Considerations on the Development of an Electric Drive for a Secondary Flight Control Electromechanical Actuator

The more electric aircraft concept aims to improve the fuel consumption, the weight, and both the maintenance and operating costs of the aircraft, by promoting the use of electric power in actuation systems. According to this scenario, electromechanical actuators for flight control systems represent an important technology in next generation aircraft. This paper presents a linear-geared electromechanical actuator for secondary flight control systems, where the safety and availability requirements are fulfilled by replicating the electric drive acting on the drivetrain. Indeed, the architecture considered consists of two power converters feeding as many electrical machines coupled to the same mechanical system. The design of both the permanent magnet synchronous machine and the power converter are addressed. Preliminary results on the electric drive prototype are also provided and compared to the design requirements. Finally, the electromechanical actuator performance at system-level is evaluated in Dymola environment, analyzing different operating modes.


I. INTRODUCTION
In recent years, demanding regulations on fuel emissions and the need of decreasing both operating and maintenance costs have encouraged the aircraft industry to move from conventional hydraulic to electrical actuation systems [1][2][3][4]. This trend is supported by the inherent benefits offered by electrical actuation systems, such as a) weight reduction due to the removal of hydraulic pipelines, b) high dynamic performance ensured by both low inertia motors and high-bandwidth controllers and c) actuation distribution degree of freedom [5,6].
Among the electrical actuation solutions, mainly two configurations can be identified, namely electro-hydrostatic actuators (EHAs) [7] and electromechanical actuators (EMAs) [8,9]. In the former, electric power is used for supplying the electric drive, which is mechanically coupled to a hydraulic pump. Thus, hydraulic power is locally generated for driving the flight control surface. On the other hand, a mechanical drivetrain (e.g. gearbox and/or ball screw) is placed between the electric drive and the flight control surface [10], as schematically shown in Fig. 1.
Both EHAs and EMAs are currently employed on large commercial aircraft. In particular, the Boeing 787 uses EMAs for landing gear braking and spoiler surfaces [11,12]. Similarly, the Airbus A380 adopts EHAs for ailerons and elevators (primary flight controls), while slats (secondary flight controls) are actuated by EMAs [13,14]. The adoption of EHAs for the primary flight controls led to a 1500 kg weight reduction on the A380 [15]. Despite the local hydraulics and the higher maintenance costs [3], currently the use of EHAs is still more widely spread than EMAs, due to the well-established "know-how" regarding hydraulic systems [15,16]. In fact, the risk of failures (e.g. jamming) and the lack of accumulated reliability experience [17] relegate EMAs to be currently implemented on secondary flight controls, such as flaps, spoilers and slats [11]. These controls are considered less safety-critical, due to the inherent surface redundancy of the aircraft wing, which is equipped with multiple surfaces performing the same aerodynamic task [18]. Nevertheless, EMAs represent an attractive technology, which is gaining interest due to its higher efficiency and further weight reduction compared to EHAs [1,3,19]. A linear geared EMA for secondary flight control systems is considered in this work. Despite the existing surface redundancy of the aircraft wing, the EMA availability has been increased by duplicating the electric drive. Hence, two electrical machines powered by two isolated power electronics converters (PECs) and controlled by two Micro Controller Units (MCUs) are used for driving the EMA drivetrain. This architecture allows to achieve the safety requirements in case of a fault occurring in the electric drive components (i.e. MCUs, PECs and motors), which have a higher failure rate [8,20]. The EMA failure probability is determined based on the fault tree methodology [21,22] and the value obtained for the described EMA electric drive configuration is compared against the ones regarding the existing technologies (i.e. EHA and purely hydraulic actuation system), in Section II.
In order to meet the challenging volume requirement, a permanent magnet synchronous machine (PMSM) is selected as the machine topology due to its high-power density and excellent efficiency [23,24]. The PMSM design is discussed along with some manufacturing details and finite element (FE) analysis results. Details regarding the PEC design choices are given throughout the paper and its thermal management is evaluated by means of experimental tests. The electric drive performance is experimentally assessed and compared against the design outcomes. Since the EMA drivetrain (i.e. gearbox plus ball screw) is not yet available for fully integrated tests, the performance at system-level is simulated in Dymola environment, using purpose built models for the electric drive components [25,26]. In particular, two EMA operating modes are investigated: 1) the active-active mode, where both electric drives are operated at the same time; and 2) the active-standby mode, where only one electric drive is powered (see Fig. 2).

II. ELECTRIC DRIVE DESIGN AND CONSIDERATIONS
In safety-critical applications, redundancy is often necessary for increasing the system availability. Despite the ever-growing importance gained by the electrical drives in modern aircraft, their components experience a higher failure rate compared to the mechanical drivetrain components [8]. Thus, a flexible "dual-branch" electric drive (i.e. redundant electric drive), acting on the same mechanical system, has been adopted for the considered EMA. The chosen fault-tolerant architecture is depicted in Fig. 2 and its failure probability is evaluated in the sub-section II.A. The two electric drives highlighted in Fig. 2 consist of an aluminium encased electrical drive and one PMSM each. The box of each drive encloses one MCU and one two-level custom inverter (i.e. PEC) fed by a 270 V DC bus. The linear geared EMA drivetrain is characterized by two stages: 1) the gearbox, which enhances the torque produced by both PMSMs and 2) the ball screw, which converts the rotational motion of the gearbox output shaft in linear motion of the EMA rod-end. The gearbox features a 200:1 step-down ratio, while the ball screw's transmission ratio is equal to 500 rad/m. The next sub-subsections deal with the design and testing of both the PMSM and PEC, along with the EMA failure probability.
A. Failure probability of the EMA On a civil airliner, the flight control systems actuated by a purely hydraulic system feature a failure probability of less than 10 -9 per flight hour [21], which represents the target value for the replacing technologies (i.e. EHA and EMA). Using the components' failure rates reported in [21,22], the fault tree for both EHA and EMA in "single-branch" configuration can be built. A "single-branch" EHA is characterized by a failure probability of 1.98x10 -4 per flight hour, whereas 1.61x10 -4 per flight hour is the failure probability of a "single-branch" EMA [21,27]. The obtained values reveal that the "single-branch" configurations cannot guarantee the failure probability demanded by the flight control system. For this reason, "dual-branch" solutions (i.e. redundant configurations) need to be employed to reduce the failure probability, despite the increment in weight and cost. The fault tree for the EMA architecture of Fig. 2 is shown in Fig. 3, where it can be observed that total failure probability is 8.12x10 -6 per flight hour. From this diagram, it can be noted that the EMA's failure probability is constrained by the mechanical system, since its failure probability (i.e. 8.1x10 -6 ) results higher than the one of the "dual-branch" electric drive (i.e. 2.36x10 -8 ). While the EMA's failure probability does not meet the required value, it is still acceptable for the application at hand due to the inherent redundancy of the aircraft wing. Indeed, the presented EMA will only act on a single flight control surface.

B. PMSM Prototype: Design and Testing
For the application under study, each PMSM is required to deliver 5 Nm at 8000 rpm and operate at a maximum ambient temperature of 70°C using natural air cooling system. Furthermore, to comply with the fast-dynamic response requirements, the moment of inertia should not exceed 10-4 kg·m 2 . According to these specifications and taking into account the limited available space (i.e. axial length ≤ 80 mm and outer diameter ≤ 80 mm), the design has been performed via FE analysis. A preliminary trade-off study has been carried out for selecting the appropriate slots/poles combination. In order to minimize both the torque ripple and the cogging torque, which might result in undesired effects on the EMA drivetrain, 9 slots / 8 poles machine configuration has been chosen [28]. Since distinguished by high fill-factor and short end-windings, a double-layer concentrated winding has been adopted, as shown in Fig. 4. This winding configuration contributes to lower copper losses [29], whereby the heat dissipation is facilitated. High permeability cobalt-iron alloy laminations have been used for the stator core, because they feature relatively high saturation flux density (i.e. 2.3 T) and lower specific iron losses if compared to silicon steel alloys. In terms of permanent magnets (PMs) material, samarium-cobalt has been preferred to neodymium-iron-boron, due to the higher operating temperature capability. To increase the torque and reduce the back-electromotive force (back-EMF) harmonic content, the PMs on the rotor surface have been mounted in a quasi-Halbach array layout [30]. Despite higher cost and manufacturing complexity, the adoption of the quasi-Halbach array configuration leads to shorter axial length, which is beneficial for both the total PMSM mass and rotor inertia reduction. The PMs' eddy currents losses have been reduced by axially segmenting the PMs in 10 sectors, thus shortening the eddy currents paths through the PMs. This approach decreases the PMs' eddy currents losses by 43% at rated speed compared to when whole PMs (i.e. single segment) are used, hence it leads to a PMSM efficiency improvement. Since the PMs are mechanically brittle, a titanium retaining sleeve has been used for holding the PMs on the shaft and protecting them during the PMSM assembling stage. For securing fast acceleration and containing the rotor inertia, a hollow shaft has been manufactured [31]. A magnetically permeable material (i.e. 17-4 PH stainless steel) has been selected for the shaft, in order to serve also as rotor back-iron. Finally, an aluminum housing equipped with radial fins has been designed for enhancing the heat dissipation by natural convection. In Fig. 5, the assembled PMSM prototype is depicted, whereas Table I summarises its parameters. The designed PMSM is characterized by an overall weight of 2.52 kg and its losses at rated operating condition are provided in Table II.   Fig. 6. A comparison between experimental and FE back-EMFs at rated speed is proposed in Fig. 7, which reveals the back-EMFs low harmonic content resulting from the implementation of the quasi-Halbach array configuration. In Fig. 7, the slight mismatch in amplitude is observable between experimental and FE waveforms. This is due to the end-effect caused by the permeable rotor endcaps, which leads to an increase of leakage flux [32]. The end-effect is not considered in 2D FE simulations, resulting in a higher back-EMF amplitude. Based on the FE results, the voltage constant is equal 11.5 mVrms/rpm, whilst 10.8 mVrms/rpm are obtained from the measurements (i.e. 6 % difference).    In order to develop the rated torque, the PMSM must be supplied with 37.5 Apk (i.e. active-standby operating mode). The flux density map and the flux lines distribution at full load condition are shown in Fig. 8, from where can be seen that the saturation limit (i.e. 2.3 T) is reached only in the tooth-shoe area. In Fig. 9, the FE instantaneous torque is reported, presenting an average value of 5.037 Nm at rated current, whereas its torque ripple is equal to 0.22 %. The PMSM torque-current characteristic has been computed by FE simulations and further experimentally validated. The comparison is shown in Fig. 10, where a good agreement between measurements and FE is highlighted. According to the FE findings, the PMSM torque constant is equal to 0.134 Nm/A, while a torque constant of 0.129 Nm/A is calculated based on experimental data (i.e. 3.7 % mismatch). Finally, a thermal analysis has been carried out on the PMSM [33,34]. Considering the rated operating condition (i.e. 5 Nm at 8000 rpm) and an ambient temperature of 70°C, the steady-state temperatures of the main PMSM components are as given in Table III.

C. MCU and PEC Prototypes: Design and Testing
The MCU and PEC designs have considered various constraints and design specifications. The total volume, weight, maximum and minimum operating temperatures, and the cooling requirements are found to be the most challenging to achieve. Furthermore, the galvanic isolation between MCU, PEC, and the auxiliary power supplies had to be taken into account. Isolation between the MCU and the PEC is provided by the use of fibre optics and current sensors with galvanic isolation between the primary circuit (PEC) and the secondary side (MCU). Similar to the machine design, air cooling has been considered albeit with forced cooling. The design has been completed aiming to minimize the impact on the overall weight. Four fans together with a Fabfin® high performance heatsink have been chosen for cooling the PEC unit.  Only Silicon semiconductor technology has been considered over Wide Bandgap Devices (WBG) as its reliability and failure rates have been thoroughly investigated. An Infineon three-phase IGBT power module (FS100R07N3E4) with a two-level voltage source inverter and an embedded temperature sensor (NTC-thermistor) has been identified as a good solution for the power stage, both in terms of current ( ) and voltage ( ) ratings as well as minimization of semiconductor losses for the required switching frequency. All auxiliary systems are powered from a 28V DC bus. Considering an insulation of 3000 between the auxiliary voltage (28V) and the DC bus (270V), the isolated DC-DC converters from Traco power (TMR3-2413HI) have been selected for suppling the gate drivers. For driving the IGBTs, the ACPL-339J Integrated Circuits (ICs), which provide for 5000 voltage insulation, an extended operating temperature range (-40°C to 105°C), and desaturation protection with soft turn-off, have been selected. Due to the high current required by the gate driver circuit, the ACPL-339J has been cascaded by a mosfet based current buffer. The MCU in Fig. 11 has been designed starting from the uCube [35] design, which is based on the Zynq-7000 System on Chip (SoC) from Xilinx. Phase currents and DC link voltage measurements can be sampled and filtered at 50 MHz using the Programmable Logic (PL) present within the SoC. Provided that the maximum fundamental electrical frequency ) is Hz, the switching and the interrupt frequency ) have been selected at 16 kHz. In order to enclose both the MCU and the PEC shown in Fig. 12, within the aluminum box in Fig. 13, the DC-link capacitor has been sized with the formula given in (1) to ensure minimal DC voltage ripple, where all the parameters are defined in Table IV.  Using (1), a capacitance of 643 μF is calculated. As the volume constraints have been met, a capacitance of 960 μF has been installed, paralleling two off-the-shelf capacitors from EPCOS (B32778J4487K000). The overall weight of the designed PEC is equal to 7.61 kg.

  
The heat paths are highlighted in Figs. 12 and 13. In Fig. 14, temperatures profiles of the NTC, PEC heatsink, air in, and air out, while controlling a current of 44 ARMS on a dummy load, are shown respectively. Also assuming the worst-case ambient temperature (i.e. 70°C), the IGBT power module works well within its temperature range, considering a maximum operating temperature equal to 150°C. The results presented in Fig. 14 prove the effectiveness of the implemented cooling system.

III. ELECTRIC DRIVE DESIGN AND CONSIDERATIONS
Following the design of the PMSM and the PEC, in order to assess the electric drive performance, a suite of integrated system tests have been conducted, where the PMSM is controlled in torque mode. Firstly, dq-current Proportional Integral (PI) controllers have been designed on the phase resistance and on the transformed inductances and reported in Table I. In Fig. 15, the simplified current loop (modelling both actuation and current measurement delay), used for designing the PI dq-current controllers, is shown.  Since the PMSM presents no-saliency, the same proportional and integral parameters have been adopted in both the regulators. For simplicity, axis de-coupling has been omitted. In Fig. 15, where represents either d or q, the actuation delay is taken into account by the pure delay , where is the interrupt period, whilst, the measurement delay is shaped as a second order low-pass filter with cut-off pulsation . In Table V, the parameters used for computing both the proportional and the integral gains, by imposing bandwidth and phase margin , are summarized [36,37]. In Fig. 16, the measured output current providing a q-current reference step of 2 A with locked rotor is compared against the output current from control diagram in Fig. 15. Sampled measurements and derived variables are saved and downloaded on the computer using a custom embedded scope coded within the MCU as shown in Fig. 11(b). Looking at the comparison, the simplified control diagram in Fig. 15 is accurate enough for designing the current PI regulators. The drive prototype introduced in the Section II is validated by imposing a load with the nominal torque at nominal speed. This is equivalent to the overall power demanded by the gearbox during active-standby operation when one drive is not supplied (5 Nm, equal to 37.5 Apk at 8000 rpm). On the other hand, in active-active operating mode, the power demanded to each drive is halved: 2.5 Nm, corresponding to 18.75 Apk, at 8000 rpm. Both active-active and active-standby operating modes have been validated by controlling the torque while spinning the machine under test at 8000 rpm by utilizing an industrial drive connected to an off-the-shelf high-speed interior permanent magnet machine, shown in Fig. 17. In  Fig. 18, the measured mechanical angle when spinning the PMSM at 8000 rpm is reported, along with the electrical angle . In Figs. 19 and 20, the measured phase currents acquired by the embedded scope in both active-active and active-standby operating modes are presented. As the sampling frequency matches the switching frequency, the switching ripple is not accounted for, which can give the impression of a small asymmetry when operating at low loads as seen in Fig 19. This has been validated as no other indications of asymmetry have been observed mechanically or electrically.    The PEC and the MCU enclosed in the box shown in Fig. 13 are able to control the PMSM current in both operating modes. Tests controlling the speed and the position will be performed pending on the availability of the mechanical drivetrain. In Fig. 21, the DC-Link voltage, during the active-standby operating mode, is shown. The measured maximum peak to peak DC-Link voltage ripple is equal to 1.3 V (i.e. 0.48%), which is smaller than the peak to peak DC-link voltage ripple shown in Table IV and used in (1) for sizing the DC-Link capacitance. Considering the mission profile shown in the next section, the speed should be ramped from 0 rad/s to 837.75 rad/s (or 8000 rpm) in 0.429 s. This is equal to stepping from 0 rad/s to 100 rad/s in 51.25 ms. In Fig. 22, the measured rotor speed at no-load while providing a step input reference of 100 rad/s is depicted. Looking at the speed response and at the relative phase currents in Fig. 23, where the maximum absolute current is equal to 12 A, the drive shows the ability of reaching the full speed at no-load, starting from standstill.

IV. EMA MISSION PROFILE
This final section of the paper dedicated to the full system performance evaluation. The fault-tolerant EMA architecture allows to perform the mission profile in two different operating modes: 1) active-active and 2) active-standby [38]. In the former operating mode, both PMSMs equally contribute to the EMA mission profile and each of them is supplied with half of the rated current (i.e. 18.75 A), generating 2.5 Nm. This operating mode is desirable, due to the lower current required for counteracting the load force. In fact, lower current results in lower copper losses, hence easier thermal management [39,40]. In active-standby operating mode, the EMA mission profile is solely performed by one PMSM, whereas the other is kept as a backup or it is intentionally de-energized in response to an electric fault detection [10]. In this condition, the active PMSM develops the rated torque by absorbing the rated current (i.e. 37.5 A). Hence, the PMSM copper losses are doubled compared to the active-active operating mode, thus, more heat needs to be dissipated. Independently from the operating mode, the EMA mission profile consists in a full extension (60 mm stroke length covered in less than 7.5 s), followed by 10 s position holding and finally the rod-end travels back in the retracted position, whilst the load force is continuously applied. The rod-end position is tracked by a cascaded control structure. In particular, the inner loops control the dq current components of both PMSMs, while the outer loop regulates the rod-end speed. The outermost loop supervises the rod-end position. The model of the linear EMA has been implemented in Dymola environment using the Modelica language, as depicted in Fig. 24. Modelica is an object-oriented language, based on algebraic and ordinary differential equations [41]. The object-oriented approach is useful, since partial model and code can be reused. Additionally, the modelling environment provided by Dymola is ideal for multi-domain dynamic simulations, which are required for studying multidisciplinary systems, such as EMAs. Custom-made models have been used for the electric drive components [25,26], whereas standard Dymola blocks have been adopted for the mechanical components.
The EMA is simulated in Dymola environment with the aim of qualitatively evaluating the performance at system-level. This choice was dictated by the unavailability of the mechanical drivetrain at the time of writing the paper. The simulation results for the active-active operating mode are summarized in Fig. 25, where the mechanical quantities at the EMA end-rod are shown. The EMA exhibits suitable dynamic performance, although a modest position tracking error is present during both acceleration and deceleration periods (i.e. absolute tracking error of 0.83 mm corresponding to a percentage error equal to 0.63%). Similarly, active-standby simulation results configuration are reported in Fig. 26. Considering the mechanical quantities at the EMA end-rod, a satisfactory position tracking error is achieved along the whole trajectory; despite the saturation of the speed controller output, during the extension acceleration. Indeed, a moderate mismatch is observable between reference and actual speed signals at the end of the acceleration phase. The saturation of the speed controller output (occurring only during the extension acceleration) leads to a higher position tracking error compared to the previous operating mode (i.e. absolute tracking error of 1.25 mm corresponding to a percentage error equal to 2.08%). V. CONCLUSIONS In this paper, the design and testing of an electric drive for aerospace EMA have been presented. The EMA was designed for secondary flight control systems and its availability was improved by adopting a redundant electric drive, which avoids single point of failure. Indeed, this solution allows to normally operate the EMA in active-active mode and move to active-standby mode, when an electric fault is detected on one electric drive. The PMSM performance has been evaluated via FE analysis and further compared against the experimental back-EMF waveforms and torque-current characteristic. The main outcome of the comparison is the influence of the permeable rotor endcaps, which leads to 6% and 3.7% reduction on the induced back-EMF and on the developed torque, respectively. Nevertheless, the PMSM performance is still in line with the design specifications. The design of the PEC was discussed and details regarding the MCU were provided. The PEC thermal management was experimentally tested and the obtained results proved its ability of safely operating also in the worst-case ambient temperature condition (i.e. 70°C). Electric drive integration tests were performed and the obtained results were outlined and discussed. In particular, both current and speed step responses of the electric drive were assessed and the fine-tuning of the controllers was accomplished. By mechanically coupling the designed electric drive to a commercial one, the feasibility of working in active-active and active-standby modes has been demonstrated. Further experimental tests are scheduled on the EMA once its drivetrain will be available for the next step of the integration process. Finally, The EMA behavior was simulated in Dymola environment, considering the active-active and active-standby operating modes. Dymola results highlighted the EMA attitude of performing the required mission profile in all the simulated conditions.