SMART: Modular Architecture for Reliable Transportation

This paper proposes a new ultra-reliable architecture for fully electric propulsion primarily aimed at the automotive sector. The system considers a direct drive arrangement, minimizing the number of moving components and is based on a network of multi-three-phase machines providing propulsion. The DC-link from each 3-phase star within each machine can be interconnected to form a network, or mesh, which would allow a variety of post-fault mitigation strategies. The system scaling is considered along with the benefits and impact of modularity on device ratings. Finally the simulated performance for one drive is presented demonstrating the core capability in redirecting power flow in case of a fault.


I. INTRODUCTION
The electrification of transportation is at the forefront of both industrial and academic development at a time when international governments have set ambitious targets for the reduction of greenhouse gases. The transport industry is attributed as one of the major contributors of carbon emissions and as such, the UK government will ban all new petrol and diesel cars as of 2040 [1]. Similar plans are afoot in France and the rest of Europe, as a commitment to clean air policy. Fully electric propulsion will be widely adopted within the next few decades, eventually making traditional combustion engines obsolete.
While important strides have been taken in improving electric vehicle technology, the disjointed nature of the automotive sector and the early stage of adoption mean that there is a lack of industry standardisation of voltage levels, system architectures and alignment of development goals.
With the emergence of technologies such as autonomous fully electric vehicles, safety and reliability is expected to become a major theme. In addition, current methods for powering pure electric vehicles suffer from a lack of reliability -a single point of failure can bring the whole vehicle to a standstill, compromising the safety of the passengers and other road users.
The aim of this paper is to propose a system architecture combining both reliability and fault tolerance in order to deliver a system focused on safety. The system also takes advantage of modularity and flexibility in the aspects of power electronics, control and machine design.
The first section of this paper will cover the literature around multi-three-phase motor research and their control which is the enabling technology for this system. The second section will introduce electric vehicle drivetrains establishing what is considered the conventional arrangement, and it will then introduce the proposed arrangement. The third section investigates how the system scaling is affected by modularity. Its impact on different aspects is quantified either analytically or via simulations. Finally the simulated performance is presented. The basic building block of the system is identified and its functionality is shown under different conditions including simulations within the full system

A. Multi-three phase machines
Multi-three-phase machines are a newly-developed enabling technology that allows for a flexible response to changing demand within the electrical machine. The machine windings are typically split into numerous sets of 3-phase systems physically displaced by a fixed angle. Typically these machines have a higher complexity in modelling and control as the interaction and cross coupling between phases must be considered, leading to a complicated inductance matrix [2][3][4][5][6][7].
Recent technological advancement in power flow control has shown that the load sharing can be controlled between the numbers of systems [2,[4][5]. Control can be either centralised or even decentralised [3] for higher reliability. Solutions typically use vector space decomposition for control in the synchronous reference frame [2][3]5]. Efforts have already been made to integrate multi-three-phase machines for directdrive and integrated in-wheel applications [8][9] where it has been shown that high torque density can be achieved [10][11]. The proposed system considers multi-three-phase machines as the enabling technology. The increase in computational power of modern digital controllers and the emergence of more efficient power semiconductors enhances the capability of this technology to achieve high efficiency and reliability compared to traditional powertrain solutions as will be outlined in the next sections. The layout of conventional drivetrain of an electric vehicle is shown in Fig. 1. A gearbox is used to increase the mechanical output torque transferred to the wheels via a differential transmission. High power density is an important factor in transportation and to achieve it in this configuration, high speed machines are opted for, since the increase of the fundamental frequency is associated with a decrease in size. Typical maximum speeds approach 20000 rpm where a gearbox ratio of around 9:1 is used to reduce the speed of the wheel to around 2200 rpm, depending on the radius of the intended wheels and the maximum speed. All-Wheel-Drive (AWD) capability, typically found in high end models, is implemented by using separate machines to drive the front and back wheels. These could be scaled equally giving a 50:50 torque capability between the front and back wheels or following an asymmetric arrangement, where a main motor is rated close to the maximum power whilst the second motor behaves as an auxiliary one to provide better traction and additional power. High end commercial vehicles are rated between 300 kW to 400 kW with around 1000 Nm of torque.

A. Conventional electric vehicle drivetrain
In the case of an electrical fault occurring in the motor, a two-wheel drive (2WD) system could be fully compromised whilst an AWD will benefit from the redundancy and could continue operating, albeit at a significantly reduced power level. When further faults are considered, the vulnerability of the system is revealed. A fault in the battery bank could bring the system to standstill if the fault is not isolated. Mechanical failures are also likely to occur due to the numerous components within the gearbox and transmission.
There are a few notable and well publicized failures of electric vehicles where the severity of the fault has been exacerbated by the flammable and toxic nature of lithiumbased batteries [12]. This emphasizes the importance of faulttolerance which must be designed into the system. The use of high current-rated converters (up to 1.5 kA) in current systems makes it more challenging to isolate a fault, particularly if a paralleled switch arrangement is opted for. Finally, more moving parts make mechanical damage more likely.

B. SMART vehicle drivetrain
The proposed drivetrain is designed bearing in mind reliability and fault tolerance from the onset. Rather than using a single machine and drive for propulsion, the proposed system envisions a mesh of identical nodes interconnected with multi-three-phase machines and multiple DC sources.
A simplified configuration is depicted in Fig. 2 considering only two wheels. Each wheel is mechanically driven directly (no gearbox) by a 9-phase electrical machine, which can be considered as 3 separate 3-phase systems. Each of the three stars contributes equally to the mechanical torque thus the rating of the converters is scaled by the number of systems, three in this case. The dc-link of each converter from the same machine is connected to a separate dc source. By interconnecting the dc sources from the two machines, three separate power paths are created. The segmentation is therefore carried on from the machine to the drives as well as the batteries.
Each machine can be defined as a power node and a "roundabout" of energy exchange, meaning that power can be exchanged among different stars. Thus when a failure occurs at any point, energy can be rerouted and still flow to functional nodes. The interconnected mesh allows for redundancy to be designed into the system from the start, ensuring that a single component failure will not cause a whole system failureguaranteeing system reliability.
Each node is locally self-controlled and can support neighbouring nodes in case of faults. A high level of reliability is achieved by combining the redundancy of the modular converters and multiple stator windings of the motors. When considering a full vehicle of four machines, multiple options are possible as shown in Fig. 3. Each option considers different paths and interconnections as well as different segmentation of the batteries, shared by two or more inverters. The decision about which system is optimal will depend on a given design and should be investigated among the multitude of possible options.
The following are the overall benefits of the proposed system: From a mechanical perspective, AWD is assumed therefore the total vehicle torque is divided by the number of wheels, four in this case, where each can provide 800Nm. The motor mechanical speed is given by the maximum vehicle velocity and the wheel diameter. Considering 18 inch (46cm) wheels and 200 km/h this is set at 240 rad/s. Segmentation and subsequently modularity is a central theme to the proposed concept. In order to assess expected benefits and implementation feasibility, and to quantify the limitations of a possible solution, multiple aspects must be considered which may be interlinked.
The impact of machine segmentation on the current ratings is straightforward. For a given machine constant the current needed to deliver the full torque is divided by the number of three phase systems N.
The machine constant is proportional to the number of poles and the machine flux. Increasing the number of pole pairs will increase the output torque. However, this will also increase the fundamental electrical frequency which relates the mechanical speed to the number of poles. Considering the inverter switching frequency must be at least 20 times higher than the fundamental frequency. This means that the number of poles will be restricted by the switching frequency as well as by physical space constraints.
The problem is visualised in Fig. 4 and Fig. 5 where the number of pole pairs and the switching frequency are shown against the nominal current and the number of three phase systems in the machine. An overall trend can be observed showing that as the number of systems increases the number of pole pairs decreases for higher nominal currents. If the nominal current is decreased too much then the number of poles required becomes unfeasible as well as the switching frequency. Furthermore, the graphs show that a wide range of solutions is possible for a given total torque for which more assessment may be necessary to identify optimal solutions.
In order to assess the impact of higher switching frequency at different nominal currents for power electronics a simulation has been used.
The thermal models for different switching devices have been included into PLECS. Devices rated at various current levels have been considered along with different technologies and packaging solutions, ranging from discrete devices to power modules and silicon based IGBT/Mosfet to Silicon Carbide (SiC) devices from a range of leading manufacturers An automated simulation has been run to evaluate the semiconductor losses at different load levels and switching frequencies for each device. A 3-phase two-level Voltage Source Inverter (VSI) topology has been used with the load adjusted at different percentages compared to the device nominal current. The efficiency accounts for the full converter. Two voltage categories have been considered with the devices compared separately for a 400V or 800V DC level. Selected cumulated results can be seen in Fig. 6 where the efficiency of the full converter is shown against the switching frequency and nominal current. Each colour represents a different device (SiC in squares, Si in asterisks). The following conclusions can be reached. Devices rated with higher currents show a higher drop of efficiency with increasing switching frequency and significantly poorer efficiency at low modulation indexes (low loads). SiC devices (squares) offer significantly superior performance, especially at higher switching frequencies compared to Si devices (asterisks). The above supports the hypothesis that segmentation enables the use of a bigger number of lower rated devices to benefit the overall efficiency of the drivetrain, and particularly at low loads.
Analytical computations have been carried out in order to find the variation of the DC-link capacitance size considering different levels of segmentation, accordingly to the methodology from [13] for sinusoidal PWM. The results are depicted in Fig. 7 for commercially available C4AE capacitor series and two total load currents assuming 5% of the DC-link voltage ripple, unity load power factor and unity amplitude modulation index. It can be observed that the capacitance volume increases for a higher number of systems, mainly attributed to the fact that higher packaging and insulation is needed for this scenario i.e. N capacitors providing the required capacitance are bigger than one big capacitor of the same value. However as the switching frequency increases, the DC-link capacitance can be minimised for a permitted level of DC voltage ripple.
A preliminary estimation has been carried out in order to capture the trend of the variation of volume and mass of the powertrain for different levels of segmentation. The plots, shown in Fig. 8 account for switching devices, batteries, inverter cooling system and DC-link capacitors. The computations are based on datasheets of commercially available devices and components. 26 Si IGBT and SiC MOSFET devices were characterised in PLECS software. Then 23,929 powertrain configurations were generated of equivalent power and capacity. Proportional approximations were used for component masses and costs estimations. Optimal mass and volume configurations were found for each N and device technology combination and plotted as a function of N. The overall results suggest a balance can be reached between the performance, size, weight and presumably the cost of the drivetrain solution.

IV. PERFORMANCE EVALUATION
In order to validate the feasibility of the proposed idea two sets of simulations are presented. Both simulations consider a system based on a 9-phase motor scaled according to the full test vehicle specifications.

A. Power Electronics simulation results
A PLECS simulation has been built in order to validate the basic functionality of a multi-three-phase machine as a multifunctional node, as required in the proposed application. TABLE 2 shows the simulated parameters of a multi-threephase electric motor coupled to a prime mover spinning the shaft of the machine under test at 240 rad/s, the maximum speed which would bring the vehicle at 198 km/h. The implemented algorithm is the Field Oriented Control for threephase Surface Permanent Magnet machines. No flux weakening has been considered. In order to control the nine phase currents, six Proportional-Integral controllers -three along the d axes and three along the q axes -have been designed accounting for the values of the first harmonic inductances. In the simulation, three isolated 2-Level Voltage Source Inverters (2LVSI) have been connected to three isolated 800V Battery Packs (BPs) based on a non-linear battery model [14]. Three operating conditions have been considered and presented in the following paragraphs. Figure 9 shows the AC currents in steady state motoring operation. The three transformed respective currents within each star, shown in Fig. 10, are producing an overall wheel torque of 806 Nm (Fig.  11) with three DC-links showing the battery voltages at 802 V (Fig.12). The motor performance achieved is in line with the designed operating point thus validating the model and confirming the functionality of the control scheme.
 Steady-state motoring operation:   In case of open circuit fault in one of the inverters, only six currents out of nine are regulated (Fig.13) and two thirds of the nominal torque are delivered to the shaft (Fig. 14). As in a real case scenario where fuses would be installed along the lines, the short circuit fault has not been considered. In case of short circuit, the phase currents would rapidly increase while blowing the fuses and isolating the faulty segment of the system [10].  Transient operation showing regeneration mode:    Fig. 15 and Fig. 16 along with the impact on the torque, which is reduced to one third (Fig.  17). The operation is validated in Fig. 18 showing the BP State of Charge (SOC) reflecting the expected change in power flow. The result thus concludes the validation of the basic functionality of a multi-three phase machine as a multifunctional node.

B. Full system simulation results
A second simulation has been set up considering an AWD system based on 9-phase machines and the vehicle mechanical model the results of which are presented here focusing on the longitudinal behaviour. The model considers the aerodynamic drag and the rolling resistance of the vehicle.  To achieve a preliminary indication of the full vehicle performance the model has been run over the Worldwide Harmonised Light Vehicle Test Procedure (WLTP) drive cycle over a period of 30minutes (1800s). The results are depicted in Fig. 20 showing the vehicle torque and velocity. It can be observed that the achieved profile follows the demand with negligible error thus validating the effectiveness of the design and suitability as an electric vehicle.

V. CONCLUSION
This paper has proposed a new reliable architecture for automotive propulsion based on multi-three-phase motors. Each motor can be considered as a node connected to identical converters which are interconnected in a mesh to provide multiple and redundant power paths. The scaling and segmentation for such a system has been outlined along with design constraints for its implementation.
Simulation results have validated the fundamental operation and functionality of one node in normal operation and under fault. Finally a full vehicle simulation has been presented over a WLTP drive cycle showing the system performance matches the design goals.
This research concludes the preliminary investigation into the feasibility of this system and opens the scenario for more thorough and in-depth investigations into this architecture. The option of having numerous system layouts has been identified and thus advanced optimisation is needed at many levels in order to identify ideas aiming towards implementation. The system will guarantee that in the scenario of one or even multiple faults occurring, the vehicle can continue its operation ensuring the trustworthiness of a fully electric propulsion future.