Variable-Voltage Bus Concept for Aircraft Electrical Power System

This paper deals with the innovative “variable-voltage” bus concept for future more-electric aircraft platforms. Using new functionalities and opportunities offered by innovative sources actively controlled by power electronics, there is an opportunity for a significant increase of their output power (within powertrain capabilities, machine thermal limits, and power converter maximum current) to supply increased load demands under certain conditions. This paper investigates applications of this concept to satisfy power demands for the wing ice-protection system for business jet. The study includes a detailed study into the control challenges of the proposed approach and suggests corresponding theoretical solutions. Furthermore, the controller design aspects are explored with considerations to achieve stable operation. The concept is tested in simulation environment and validated through experimental studies. This paper might be of interest for the researchers and engineers focused on innovative solutions for future aircraft platforms.


I. INTRODUCTION
he design of next-generation aircraft platforms has been targeting for more efficient and hence more environmentally friendly, greener solutions to ensure sustainable future. In more-electric aircraft (MEA), this is achieved by replacement of on-board systems powered by hydraulic and pneumatic energies by electrically powered ones [1][2][3]. Because of this transfer, the total installed on-board electrical power demand in future aircraft will significantly increase. One of these "new" loads on-board MEA is the wing ice-protection system (WIPS) which employs electrically produced heating using resistive thermal mats. In larger aircraft this load system can typically demand about 125kW per wing [4] and this is within the available poser budget. Conversely, for smaller platforms such as regional and business airplanes, the total generated on-board power may not be enough to cover the WIPS demand. Therefore, hybrid solutions are considered where these assumes employment of bleed air from the engine in combination with the electrothermal mats. Some other new/existing designs consider electro-expulsion or electro-mechanical methods of removing formed ice [5], [6] which requires the least amount of energy but are not very reliable and may lead to the wing structural damages. Others have chemical de-icing systems [5], which are of limited availability during a flight cycle, and piezoelectric based technologies that has drawbacks in electromagnetic interference and wing structural fatigue [5].
In current fully electrical WIPS solutions (eWIPS) the electric power distribution is arranged via the main buses of an aircraft electrical power system (EPS) [7][8][9]. As illustrated in Fig. 1(a), the eWIPS is connected to a fixed high voltage bus along with other electrical loads. Local power converters/controllers can be used to vary the amount of power to the eWIPS depending on the temperature requirement and zonal heating capabilities [10]. One thing to note is that both ac and dc power transmissions are viable as the eWIPS is mainly resistive in nature. For the topology of this EPS, the voltage levels should be kept constant as the bus may be shared with other more sensitive electrical loads. Hence, the bus voltage level and total resistance of the thermal mats limit the maximum eWIPS power.
There are possible control methods of transmitting power from the power source to the eWIPS if the primary sources are based on innovative topologies such as permanent magnet machine (PMM) controlled by active front-end power electronic converter [3,11]. In literature, [12] has described a vector based control scheme including its controller design steps for PMM based aircraft starter-generator systems. [13] proposed a control scheme based on voltage angle regulation for high speed PMM based drive systems. The control scheme allows full utilisation of the dc bus voltage and improved control stability using d-axis feedback loop. A similar control concept was also presented in [14] based on voltage angle with six step operation and the inclusion of d-axis current loop for added controller stability. De-centralised droop based control strategy for paralleled multi-generator system for MEA has been looked at in [15] and [16].
If the bus voltage level is not restricted, there is a natural opportunity to increase its output voltage by narrowing the region of flux-weakening operation within the current limits of power converter and torque capabilities of the power train T (accessories gearbox). Within the current limit, this would mean that there is reduced reactive current component and hence increased allowance for active part while not affecting the machine thermal performance. This can be achieved via a dedicated bus for the eWIPS as seen in Fig. 1(b). When the eWIPS demands for very high power level, e.g., if anti-icing required, then "above the rated" voltage levels can be set and this can be called variable voltage, or "voltage wild" bus. Furthermore, such configuration eliminates the need for local eWIPS power converters and may yield potential reduction in overall aircraft weight. The drawback for this concept is that the increase in bus voltage also increases the semiconductor losses assuming same current operation levels and the need for appropriate circuit protection devices. Hence, the converter, cooling system, and circuit protection devices may have to be overrated to compensate for the losses. Moreover, the voltage variation should be within the maximum limits of the electrical machine as partial discharges may occur.
This paper investigates the theoretical aspects of "voltage wild" EPS concept, including overall control approach, controllers design and stability analysis. The key theoretical findings are supported by the EPS simulations at different operational modes and verified by a mock-up EPS. Details of the studied EPS is in Section II and its proposed control scheme in Section III. In-depth controller design and analysis are reported in Section IV, followed by both simulation and experimental results in Section V and VI respectively. Section VII concludes the paper.

II. POWER SYSTEM
The base EPS explored for this study is illustrated in Fig. 2. ωr is the mechanical speed, iabc is the three phase stator current, C is the dc bus capacitance, and Rw is the eWIPS load. Here, the generator is assumed to be connected only to the eWIPS as to determine and test a suitable control scheme. The high-level control scheme covering the interactions between multiple buses is not covered in this paper. A generator system considered for this study consists of a PMM that can be mechanically coupled to the aircraft engine.
An active front end rectifier (AFE) is connected to the PMM which converts ac power generated by the PMM to dc power and transfers to the bus. The AFE can be controlled via pulse width modulation (PWM) generated from the control scheme signals. A resistor is used to represent the eWIPS that is assumed fully resistive in nature.
The key equations representing the system can be seen from (1) to (4). They are denoted in the widely used rotating reference dq frame for three phase drive systems [17]. vd, vq, id, and iq are the ac voltages and currents in dq frame. Rs, Ld, Lq, and ψm are the PMM stator resistance, inductances, and machine flux respectively. idc and Edc are the dc current and voltage while ωe is the electrical speed.

III. CONTROL SCHEME
Typically, the control priority for an aircraft generator system is the bus voltage for nominal load operation and current limit regulation within the EPS rating. In [12], speed control was selected for starter operation while bus voltage control with droop coefficients was used in generation mode. For both modes, flux weakening control is always present in the event of operation at high speeds to preserve power converter controllability. Similar control scheme was introduced in [15] with the use of droop based bus voltage controller focusing on its design and stability analysis. Classic bus voltage control was used in [18], [19] and [20]. Their control schemes seemed sufficient for the priorities mentioned earlier, however an alternative control scheme has to be considered to introduce the "voltage wild" concept. An additional function to control is the power of the eWIPS load that will vary the bus voltage levels.
As such, there are four different functions that have to be performed by the control scheme: 1) Flux weakening for the PMM during high speed operation. 2) Keeping Edc constant when connected to the normal bus.
3) Regulate eWIPS power when connected to the eWIPS bus. 4) Regulate stator current within the designated limit. The proposed control scheme will facilitate all four functions in order to enable the variable voltage concept. The functionalities allow constant/variable bus voltage control depending on the bus it is connected to. Flux weakening is employed when needed to maintain converter controllability and in addition to that, stator current limit controller. The control structure would then be able to prioritise these functions depending on the operating modes (speed and load) and reference values.
The control structure for this study is shown in Fig. 3. The structure can be divided into two parts; the inner and outer loops. The inner loop consists of the dq current controllers (Wid and Wiq) and a dynamic modulation index limit. mabc, md, and mq are the modulation index in three phase and dq frame respectively. mlim is the maximum modulation index and mdlim is the limit imposed on md. The value of ks depends on the modulation scheme to be used for the power converter [21]. The outer loop is made up of dc voltage, current limit, and dc power controllers (We, Wis, Wp) with back-tracing (BT) algorithm. iq1 * , iq2 * , and iq3 * correspond to the outer loop controllers output respectively. is and ismax are the stator current and its maximum value, while Pdc is the dc power. Variables with the * superscript denotes its corresponding reference signal. This control scheme covers the four functions listed previously with the dynamic limit to maintain converter controllability and the remaining outer control loops for their respective functions.

A. Inner control loops
The operating principal for the dynamic limit is based on the following equation: The modulation index is the ratio between the AC voltages to Edc through: For this study, mlim is set to 1 for full utilisation of the available voltage supplied by the PMM. The q-axis control loop should have higher priority than d-axis, hence md can be limited to fulfil the conditions of (5): Flux weakening is achievable through this modulation index limit (5), where mdlim dictates the output md which determines the amount of reactive current represented by id into the PMM to reduce the back-emf. The current loops are set to control the stator currents in dq frame. Wid has a fixed reference value of zero. This will ensure that id is controlled to be zero when the modulation index is less than its limit. On the other hand, iq * is dictated by the outer loop controllers and determines the value of mq and mdlim. The output of the current controllers are scaled by Edcks to obtain the respective modulation indexes using (6).

B. Outer control loops
The function of We is to maintain constant Edc (typically 270V for high voltage dc buses) in the event that the EPS is connected to a normal bus. It can also be used with the eWIPS bus if only nominal power is required. Wis serves as a current limit to address the disadvantage of this type of control structure. The output of Wis (iq2 * ) is multiplied by the sign of iq * to ensure correct power flow direction. When is exceeds the reference value, iq2 * reduces in order to meet the requirements of: If is is less than ismax, then Wis output is limited to zero to prevent control conflict with the other outer loop controllers. The third outer loop controller is Wp that regulates the power required for the eWIPS load. This controller is setup for uni-directional power flow to only send power to the dc bus. The control idea is straightforward, which is to control the power sent to the eWIPS load. The power demand should be derived from the temperature requirement of the eWIPS load with respect to the surrounding temperature. However, in this paper it is assumed that the power demand is determined by the user in order to simplify the control design process. This controller becomes a priority when the power demand exceeds the power supplied using We (Pdc with given Rw at Edc * ). In general, the magnitude of error of the controllers determine the controller priority.
Each of the controllers can adopt the general PI configuration, utilising proportional and integral gains (kp and ki) to regulate the respective control variables. The controllers also use anti-windup scheme to prevent integrator saturation issues. The outputs of these controllers can be selected as reference values for iq. A minimum function is used to compared all three signals and the smallest value is selected as iq * . The minimum value is used as the direction of current flow from the PMM towards the dc bus is considered to be negative. To ensure smooth transitions between the controller signals, BT algorithm is employed for all three controllers [22]. This algorithm takes into account the output iq * and maintains each controller integrator states to be of similar value. Fig. 4 shows a flow diagram that summarises in general the control action of the proposed control scheme.

IV. CONTROL PLANT
In this Section, the control plants are defined to aid the control design process. It can be observed that there are nonlinearities in the equations of the EPS such as (3), (5), and (6). They are linearised using Taylor's approximation around a particular operating point for small signal analysis. An assumption made is that ωe is constant for small signal analysis as the speed changes are much slower than the electrical variables. The following are the small signal equations to be used to derive the outer control loop plants, and the corresponding small signal variables have the symbol Δ. Any variable with the o subscript denotes its initial operating point value.
A. Inner current control loop The inner current loops design have been well established in literature [23]. One can form the closed loop plant for the current loops with PI type controller. For the iq control loop: The current controllers have been designed to achieve bandwidth of 400 Hz, damping factor ζ = 0.95 and the controller gains can be found in Appendix I.

B. DC voltage control loop
From (3), it can be seen that Edc can be controlled through the variation of iq (id is exclusive for flux weakening function) [21]. Therefore, the plant for this control loop can be developed starting from the linearised equation of (3):  Each small signal variable can be replaced with equations (9) to (15) such that an open loop plant relating single input iq * to single output Edc can be established in (17). The coefficients for n1, n2, n3, d1, d2, and d3 are located in Appendix II.
A comparison step response has been made between the derived transfer function with a non-linear model that consists of the equations (1) to (5). At a same operating point, both responses show good correlation and hence the transfer function can be considered as the plant for We.
The root locus at different loads were also plotted as depicted in Fig. 5. As the load power increases, there is a tendency for one of the poles to move towards the left side. A more important discovery is that there is a pair of conjugate zeroes that move closer to the imaginary axis as the load power increases.
If a pure integrator controller is used to form a closed loop transfer function, the trajectory of the closed loop poles can cross the imaginary axis to the left hand plane which causes instability. This can be seen in Fig. 6, and the controller should be designed carefully as there is a limited gain range that can guarantee stability. However, if a proportional term is added to the controller, then the root locus trajectory is confined within the right hand plane (RHP). This means that the controller is stable for all gain values and is selected for Edc control.

C. DC power control loop
The approach to derive the plant for Pdc control is similar to the Edc plant, and can begin from the linearised equation for Pdc: Using the same set of equations, the plant can be derived as: The derived transfer function was also verified with an equivalent non-linear model. Since the open loop plant is closely related to the Edc plant, it shares similar poles and zeroes location. Using an operating point when Pdc control is operational, the plant showed zeroes that are on the RHP as seen in Fig. 7. This shows non-minimum phase characteristics, and the controller will have a limited gain range that allows stable operation. Since the power fed to the load can be done by increasing Edc, it is possible that m can be less than 1. As a result, the root locus has real positive zeroes instead as observed in Fig. 7. For the purpose of this study, it is assumed that the EPS always operates with m = 1, hence only the associated operating points are considered. For the closed loop analysis, the gain limit for Wp is inversely related to the amount of power requested. The controller would have a small gain range for stable operation if the need for high power is required. Hence, it is recommended that small controller gains are selected for Wp. Fig. 8 shows the closed loop root locus for Pdc plant with PI type controller. The gain range was verified with an equivalent non-linear model where increased oscillations was observed when the controller gain exceeds the range. It can be seen that if proportional gains are used for the controller, its value has  to be small or instability may occur when the pole pairs (pink dots) cross to the RHP. Therefore, it is recommended to have an I controller instead where the controller gain has more stability range.

D. Current limit control loop
Since the stator current is the control variable for this loop, hence the plant can be derived from the linearised current limit equation (8): Similarly, using the equations (9) - (14), the plant for Wis can be found as shown in (21). The coefficients for n4, n5, n6, d4, d5, and d6 are located in Appendix II.  Fig. 9 shows the derived open loop plant root locus at different load points when the EPS is operating at current limit. This plant exhibits non-minimum phase characteristics as well, and the controllers would need to be designed carefully. The variation of load shows small changes to the positive zero locations that gradually moves to the real axis. Furthermore, there is a presence of a plant pole very close to the origin. This would mean that the controller bandwidth has to be slow (less than 10Hz), and the controller gain range can be very small. The use of I controller allows a slightly larger gain range, however at the cost of damping. The closed loop analysis using PI type controllers can be seen in Fig. 10. The addition of proportional term, allows more damping and also increased bandwidth allowance along the stable gain range. Hence, the PI type controller is selected to regulate is. In summary, both Pdc and is plants have non-minimum phase characteristics and their controllers have limited stability gain range. The gain range can be determined through the derived plants at the desired load operating point. The closed loop root locus of the derived plants can be used to select suitable gains for the respective controllers.
Taking into account the gain stability range (before having positive poles on the right half plane), the gain can be selected when the closed loop poles are ¾ distance away from the imaginary axis. This is to ensure stable operation of the controller while having reasonable dynamic response.
The following gains have been selected for the outer loop controllers to achieve reasonable bandwidth response within the stable ranges and they are displayed in Table I. The backtracing gains have been selected as kte = ktp = kti = 200 that is much faster than the controller bandwidth to achieve fast tracking during controller transient changes.  Fig. 9. is open loop root locus at different power demand levels.

V. SIMULATION RESULTS
The "voltage wild" control scheme has been tested in simulation environment. Fig. 11 shows the results demonstrating the proposed control scheme achieving the four functionalities highlighted earlier. The operating points were selected such that each of the outer loop controllers have a distinct working region. They were picked to proof the concept of the control scheme and to not test the performance of the EPS (which will be subject for future publications).
The initial operating condition is simulated at 3100rpm, mlim = 1, Pdc * = 0W and the ismax = 4A. Pdc * is set to be lower than the actual Pdc so that We has priority on the outer loop control, unless the current limit is reached. At t = 0.3s, a step increase of Pdc * = 450W is applied and reduced back at t = 1.3s. During this period, it can be observed in the figures that Wp is selected by the minimum function as iq3 * is larger than iq1 * in the negative direction and Pdc is controlled to 450W. The control action causes Edc to increase in order to satisfy the power demand.
Then, the speed is increased from 3100rpm to 3200rpm at t = 1.7s and is reduced back again at t = 3s. As the speed increases, more id is required for flux weakening in order to maintain m = 1. This causes is to increase as well and eventually reaches the current limit ismax = 4A. The initial overshoot of is is due to the nature of the open loop plant that contain double RHP zeroes. In addition to the slow time constant, the overshoot can be contained but can be difficult to be eliminated. However, some level of current overshoot are allowed in this control scheme and throughout the operation, m has been kept at 1 using the dynamic limiter.
From a general observation, it can be seen that there are different transient dynamics when the outer loop transitioning in and out of different controllers. This is due to the transient between the outer loop controllers. The bandwidths of both Wp and Wis are slower than We due to the gain limits as discussed in Section IV. As a result, transitions into Wp and Wis are slower in comparison to transitions into We. The back-tracing gain within the controller does not have significant influence during the transition due to the slower bandwidths of Wp and Wis.

VI. EXPERIMENTAL RESULTS
An experimental rig has been used to verify the findings in this paper. The PMM (Emerson, 115UMC300CACAA) is connected to a two-level IGBT power switch AFE converter built in-house and the control platform is a DSP (Texas Instrument, TMS320C6713 DSP Starter Kit) and FPGA board. The PMM is driven by a dc brushed machine (TT Electric, LAK 2100-A) using a commercial four-quadrant dc drive (Sprint Electric, PLX 10) that acts as an active load. The power rating of the rig is much smaller in comparison to the high power demand of eWIPS as the tests aim to validate the applicability of the "voltage wild" control scheme. The operating points used are similar to the ones used in the simulation analysis. Fig. 12 shows the results transition from constant Edc control to Pdc control. This can be done with a change of Pdc * as shown at t ≈ 0.9s. Pdc can be seen being controlled to 450W as indicated by Pdc * . Edc increases to satisfy the power demand that can be observed in iq. Meanwhile, id adapts accordingly to maintain m = 1 during operation. The transition between the controllers can be identified when iq * initially following iq1 * changes to iq3 * .
The transition of Pdc back to Edc control can be seen in Fig.  13. Pdc * is reduced to 0W which is less than the power supplied at 350V and iq is reduced. This triggers the constant Edc controller output (iq1 * ) to be the priority once again as it is the smallest among the three reference signals. The different transient response is observed at about t ≈ 0.65s compared to when Pdc is operational (from Fig. 12). This is due to iq * being dictated by the Edc controller output once Pdc * was changed. Some steady state error are present when the controller is in Pdc control. This can be explained due to the amount of power consumed by the resistor. The resistance value can change with respect to operating temperature which affects the amount of power drawn. This error is handled by the integrator term of the controller which ensures that the correct amount of power is supplied to the load. Fig. 14 and Fig. 15 show the results for when the outer loop control transitions between Edc and current limit controllers. The operating condition was initially at 3100rpm which allows for is to be very close to is * = 4A. The speed is then increased to 3200rpm at t ≈ 1s in order to increase is demand. is is perceived to be limited to is * and there is a slight increase in Edc to compensate for the limited current. The transition is highlighted when iq * changes between iq3 * (Pdc control) and iq2 * (current limit control). The different transient dynamics between the outer loop controllers can also be observed, similar to the simulation results in Fig. 11. Fig. 11. Simulation results with demonstrated Pdc control (0.3s < t < 1.3s), is control (1.7s < t < 3s), and Edc control.
It is to be noted that throughout the control operation, m has been kept within the limit of 1 via flux weakening. In general, the experimental results demonstrate the functionalities of the proposed control scheme.

VII. CONCLUSION
The new 'voltage-wild' concept has been investigated for an aircraft EPS with eWIPS load. Its novel control scheme was proposed with the strategy to cover the concepts four main functionalities; Edc, m, Pdc, and current limit control. The small signal control plants for the controllers have been derived for controller design process. Non-minimum phase characteristics was inherent for both Pdc and current limit plants, which meant limited gain stability for the associated controllers. Simulations have been performed in order to demonstrate the control operation for each of the functionalities. Experimental testing has also been done to proof the variable voltage concept on a mock-up EPS. For future studies, the EPS shall be expanded to cover multiple buses to emulate a larger aircraft EPS. This allows