Electrical Drives in Automotive Systems
An electrical drive can be defined in terms of ability to efficiently convert energy from an electrical power source to a mechanical load [DeD2011]. It is adaptable to almost any operating conditions and can be designed to operate in all four quadrants of the torque-speed-plane.
Transportation sector is a high mechanical energy consumer. The transformation of primary energy (fossil) produces pollution and according to the International Energy Agency [Iae2013], CO2 emissions have increased by 1.4% in 2012.
Nowadays, electric power is coming to be seen as a solution to the pollution caused by cars, e.g. reducing the CO2 emissions [Ber2010]. In new vehicles the number of electrical actuators is growing steadily. They are used for example (Figure 1.1):
• To compress the cooling gas and provide the conditioned air.
• actuate the windscreen wipers.
• to assist the driver for steering; it gives more assistance as the vehicle slows down, and less at higher speeds and the system has an advantage in fuel efficiency compared to a hydraulic system because there is no belt-driven hydraulic pump constantly running.
However the challenge is to develop electrical powertrain for vehicle propulsion. After the first attempts at the beginning of the 20th century, the main challenge was the introduction of cost effective electrical power in the propulsion. In fact, the key points are to:
– Embed enough energy in the vehicle to guarantee enough autonomy.
– Guarantee at least the same level of security and performances as a vehicle propelled with Internal Combustion Engine (ICE).
The most mature technologies for energy storage devices in electrical propulsion are electrochemical batteries and fuel cells. Figure 1.2 shows the different solutions sorted depending on their level of maturity and on their high energy or high power capability.
Many cars using electrical propulsion are already commercially available (Honda Fit EV, Nissan Leaf, Renault Kangoo, Zoe and Fluence, Toyota Prius, Peugeot iOn, Citroen C-Zero and Mitsubishi i-Miev). However, the first step in the use of storage device was its combination with ICE: hybrid vehicle or hybrid electrical vehicle.
Hybrid vehicles are often classified into four categories: Micro hybrid, Mild hybrid, Full hybrid and plug-in hybrid (Figure 1.3). Micro hybrids are essentially conventional vehicles with an starter- alternator allowing the engine to be turned off whenever the car is coasting, braking, or stopped, yet restart quickly and cleanly [Bru2009]. In the Mild hybrids, the starter-alternator is replaced by a more powerful electrical motor improving the regenerative braking [Mor2012]. A full hybrid EV can be propelled by the ICE, the electrical motor or both [Sta2006]. A plug-in hybrid electric vehicle (PHEV) is a full hybrid able to run in electric-only mode, with larger batteries and the ability to recharge from the electric power grid. Their main benefit is that they can be gasoline-independent for daily commuting, but also have the extended range of a hybrid for long trips. Recently, it has been observed many improvements for conventional vehicles in terms of consumption (injection) or pollution (catalytic converters). Although these changes have been applied to hybrid vehicles, their electrical part grows. Obviously, there is a logical trend toward complete electrification of the entire vehicle. In this evolution, the structure of the powertrain will be simplified by the use of only electric motors, which will allow a more significant reduction of the pollution [Raj2013].
Fault detection and diagnosis principle
Let us first underline the difference between a failure and a fault. A failure is the inability of an element or a system to perform a function as required while a fault is an abnormal condition that can cause an element or a system to fail [Iser2006]. For example, a steering column locked is a failure and an open resistor a fault.
In industrial systems, the monitoring system, which includes hardware (data collection) and software processes the data (monitors fault conditions and generates alarms) and gives a graphical representation updated periodically. Its role is to assist the human operator for emergency management to increase the reliability, availability and functional safety. Supervision takes place in a hierarchical structure (with at least two levels), and covers the occurrence of normal and abnormal operation: the faults are modeled based on events or binary states [Bat2011]. The main activities involved in the supervision of a continuous system are grouped in the fault tolerant control which is divided into three stages: detection, isolation and reconfiguration. The detection consists in identifying online changes on variable behavior. The isolation is then to discriminate which faulty variable has generated the symptom detected from the alarms enabled in the multi-sources case. Then the reconfiguration is to provide corrective control action to bring the system to normal operation (or in a limp mode that is to say, tolerated by the system).
The concept of integrating automated fault detection and diagnosis came at the beginning of the 1980’s, as a functionality of supervision systems. The specification for a monitoring system is to determine the relevant equipments to supervise, while minimizing false alarms, non-detections and delays in the detection. Obviously, constraints in terms of computer memory and costs must be taken into account. Therefore, there is always a compromise between the number of faults to monitor and diagnosis performance. Generally, relevant equipments can be determined off-line by the FMEA (Failure Mode and Effects Analysis) [SAE1967]. Then after the fault detection, different actions can be taken to ensure the continuity of the process even if in some cases only a safe stop is possible. However, a little attention has been paid to the analysis and design with the overall system structure and interaction between Fault Detection and Diagnosis (FDD) and Reconfigurable Control (RC).
How to analyze systematically the interaction between FDD and RC? How to design the FDD and RC in an integrated manner for on-line and real-time applications?
These challenging issues still remain opened for further research and the solutions vary, depending on the severity and the fault type (gradual, abrupt and intermittent). Historically, a significant amount of research on FDD systems was motivated by aircraft flight control system designs and nuclear power plants. The goal, therein, was to provide a safe mode in the event of severe faults. More recently, the fault-detection and diagnosis problem has begun to draw more and more attention in a wider range of industrial and academic communities such as automotive industry, due to increased safety and reliability demands beyond what a conventional control system can offer [Cap2007]. For high-critical safety systems (nuclear plants, aerospace industry), the designers multiply the hardware system so that, for a failure in one system, another system will immediately replace it. This type of maintenance (material redundancy) is clearly too expensive and adds complexity to the design. One can also use an analytical redundancy called software redundancy which models the healthy operation of a certain part of the system. With recent progress in high performance processors, software redundancy has become a promising alternative. This last solution is much more attractive because of its flexibility and its capability of evolution. The diagnosis methods are mainly classified into two categories: model-based (quantitative) and data-driven-based (qualitative) methods (Figure 1.5).
Fault Tolerant Control or Design
For certain faults, the two-step classical approach, that consists in designing the process and then build the controller (including diagnosis and reconfiguration) has to be thought differently. So, in some cases, the question is how to design the component (electrical machine or inverter) so as to facilitate the diagnosis and improve the reconfiguration and availability capability. In aerospace applications, machines that have been designed with fault tolerance in mind are found in [Atk2005, Ram2011 and Ben2011]. The case of automotive (electric cars) industry is also mentioned by [Bia2003, Wel2004].
For the design of the actuators, increase the number of stator phases to achieve high reliability for the power stage has shown his effectiveness and in addition, a functional post-fault structure may lead to generate the references for the switching patterns [Bau2012].
Modern technological systems rely heavily on sophisticated control systems to meet safety requirements. This is particularly the case of safety critical applications such as automobile industry where a minor and often fault could potentially lead to unwanted events. To prevent fault induced losses and to minimize the potential risks, new control techniques and design approaches need to be developed to cope with system component malfunctions while maintaining a good limp mode. In automotive industry, it is highly appreciated that the faulty drive could remain operating, of course apart the cases of unrecoverable damages to the motor or to the power supply. The most common way is to do complete hardware redundancy but it has an additional cost that cannot be ignored and the objective of FTCS is to reduce this cost while having a safe system. Generally, for sensors, a fault tolerant design is not needed but for a specific method (High-Frequency signal injection for position estimation), the design has its importance. In this case, the rotor position is estimated by exploiting the magnetic saliency. The degree of saliency, and therefore the resulting inductance variation, is highly dependent upon machine design, and comparisons between different PM machines have shown that, under specific operating conditions, certain rotor topologies and saliency profiles are more suitable to sensorless control than others. So, the rotor topology influences the selection of an appropriate injection-based sensorless control strategy, particularly for use under high-load conditions and rotor geometry has significant influence on the accuracy of the estimation method. Therefore, consideration may be given to the saliency profile at the design stage of the PM machine. In [Wro2011], a high-frequency injection sensorless position estimation algorithm has been directly incorporated into the Finite Element design procedure for an IPM machine, resulting in a machine that exhibits the required electrical characteristics while being tailored for sensorless control.
For the power converter failures, many strategies have been proposed to reconfigure the control of the machine. The best way here is to perform hardware redundancy due to the difficulty to have an effective component model. Moreover, these strategies need mostly, for the reconfiguration, to have access to the neutral point of the machine (an access to the neutral point should have been planned or the phases of the machine decoupled) to continue to operate safely in a degraded mode. As an example, for a good fault tolerant control of a permanent magnet machine drive, the machine should be driven from separate single phase bridges, and must be capable of withstanding a short-circuit terminal fault.
For sensor fault, FTC doesn’t need, in general to reconfigure the control but only ensuring a quick switch from the faulty sensor to a software sensor, which is a state reconstruction. However, the switch strategy must be performed so as to ensure a smooth transition and avoid a harsh transient. According to this state of the art, for fault detection and diagnosis these following issues still need to be addressed:
– Initialization and low speed operation of rotor position observer.
– Fault Detection and identification of DC-link voltage and phase current sensors.
– Information on the detection time of FDD methods and their consistency with functional safety rate of coverage.
– Estimation of DC-link voltage.
Combined architecture for traction and battery charging
The objective of the SOFRACI (high eﬃciency Inverter with integrated charge function) project was to develop an innovative architecture for achieving the functions of traction, regeneration and charging [Sil2009]. To do so, a unique powertrain is used for the charging and the traction of the vehicle, allowing to:
– Reduce the global number of components.
– Integrate charging function without additional electronic component.
– Provide a limp mode to the user in case of critical faults.
Because the traction and charging modes cannot be simultaneous, the inductances of the machine and the power converter can be used to perform the battery charging. The level of current is enforced by the traction design, which requires high power to drive the vehicle; then it is also possible to sustain high power in charging mode. The power converter used for traction has already been used for battery charging [Shi1994]. However, high current relays were required to pass through one mode to the other. In addition, these electromechanical components increase the design complexity.
Figure 2.1 shows the proposed topology. In charging mode, the three-phase grid is connected to the midpoints a’’, b’’ and c’’ of each machine’s coils and no relay is necessary. It could be possible to connect the grid to the extremities a, b, c or a’, b’, c’ but the charging three-phase currents would generate an undesirable rotating magnetic field [Bru2010]. By connecting the grid to the midpoints of each phase, the current will be split in two equal and opposite components and therefore nullify the Magneto-Motive Force at the stator level. This cancellation ensures a magnetic decoupling between the rotor and the stator. Another consequence is that the same amount of current flows in each leg of the same bridge and is in phase with the 50Hz or 60Hz input voltage [Lac2013]. The ac current is therefore rectified to source the DC link (Figure 2.1b). Then each power switch can be sized with the half maximum current.
PMSM and Sensors
The energy conversion chain is composed of: an electrical machine (PMSM), a power supply, a power converter and sensors for the measurements required by the control loop. Synchronous motors with rare earth permanent magnets have higher power density than comparable DC motors because there is no limiting effect due to the mechanical commutator [Ref2006]. The SOFRACI electrical machine is a poly-phase Permanent Magnet Synchronous Machine (PMSM) designed by Leroy SOMER. The machine is a radial flux type and has been chosen for its robustness and reliability properties. It can develop high torque density and has the ability to reach high speeds [San2013]. This makes it practical to magnetise the pole pieces separately rather than magnetising the complete assembly and requires also much less space.
The windings are distributed around the stator teeth with concentrated coils (Figure 2.2). This allows to limit short-circuit phases and to reduce conductors Joule losses [Kha2011]. The PMSM has 12 stator slots and 4 pairs of poles on the rotor with a rated nominal power of 15kW and a maximal speed of 13000 rpm. The eight magnets with a length of 40 mm are in Neodymium Iron Boron (NdFeB).
Implementation of the regulators requires measurements of the rotor position, the phase currents and the DC link voltage. An incremental encoder BEI KHO5S14 with block commutation (bandwidth of 300 kHz, resolution up to 5000 points) gives the mechanical position of the rotor. It provides three signals: A and B shifted by 90° and each step of A (and then also B) indicates that the encoder has performed one step; the third signal is a top zero which is used to readjust the electronic counter in synchronism with the zero position of the encoder. The zero position indexing is obtained by rotation of the sensor, following an angle of ±10°. The block commutation signals are transmitted as TTL square-wave signals. Then an incremental encoder interface DS3001board provides the position in radians.
The current measurements are performed by 3 Hall Effect LEM transducers (LEM LA 55-P). They can measure DC and AC currents up to 50 A (bandwidth of 200 kHz, accuracy of ±0.65%). The DC bus voltage is measured by a Varistor V661BA60.
Inverter and Power supply
The ARCEL power converter consists of six IGBT legs (Figure 2.3). The decoupling of the DC bus is performed by a series of polypropylene capacitors with a total value of 2.3 mF.
The IGBTs from Fuji Electric Device Technology (1200 V, 450 A) are placed on an air-cooled radiator using a centrifugal turbine (Figure 2.4). The minimal dead-time between the opening and closing of the switches is 1.5 µs.
For the converter control, the 2SD316EI dual-channel driver based on CONCEPT’s proprietary SCALE chipset is used. It allows managing the half bridge mode or a free control of the two switches of the leg. Each channel of the SCALE driver is equipped with a monitoring circuit for over-current and short-circuits issues. When the four windings are coupled in parallel for each phase, the maximum DC voltage is limited to 450V while the maximal RMS current per phase is 30A.
Real-Time implementation of the controller
The control of electrical drives is executed by microelectronic components, which are manufactured from semiconductor materials. The important progress in semiconductor technology combined with high speed signal processing at reasonable cost has been the essential force behind the development of high performance electrical drives controller [Men2011]. Over the last decades, the micro-processors have also become both increasingly powerful and affordable.
Embedded systems require flexible and high-performance architectures to support the increasing amount of programmability. Digital implementation has allowed limiting the high sensibility of systems with respect to thermal and electromagnetic constraints in contrast to analogical implementation [Gat2010]. To this purpose, hardware and/or software design are used to implement the controller that fits to a given algorithm. The DSP (Digital Signal Processor) and the micro-controller are well suited for most of the software designers because they are easier to reconfigure and are cost effective. Hardware implementation is defined by the use of Field-Programmable Gate Array (FPGA) or Application-Specific Integrated Circuit (ASIC). They are very useful for their ability to process signals beyond high frequencies (beyond GHz) and below low frequencies (a few Hz), with a short execution time.
Recent advances in the synthesis have prompted mixed implementation referred to as hardware/software co-design [Bah2013]. It consists to allocating to each target a functionality of the algorithm. The Hardware/Software is partitioned with respect to the control requirements (bandwidth and stability margin) and the architectural constraints (e.g., available area, memory, and hardware multipliers). For e.g., in electrical machine control, the tasks requesting a critical computational time (generation of PWM signals) are run on hardware targets while the regulation of the currents and the speed are executed by the processor.
A mixed implementation will be used for the experimental tests on SOFRACI bench, with a DS1006 Processor board and a DS5203 FPGA board. The illustration of the mixed implementation applied to the electrical drive is shown in Figure 2.5.
Table of contents :
Chapter I: State of the art of electrical drives fault tolerant control
1.1 Electrical drives in automotive systems
1.2 Fault Management
1.2.1 Fault Detection and Diagnosis principle
1.2.2 Fault Tolerant Control principle
1.3 Application to an electrical system
1.3.1 Fault types in electrical drive
184.108.40.206 Actuator faults
220.127.116.11 Sensor faults
18.104.22.168 Electrical Machine faults
1.3.2 Fault Diagnosis and Fault Tolerant Control for electrical drives
22.214.171.124 Fault Diagnosis
126.96.36.199 Fault Tolerant Control or Design
Chapter II: SOFRACI Platform modelling and description
2.1 SOFRACI Structure: A system designed for Fault Tolerance
2.1.1 Combined architecture for traction and battery charging
2.1.2 Diagnosis capabilities
2.2 Bench description
2.2.1 PMSM and Sensors
2.2.2 Power supply and Inverter
2.2.3 Real-time hardware implementation of the controller
2.3 PMSM modelling for control
2.3.1 Electrical equations
2.3.2 Mechanical equations
2.3.3 Nonlinear model for state space representation
2.4 Electrical Drive Control in traction mode
2.4.1 PI controller synthesis
188.8.131.52 Synthesis by identification to a 2nd order system
184.108.40.206 Speed Controller
2.4.2 3H bridge Inverter model
2.4.3 Simulation and Experimental results of the Torque Control
220.127.116.11 Simulation Results
18.104.22.168 Experimental Results
Chapter III: Position/Speed Sensor Fault Tolerant Control
3.1 Impact of a sensor failure on the PMSM Control
3.2 Position /speed Estimators and their various uses
3.2.1 Extended Kalman Filter
3.2.2 Back-EMF based Observer
3.2.3 High Frequency Signal Injection
3.3 New Estimator based on a Differential Algebraic Approach
3.3.1 Observation Principle
3.3.2 Differential Algebraic Estimation in PMSM
22.214.171.124 Position/Speed Estimator Synthesis
126.96.36.199 Stability Analysis
3.3.3 Position/Speed Estimation Results of the Differential Algebraic Estimator
188.8.131.52 Operation with sensor
184.108.40.206 Sensorless Control
220.127.116.11 Robustness Issue
3.3.4 Comparison of the three estimators
3.4 Position /Speed Sensor Fault Detection, Isolation and Reconfiguration
3.4.1 Sensor Fault Enabling based on Observers
Chapter IV: Phase Current and DC bus voltage sensors Fault Detection and Diagnosis
4.1 Needs on phase currents sensors and DC link voltage sensors diagnosis
4.1.1 Current Sensor fault origins and consequences
4.1.2 DC Voltage Measurement
4.1.3 Existing methods based on Observers and Signal processing
4.2 Developed methods for Phase Currents Sensors Fault Detection and Isolation
4.2.1 Diagnosis by an algebraic approach of fault estimation
18.104.22.168 Design and Simulation Results
22.214.171.124 Experimental Results in FDI scheme
4.2.2 Current vector analysis for Fault Detection and Diagnosis
126.96.36.199 Current residuals analysis in the (d,q) frame
188.8.131.52 Simulation Results
184.108.40.206 Experimental Results
4.3 DC Link Voltage Observer
Conclusions and Perspectives