lithium iron phosphate LFP, and lithium nickel cobalt aluminum oxide NCA).
Considering that the taxiing energy required is 22.6 kWh, and that optimization is computationally intensive, a set of battery usable energy values (E bat ) in the range between 5 and 30 kWh (steps of 5 kWh) are proposed; thus, six energy sizes per cell type are considered. As a result, a pool of 252 different batteries is generated and the corresponding weight for each battery is reported in Fig. 7. The system's weight increases with the system rating and differs based on the battery brand and type. Using information from Table III, each battery's maximum continuous charge (P max,dsch S ), discharge power (P max,ch S ), and mass are calculated. The optimization formulations were solved using CPLEX Optimization Studio 12.9.0 [51] running on a Windows High Spec PC Intel Xeon 64-bit processor at 3.60 GHz with 32 GB RAM. Due to the reason that all the constraints are linear, the quadratic objective in (1) was linearized [26], [52] in order to exploit the advantage of solving the optimization as a mixed integer linear programming (MILP) problem [53]. In some cases, optimization is not feasible due to the battery's limitations, and these cases are discarded. The computation time limit to solve an off-line optimization was set to 3.5 min, the average solving time observed is 97 s (1.6 min), and 30% of the optimizations were completed in under 13.5 s (0.2 min), which is adequate for an early stage design problem. A discussion of the applications on real-time operational taxiing is presented in Section V-D. Although m A is introduced as a suggested weight allowance, the results are presented for a broader range of battery weights depending on their E bat , such that the influence of a different m A can still be appreciated and further explored if required. The main findings for the rest of the options are discussed below.

A. Strategy (i): Sustained Condition
In strategy (i), the APU's electrical generator supplies a certain amount of energy to bring the battery's energy to a specific value at the end of taxiing, i.e., E(k = T ) = E 0 , and some regeneration power also charges the battery when possible. The optimization results are illustrated in Fig. 8; Fig. 8(a) depicts fuel saved with respect to E bat , while Fig. 8(b) shows fuel saved in terms of battery's weight. In addition to the optimal results, a curve that best fits the optimal points has been included. The expressions for these curves are quadratic and are given in the following equations, where f * (E bat ) and f * (m bat ) represent the optimal fuel saved in kg, E bat is in Although the battery sizing exercise depends mainly on the fuel economy, usable energy, weight, P/E ratio, and brand, (16) and (17) aim to provide a first approximation for battery sizing in terms of a single parameter, where the rest of parameters are implicitly represented. Because the essential sizing parameters are energy, weight, and P/E ratio, the results in Fig. 8 can be fitted in terms of these three parameters to obtain a more accurate battery size estimation. This approximation is a two-variable fuel saved function in terms of E bat and m bat , as in the following equation, for which its coefficients change depending on the P/E ratio, as tabulated in Table IV: Expression (18) and coefficients in Table IV are valid on the vicinity of the batteries selected for each P/E ratio; E bat is in kWh, and m bat is in kg. In other words, (18) is the result of curve fitting exercise with the highest goodness-to-fit ratio (e.g., SSE = 0.0238, R 2 = 0.9859, and  20). For batteries whose useful energy is below 5 kWh, the total APU's fuel consumption is reduced by less than 1 kg, as depicted in Fig. 8 (options shown with "o" marker). However, even with a 30-kWh battery (six times larger), the saved fuel amounts to less than 2 kg, being 1.68 kg the maximum saving from an A123 30-kWh battery that has a mass of 307.5 kg (see "∇" blue markers in Fig. 8). The relative fuel reductions (compared with CET) fall in the range between 2.0% and 4.0%. Below m A (weight allowance), a 30-kWh Kokam HE battery with P/E = 2 and 164.0 kg provides the maximum fuel saved of 1.54 kg, just 8.3% less fuel than the A123 30-kWh batteries at almost half of the weight.
The minimum saved fuel is 0.93 kg corresponding to a 5-kWh Kokam HE battery with P/E = 5 and 41.2 kg. In terms of the total added weight, the minimum corresponds to a 5-kWh Kokam HE with P/E = 5 and 41.0 kg. For the rest of the 5-kWh cases, battery mass is between 27 and 65 kg and it achieves fuel reductions of 1 kg at most (i.e., 26-64 kg of total added weight). The Kokam HP batteries display an optimal fuel consumption of 1.42 kg at most for a 20-kWh battery with P/E of 8 and a weight of 179.3 kg.

B. Strategy (Ii): Nonsustained Condition
In strategy (ii), the battery initial energy (E 0 ) is the same as in strategy (i), but its final value is unconstrained. The optimization results are demonstrated in Fig. 9; Fig. 9(a) represents fuel saved versus E bat , whereas Fig. 9(b) displays fuel saved to m bat . Similar to strategy (i), a quadratic equation for fuel economy approximation is used to describe a curve fitting for these results, as listed by the following equations: bat + 0.0400m bat + 0.3382. (20) In the same manner as in (18), the two-variable fuel saved function in terms of E bat and m bat is used for this strategy, and the values for its coefficients are summarized in Table V.
The trend of saved fuel is generally proportional to the battery energy rating, such that the minimum saving is corresponding to the lower kWh values and vice versa. At 5 kWh, the maximum saving is equal to 2.1 kg of fuel. Conversely, the maximum saving is 16% of the CET level for batteries rated at 30 kWh. Having in mind the weight allowance m A , two batteries-Kokam HE and Panasonic-rated at 30 kWh emerge, where 6.3 kg of fuel is saved, weighing 166 and 163 kg, respectively, i.e., a saving 8.7% less than the maximum possible. For the total added weight criterion, a 5-kWh Kokam HE with P/E = 2 and 39 kg of total added weight has a saving of 2 kg of fuel.

C. Strategy (Iii): APU Turned on at t A
In this strategy, the APU is completely turned-off during first taxiing period and ETS is fully powered from the ESS. In Fig. 10, the APU activation time is shown for all cases with respect to E bat and m bat . The instant t A when the APU turns on depends on the energy content of the battery as well as on its power rating. For example, two Kokam HE batteries with the same E bat of 5 kWh and two different P/E ratios yield different APU activation times: one with P/E equals to five and it turns on after t = 605 s, whereas the one with P/E of two becomes operational after t = 1374 s. This is because the discharge power of the latter one is lower and, consequently, the discharge time is higher for the same energy content. Also, it can be concluded that there are cases when the  Even though the APU generator and ESS are managed in optimization mode after t A , most of the fuel savings are obtained before the turn-on time. The more delayed t A is, the higher the fuel savings are. Thus, the turn-on time t A can be considered as a preliminary estimate of the expected optimal fuel savings, and the following equation approximates this time instant as a function of E bat and m bat . In Table VI, the coefficients used in the following equation are provided: bat . (21) In this strategy, in addition to the optimal saved fuel and battery weight criteria, the battery must be capable of providing the taxiing power requirements alone, i.e., P max,dsch S > P max taxi . Batteries that fulfill this condition are introduced as category I (cat. I). In the other cases, when a battery is not sufficient (cat. II), an auxiliary bidirectional source, namely the add-on device, must be added to assist the battery and its weight also should be assessed. Thus, further analysis on the categories cat. I and cat. II is addressed.

1) ESS With Battery Only (Cat. I):
The simple following formula can be used to analytically identify batteries that are within this category: Considering the usable energy range used in this study (i.e., 5-30 kWh), the minimum P/E ratio that fulfills condition (22) is 4, with E bat of 30 kWh. As the P/E increases, the set of possible usable energies increases as well. For example, all batteries with P/E ratio equal to 20 and with an E bat of more than 10 kWh comply with (22). Brands that are in this category are the high-power A123 batteries (∼ 83% of selected A123 batteries), most of the Kokam HP batteries (∼64%), and few Kokam HE batteries (∼9%) that exhibit P/E ratio of 5 and have E bat of 25 kWh or 30 kWh. All batteries from cat. I and their associated fuel savings are represented in Fig. 11.
Contrary to the previous two strategies (i) and (iii), a clear saturation trend is noticeable. Practically, all batteries with weights higher than 286.5 kg result in 100% fuel savings. In fact, a significant number of options yield savings higher than 95%. The lightest battery that provides 95% of fuel savings (i.e., 42.08 kg of fuel) has a mass of 205 kg, 14% higher than the weight limit. This is Kokam HE battery with P/E of 5 and usable energy of 30 kWh.
On the other hand, there are only a few batteries that are within the maximum weight limit. Most of them are 10-and 15-kWh A123 batteries and 10-and 15-kWh Kokam HP with P/E ratios of 8 and 20. The battery that saves most fuel (∼70%) within m A limitation is the 180-kg heavy Kokam HP with P/E of 8 and 20 kWh. The fuel savings of cat. I batteries with respect to battery weight and energy rating can be approximated using (18) with coefficients from Table VII. In this category, the minimum total added weight is achieved with Kokam HP, E bat = 10 kWh, P/E of 20 and 101.5 kg.  In this case, the APU turns on at 605 s (i.e., 30% of the taxiing is performed on ESS itself); thus, a significant amount of fuel is burned 28.4 kg (15.8 kg savings compared with baseline scenario CET, i.e., 44.3 kg). Therefore, the minimum total added weight amounts to 85.7 kg. Similar results are obtained using two Kokam HP batteries with the same usable energy and P/E of 20, resulting in 93.6 and 86.5 kg of total added weight.

2) ESS With Both Battery and Add-On Device (Cat. II):
In cat. II, there are predominantly batteries created from HE Saft, Panasonic, and Kokam HE cells. Also, a small number of HP batteries with low energy content are presented, as shown in Fig. 12. Since HE cells typically exhibit excellent energy densities, these solutions are lighter compared with cat. I batteries. Also, it might appear promising at first glance that almost 100% of fuel savings can be achieved with lightweight batteries. However, as mentioned earlier, batteries in this category are not powerful enough to sustain ET alone when the APU is turned off. Therefore, they must be integrated with some other source to bridge this power gap and the weight of that device must be properly assessed and included.
To this purpose, P W (t) is analyzed on each case. The main outcome of this analysis is the obtention of the required maximum charge-and discharge-powers, and the total energy provided by the add-on source. Then, these parameters are used for sizing this additional device. The necessary energy is calculated using the maximum swing in energy profile [54], as demonstrated in Fig. 13.
Obviously, the energy and power requirements of the add-on device are less stringent for the batteries with higher energy content. For example, in Table VIII, the requirements of an add-on device are shown for all Panasonic batteries. Furthermore, for HP batteries, the requirements are even more relaxed in terms of energy and power rating, as can be observed in Fig. 14.  The add-on device could be any electrochemical device such as SCs, lithium-ion capacitors [55], and new battery types such as that proposed in [56], consisting of a hybridization between HP and HE batteries. It is out of the scope of this article to analyze all possible options; thus, an SC is going to be considered as it is the most common device to be integrated with  batteries due to advantages of that configuration [57], [58]. A Maxwell SC [59] cell with a capacitance of 3000 F and with energy and power density of 6 Wh/kg and 5.9 kW/kg, respectively, is selected as the building component of the SC stack. Obviously, due to its inherently low energy density, the SC weight is extremely high when the energy requirement is high as is the case with the Panasonic batteries. In Fig. 15, the fuel savings are presented with respect to the total ESS weight (i.e., battery and add-on device weights), whereas the coefficients of the fuel estimation function (18) are presented in Table IX. By inspecting Fig. 15, only five batteries and SC combinations fall below m A (180-kg limit). All battery types are HP ones with P/E ratio of 20 and E bat of 5 kWh. Therefore, the best result is achieved when an HP battery with low energy   Table X indicate that the total added weight is between 46.16 and 60.7 kg, which are all lower than the minimum added weight of cat. I of 85.5 kg (∼50% less). Thus, in this strategy (iii), it is more beneficial to use HP batteries with low energy content in conjunction with SCs.

V. DISCUSSION AND RECOMMENDATIONS
Having obtained the results presented in Section IV, Section V provides a four-part discussion. The first part considers the implications of selecting one of the three energy management strategies. Then, the influence of different charging/discharging efficiencies in the validation of the results is briefly addressed. Due to the reason that only one high-demanding taxiing cycle has been examined, the impact of other drive cycles is reviewed. Finally, a brief account on the potential for the optimization formulation presented in this article, for operational applications in real-time taxiing, is assessed.

A. Energy Management Strategies Analysis
Strategies (i) and (ii) are conceived as a potential early step in hybridization toward AET, bringing closer together CET and AET. Besides fuel reduction, ESS also helps to achieve taxiing kinematic requirements without overloading the APU's generator. These strategies can be implemented on current aircraft where the APU is an integral part of the aircraft power generation. Alternatively, strategy (iii) assumes that other aircraft systems are supplied by nonfossil fuel power sources (next generation of aircraft). In fact, it is conceived as the highest level of hybridization among all strategies (closest to AET) because it allows standalone battery mode until the ESS reaches the lowest permissible energy level, the instant when the APU turns on.
The first two strategies depict the ESS as a single battery whose purpose is to diminish the amount of taxiing fuel supplied by the APU, which is always turned on and also supplies other aircraft systems (different from ETS). Therefore, the theoretical maximum saved fuel in these two strategies is limited by the effective taxiing fuel (6.84 kg), i.e., 16% in terms of CET fuel consumption. The fuel reductions in strategy (ii) are larger than in (i), due to the reason that in (i) the APU has to deliver a certain amount of energy to achieve a prespecified energy content at the end of the taxiing period. Consequently, the fuel reduction in (i) is less than 2 kg in all of the cases, while it starts at 2 kg and reaches approximately 6.8 kg in (ii). In terms of percentage of CET fuel (baseline), strategy (i) saves between 2.0% and 4.0% compared with (ii) which saves between 4.0% and 16% of taxiing fuel. In both strategies without exception, the minimum total added weight [27.2 kg in (i) and 39 kg in (ii)] corresponds to the lightest batteries but the least saved fuel. The solutions with the highest fuel reduction unambiguously result in higher total added weight.
More detailed analysis could be performed by obtaining the gradient of the optimal saved fuel function defined in (18). For example in strategy (i), at half of m A (battery weight), an increase of 1.8 × 10 −3 kg of fuel per kg of weight is expected for a 10 kWh battery with P/E of 20, which is a sensitivity parameter relating two parameters, i.e. saved fuel vs. battery weight, for a given usable energy. In general, this small number applies to all strategies, which means that the optimal saved fuel depends mainly on the P/E ratio and E bat , rather than on the battery's weight. For instance, at approximately 7.5 kWh, the sensitivity is 31.4 × 10 −3 kg fuel/kWh energy for a weight of 90 kg (half of m A ) and a P/E of 20.
In terms of battery size, strategy (i) introduces the least added weight but the least reduction in fuel and consequently emissions. However, the benefit is that the battery requires no time-consuming charging after a taxiing cycle, due to the sustained charge mode, which could save turn-around time and charging electricity cost. Furthermore, with the possibility that the ESS could be switched to a different bus to provide power to other electrical systems during the flight, strategy (i) could be conveniently introduced. In the case of strategies (ii) and (iii), the battery could require a steady charge after taxiing just before switching to other bus. Nevertheless, strategies (i)-(iii) must be carefully justified in terms of regulatory compliance and economic advantages. However, these analyses can be extended in the future. In terms of emission (pollution), the first two strategies provide the figures shown in Table XI for the minimum total added weight scenario. From this table, it can be seen that strategy (ii) has lower emissions compared with (i) (≈1.5% lower). It also shows that strategy (iii) reduces around 18% of emissions compared with (i). All of this reflects to cost savings related to noise and pollution penalties that could be avoided. In some cases of strategy (iii), an add-on device is required to supply the traction demand completely. This strategy provides savings in the order of 10-44.3 kg, which are promising values considering the absence of noise (APU off). However, opposite to (i) and (ii), the total added minimum weight is 85.5 kg for cat. I and 46.1 kg for cat. II that are integrated with SC. Even though the total added weight is the highest in this case, from an operational perspective, this approach is the best as it saves the highest amount of fuel, it has the highest pollution reduction (Table XI) and it is practically noise-free at the beginning of the taxiing phase.

B. Different Charging/Discharging Efficiencies
The battery's efficiency depends on multiple factors, such as the state of charge (SoC), operational temperature, and discharge/charge rates. Hence, it is compulsory to have a significant amount of specific information on the batteries' characteristics to develop a comprehensive efficiency model. The assumption of ideal and constant charging/discharging efficiencies (i.e., 1.0) allows an unbiased comparison of the potential of several types of batteries under the same conditions for the implementation of a hybrid ETS. In general, the utilization of a common discharge/charge efficiencies (other than 1) for all batteries would offset all the results toward lower saved fuel regions, or increased fuel consumptions, regardless of the battery type, due to the reason that discharging/charging would account for power losses mainly compensated by the APU's electrical generator. This offset in fuel consumption is roughly constant in magnitude, so that approximations in (16)- (20) are still valid. On the other hand, the optimization assessment with different efficiencies is possible; however, it requires a set of detailed parameters for each of the 252 batteries considered in this article, in order to perform the comparison on a more accurate basis. This task could be challenging as several manufacturers do not provide detailed information on their batteries, thus preventing the development of an efficiency model for each separate battery.

C. Impact of the Drive Cycles
It is widely accepted in the automotive sector that drive cycles are useful tools in assessing the performance of vehicles in terms of pollutant emissions and fuel consumption. For example, in the EU, the legislative drive cycle called New European Drive Cycle (NEDC) is in use since 2000, whereas in the United States, numerous cycles exist such urban dynamometer driving schedule (UDDS) and New York city cycle (NYCC) [60]. All these cycles try to represent common patterns encountered during motorway and urban (city) driving. Unfortunately, such legislative cycles do not exist for ET applications yet. However, most of the ET manufacturers designed their systems to follow electric green taxiing system (EGTS) requirements, namely: 1) Max. speed of 20 knots that should be achieved in 90 s.
2) Speed of 10 knots in 20 s for active runway crossing.
3) Breakaway torque at 1.5% grade at MTOW. The cycle used in this study is constrained by these typical ET taxiing speeds and accelerations. Furthermore, it includes the pushback phase, during which the speed is negative (aircraft is reversing), at the beginning of the cycle, and the queuing phase before takeoff at the end of the cycle. The cycle used in [61] follows similar kinematics. In terms of length, the most conservative taxiing time is considered in this article, i.e., 30 min compared with average taxiing time of 15 min [62]. This ensures that ESS is assessed for the worst case scenario, i.e., the longest and highest power demands.
Nonetheless, the driving pattern of ET can affect the EMSs and optimization itself. As pointed out in [63], there are two types of taxiing cycles, unrestricted and restricted. During the unrestricted one, the aircraft taxis to the runway without any stops. On the other hand, during the restricted taxiing cycle, multiple stops occur due to traffic and active runway crossings. Thus, during one cycle, more than one HP demanding event can occur, leading to different optimization outcomes. As it can be seen from Fig. 3 (between 158 and 220 s), the most power demanding event is the acceleration from 0 to 20 knots (indicated as a 0−20 ), during which 144 kW is needed for 30 s. During this occurrence, the APU is at its limit of 120 kW, while the ESS provides the rest of 24 kW for only 10 s. Therefore, for instance, if during strategy (iii), a 0−20 occurs just when the APU turns on, due to optimization constraints, the power will be provided by the add-on device and APU, as the battery is at its minimum allowed energy when the APU is turned on. However, the battery is still at 20% of its maximum level, and depending on its size, it can provide either a fraction or all 24 kW for a time period of 10 s. So, in this case, additional constraints are required to favor supplementary battery discharge rather than using the add-on device.

D. Applications on Real Time Operational Taxiing
While the optimization formulation presented in this article (see Fig. 6) provides a lower boundary for fuel consumption in the early design of a hybrid electric system, it can be used as a starting point to establish a power management system that optimally distributes the online electrical power (real-time taxiing operation). In this case, having deployed the most promising option from the available ones depicted in the results of Section IV, the taxiing operation requires a control system that efficiently distributes energy between APU and ESS in order to meet the actual taxiing load demand. Given the extensive use of predictive control methods (e.g., receding horizon control), an operational power management could use a two-stage control methodology consisting of some form of EMS optimization and a continuously modulated mechanism (e.g., PID, predictive control) in order to exploit predriving and online power allocation [22], [40], [64]. In the former stage (predriving), the system could use a ground movement forecast [4], [65], [66] (e.g., optimal speed profile generation), which is updated periodically based on the changes in airport operation. For example, on a quiet day the initial forecast could even be exempt from further changes as the runway has reduced traffic. In the latter stage (online power allocation), control techniques should be rapid and effective to provide fuel economy along sudden traction demand changes.

VI. CONCLUSION
This article explored the feasibility of hybridizing the conventional ETS supplied by the APU with an additional ESS. The analysis is performed on a pool of several types of batteries and, in some cases, it is complemented by an SC add-on device. Three different energy management strategies, based on the level of hybridization, were studied for a typical ET duty cycle of half an hour. The first strategy considers constrained battery's energy at the end of taxiing, while the second strategy relaxes this condition and allows battery depletion to a minimum energy level. In the third strategy, APU is turned off in the first period of the taxiing cycle and then a relaxed energy battery condition is ensured for the rest of the cycle; the percentage of time during which the APU is shut down depends on the battery's characteristics. These strategies were formulated as a taxiing fuel minimization problem constrained by the APU characteristics, battery ratings, and power flow balance with the aim of selecting a feasible energy storage for a hybrid ETS.
The results showed that the highest fuel reductions are achieved for batteries with a HE rating in the first two strategies. These are the closest to conventional ET because they are majorly reliant on the APU operation during taxiing. It has been found that up to 14% of taxiing fuel can be saved, when compared with conventional ET, by implementing an ESS of 163 kg, which is below the proposed weight allowance of 180 kg. However, considering total added weight, i.e., the difference between battery weight and saved fuel, the best results are achieved for HE batteries with low usable energy ratings, resulting in a 4.4% reduction only. In the case of the third strategy, it is possible to achieve a 67% reduction for a standalone battery option within the weight allowance limit. For a 100% fuel economy, i.e., the APU is turned off, an ESS of 245 kg must be implemented, which is 35% heavier than the suggested weight allowance. The most promising implementation in terms of total added weight in the third strategy demands an ESS consisting of a battery with HP and a low usable energy rating in combination with an SC.
The results can be further explored in terms of regulatory compliance and economic benefits, airborne fuel consumption implications, different storage types for the complementary add-on device, varying taxiing cycles, and environmental operating conditions.