You are on page 1of 8

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/327721264

Torque Vectoring Control

Conference Paper · July 2018

CITATION READS

1 2,585

3 authors, including:

Derong Yang Mats Jonasson


Volvo Car Corporation Chalmers University of Technology
20 PUBLICATIONS   87 CITATIONS    46 PUBLICATIONS   283 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Evaluation of Force Allocation View project

Brake blending control strategy in the hybrid/electric vehicles View project

All content following this page was uploaded by Mats Jonasson on 10 October 2018.

The user has requested enhancement of the downloaded file.


AVEC’18

Torque Vectoring Control for Progressive Cornering


Performance in AWD Electric Vehicles
Derong Yang1, Martin Idegren1, Mats Jonasson1,2
1Volvo Cars Corporation, Department of Active Safety and Vehicle Dynamics, Gothenburg,
SE-405 31, SWEDEN
2The Royal Institute of Technology (KTH), Vehicle Dynamics, Department of Aeronautical

and Vehicle Engineering, Stockholm, SE-10044, SWEDEN


E-mail: derong.yang@volvocars.com

In the present paper, a real-time integrated Torque Vectoring Control function is designed and
implemented in an AWD axle-split hybrid vehicle. The front axle has a conventional combustion
engine, and two individually controlled electric motors are located at the rear axle. The function
aims to enhance the vehicle cornering performance by yaw torque control allocation, at steady-state
and transient steering maneuvers, with different propulsion inputs. Preliminary results have shown
that the controller can effectively increase turn-in precision, prevent the vehicle understeer, thus
reduce the steering effort and oscillation at progressive cornering.

Topics /Yaw Response, Torque Vectoring, Vehicle Dynamics, Integrated Control, Electric Motors

1. INTRODUCTION 2) Due to combined-slip effect at the tire contact


patch, a direct yaw torque from eTV will inevitably
In the recent decade, the modern car industry has
generate a consequent indirect yaw torque, which is one
never been facing a more demanding challenge than
important principle of electric all-wheel-drive system
before, to minimize the emission level and thus the
(eAWD). The eTV presented here takes this effect into
environmental impact. Along with the enhanced
account and is therefore naturally integrated with an
performance of energy storage units as well as the power
eAWD system using control allocation.
of electric motors, a fleet of mild/primary hybrid and full
3) Steady-state and transient yaw response during
electric cars with competitive mile range are stepping
acceleration, constant-speed, and deceleration. A precise
into the markets, e.g. Toyota Prius and Tesla [1, 2]. The
turn-in and progressive cornering performance should
unique powertrain architecture of electric cars with
deliver a truly sporty electric car driving experience. This
multiple electric motors opens up new possibilities to
is ought to be achieved with no conflict with vehicle
control the vehicle wheel forces independently at fast
controllability and stability margins.
response and high precision. This type of control function
that modulates the individual-wheel forces at dynamic
2. TEST VEHICLE
driving is often called Torque Vectoring (TV). Previous
research and engineering work have popularly The research vehicle is a hybrid two-door sports
investigated various TV actuator mechanisms at both coupé. The car is equipped with a combustion engine at
electric and non-electric vehicles, where mechanically or the front axle, and two electric motors at the rear axle.
electro-mechanical controlled clutch and differential are Each electric motor has the top power and torque outputs
used to distribute the drive torques left/right wise [3-6]. approx. as 85 kW and 240 Nm. This corresponds to about
Owing to the advancement of TV technologies, vehicle 1800 Nm at the wheel. In total, the car powertrain could
stability and yaw response can be simultaneously deliver approx. 600 hp and 440 kW. Fig.1 below shows
optimized based on the standard direct yaw moment the prototype model displayed in the vehicle dynamics
control algorithm for example in [7-10], without simulation tool CarMaker®.
comprising other crucial vehicle properties such as fuel
economy and ride comfort.
The present paper will focus on:
1) Electric torque vectoring (eTV) where two electric
motors will be controlling the drive/brake torques at the
left and right wheels respectively. The controller is
demonstrated at a primary hybrid car with the electric
motors only at rear axle, although the algorithm itself is
expected to be applicable to full electric cars with
different amount of electric motors at the drivetrain.
Fig. 1 Concept prototype in CarMaker®
AVEC’18

The basic vehicle parameters are listed in Table 1: 4. REFERENCE YAW TORQUE CALCULATION
Table 1 Vehicle Parameters 4.1 Yaw Response Control
vehicle mass m 2414 kg The first contribution of the yaw torque is for the
vehicle CoG height hcog 0.492 m yaw response control, namely a feedforward control part
to prevent understeer and a feedback control part to
front axle to CoG lf 1.472 m
correct and minimize understeer if that ever happens.
rear axle to CoG lr 1.268 m The vehicle model used here for control allocation is
vehicle track width w 1.63 m a common two-track dynamic model:
vehicle yaw inertia Izz 6138 kgm2
dynamic wheel radius rw 0.33 m 𝛿𝑓 𝑣𝑥 𝛿𝑓
𝑓𝑥𝐹𝐿 𝑓𝑥𝐹𝑅
3. CONTROL ARCHITECTURE
𝑣𝑦
𝑓𝑦𝐹𝐿 𝐹𝑅
Fig. 2 below shows the control architecture of the 𝑇𝑥 𝑓𝑦
proposed integrated controller. It is integrated from three 𝑙𝑓
aspects: 𝜔𝑧
1. Control allocation method is applied to fulfill both
drive torque and yaw torque requests at the same 𝑤 𝐿
𝑀𝑧
step; 2
2. The two customer functions, i.e. eAWD and eTV are
𝑤 𝑙𝑟
integrated with the common purpose to improve yaw 𝑓𝑥𝑅𝐿 = 𝑢3 𝑓𝑥𝑅𝑅 = 𝑢4
motion control; 2
3. At dynamic driving, there are both indirect yaw 𝑓𝑦𝑅𝐿 𝑓𝑦𝑅𝑅
torque due to the change of lateral tyre forces at the
friction circle, as well as direct yaw torque due to the
Fig. 3 Two-track dynamic vehicle model
differential brake/drive torques from the longitudinal
tyre forces. The control allocator here is augmented
The system state-space model can be formulated
such that the indirect yaw torque is taken into
using a single-track version of the vehicle model in Fig.3:
account. The control allocation matrix is therefore
augmented with two extra parameters i.e. a and k.
𝑥̇ = 𝐴𝑥 + 𝐵𝑢 (1)
See Section 5 below for detailed synthesis.
There are in principle two vehicle motion control 𝑣𝑦 𝛿
𝑥 = [𝜔 ] , 𝑢 = [ 𝑓 ] (2)
modules involved in the architecture. Engine Control 𝑧 𝑀𝑧
Module (ECM), which calculates the total longitudinal
torque request from the driver pedal inputs (TxReq), and 𝑣̇𝑦 𝑣𝑦 𝛿
delivers torque requests to the different electric motors. [ ] = 𝐴 [𝜔 ] + 𝐵 [ 𝑓 ] (3)
𝜔𝑧̇ 𝑧 𝑀𝑧
Vehicle Dynamics Control Module (VDCM), which
calculates the reference yaw torque (MzReq), stability 𝐶𝑓 +𝐶𝑟 𝑙𝑓 𝐶𝑓 −𝑙𝑟 𝐶𝑟
limits, indirect yaw torque parameters (a, k), and − (−𝑣𝑥 − )
𝑚𝑣𝑥 𝑚𝑣𝑥
eventually the allocated torque requests to individual Where 𝐴 = [ 𝑙𝑓 𝐶𝑓 −𝑙𝑟 𝐶𝑟
2
𝑙𝑓 𝐶𝑓 +𝑙𝑟2 𝐶𝑟
]
actuators, in our case, the ICE engine (TqReqICE) and − −
𝐼𝑧𝑧 𝑣𝑥 𝐼𝑧𝑧 𝑣𝑥
the two electric rear axle drives (TqReqErad). At this
stage, individual wheel friction brakes interface is not 𝐶𝑓
established in the proposed controller. The following 0
𝑚
sections will provide a dive into each component 𝐵= 𝑙 𝐶 1
𝑓 𝑓
presented in this architecture. [ 𝐼𝑧𝑧 𝐼𝑧𝑧 ]

where 𝑀𝑧 is seen as the generated yaw torque by


longitudinal forces, including both the direct yaw torque
from differential drive/brake forces left/right wise:
𝑤
𝑀𝑧𝑑𝑖𝑟 = (𝑓𝑥𝐹𝐿 𝑐𝑜𝑠 𝛿𝑓 − 𝑓𝑥𝐹𝑅 𝑐𝑜𝑠 𝛿𝑓 + 𝑓𝑥𝑅𝐿 − 𝑓𝑥𝑅𝑅 )
2
+ sin 𝛿𝑓 𝑙𝑓 (𝑓𝑥𝐹𝐿 + 𝑓𝑥𝐹𝑅 ) (4)

The indirect yaw torque from tyre lateral forces become:

Fig. 2 Control architecture of the integrated eAWD+eTV 𝑀𝑧𝑖𝑛𝑑𝑖𝑟 =


controller (𝑓𝑦𝐹𝐿 − 𝑓𝑦0
𝐹𝐿
) · 𝑐𝑜𝑠 𝛿𝑓 𝑙𝑓 + (𝑓𝑦𝐹𝑅 − 𝑓𝑦0
𝐹𝑅
) · 𝑐𝑜𝑠 𝛿𝑓 𝑙𝑓
−(𝑓𝑦𝑅𝐿 − 𝑓𝑦0
𝑅𝐿
) · 𝑙𝑟 − (𝑓𝑦𝑅𝑅 − 𝑓𝑦0
𝑅𝑅
) · 𝑙𝑟 (5)
AVEC’18

The control laws for steering and torque vectoring larger the indirect yaw torque. The same effect goes with
yaw torque request are then expressed as a steady-state the total drive torque or say the longitudinal acceleration.
solution of Eq.3 above: Fig. 4 shows one example case of a static analysis given
𝑟𝑒𝑓 𝐿 different lateral accelerations. Fig.5 shows the result
𝛿𝑓 =( + 𝐾𝑢𝑠 ) 𝑎𝑦 (6) given different drive torques. Hereby we use the Eq.10
𝑣𝑥2
𝑟𝑒𝑓 1 below to support the augmented matrix in Eq. 9:
𝑀𝑧𝐹𝐹 =− (𝐴21 𝑣𝑦 + 𝐴22 𝜔𝑧 + 𝐵21 𝛿𝑓 ) (7)
𝐵22
𝑇𝑥𝑟𝑒𝑎𝑟
𝑀𝑧 𝑖𝑛𝑑𝑖𝑟 = 𝑘(𝜀 − 𝑎) , where 𝜀 = (10)
𝑇𝑥
𝑟𝑒𝑓
Here the reference steering 𝛿𝑓 is used for the feedback
𝑟𝑒𝑓
understeer control (𝑀𝑧𝐹𝐵 ),
and it includes the parameter
vehicle understeer coefficient 𝐾𝑢𝑠 , which is a tuning
parameter based on the online cornering stiffness
estimation. As shown, the reference yaw torque is based
on the vehicle side slip angle, which can be a fusion
between the steady-state estimation and the
measurement. Here 𝑎𝑦 is the vehicle lateral acceleration,
𝜔𝑧 is the vehicle yaw rate.

4.2 Yaw Damping Control


To prevent any conflict with the existing stability Fig. 4 A conceptual illustration, change of indirect yaw
functions on board, a common practice is to design some torque at varying ratio of longitudinal torque distribution
sort of transition controller for control smoothness, rear/total.
performance and system safety. Here we ramp out the
excessive yaw torque by applying a similar algorithm as
the ESC oversteer control function, where the vehicle
yaw rate and rear axle side slip angle are monitored. A
simple proportional feedback controller is designed to
generate a yaw damping control component. In total, the
reference yaw torque becomes a weighted sum of the
control intervention for yaw response in Section 4.1, and
𝑟𝑒𝑓
yaw damping (𝑀𝑧𝑂𝑆 ), OS here stands for oversteer:
𝑟𝑒𝑓 𝑟𝑒𝑓 𝑟𝑒𝑓 𝑟𝑒𝑓
𝑀𝑧 = 𝑤1 𝑀𝑧𝐹𝐹 + 𝑤2 𝑀𝑧𝐹𝐵 + 𝑤3 𝑀𝑧𝑂𝑆 (8)

5. CONTROL ALLOCATION
As shown in the Fig.2 above, the calculated
reference yaw torque and drive torque are then cascaded
to the actuators. We augment hereby the allocation Fig. 5 Change of indirect yaw torque at varying ratio of
problem using the contribution from indirect yaw torque: longitudinal torque distribution rear/total, with different
𝑥 = 𝐵𝑢 (9) lateral accelerations @ 𝑇𝑥 𝑅𝑒𝑞 = 1500 𝑁𝑚. Static
where: analysis, where small steering angles is assumed.
𝑟𝑒𝑓
𝑇
𝑥 = [ 𝑟𝑒𝑓𝑥 ]
𝑀𝑧 + 𝑎𝑘
cos 𝛿𝑓 1 1 1
𝐵 = [sin 𝛿𝑓 𝑙𝑓 𝑘 𝑤 𝑘 𝑤 𝑘 ]
𝑟𝑒𝑓
− + 𝑟𝑒𝑓 + 𝑟𝑒𝑓
𝑟𝑤 𝑇𝑥 2𝑟𝑤 𝑇𝑥 2𝑟 𝑤 𝑇𝑥

𝑢 = [𝑢1 𝑢2 𝑢3 𝑢4 ]𝑇
Here 𝑢1 is the ICE torque, 𝑢2 is the Erad torque (if only
one Erad at the axle), 𝑢3 is the Erad torque at the left
wheel, 𝑢4 is the Erad torque at the right wheel. The
construction is done in this way to show capability of a
scalable electric drive configuration with different
amount of electric motors at the drivetrain.
We found a (quasi) linear relationship of the
front/rear distribution of drive torques and the Fig. 6 Change of indirect yaw torque at varying ratio of
consequent indirect yaw torque due to combined slip. As longitudinal torque distribution rear/total, with different
shown in Fig.3, the larger the lateral acceleration the driving torque requests @ 𝑎𝑦 = 5 𝑚/𝑠 2 . Static analysis,
where small steering angles is assumed.
AVEC’18

The cost function can be then written as:


𝐶𝑜𝑠𝑡 = min (𝑥 − 𝐵𝑢)𝑇 𝑄(𝑥 − 𝐵𝑢) (11)
𝑢𝑚𝑖𝑛 ≤𝑢≤𝑢𝑚𝑎𝑥

where 𝑢𝑚𝑖𝑛 and 𝑢𝑚𝑎𝑥 are the lower and upper limit of
the actuator capacity and stability margins. 𝑄 is a
𝑟𝑒𝑓 𝑟𝑒𝑓
diagonal matrix that weights between 𝑇𝑥 and 𝑀𝑧 .
This constrained optimal control problem is relatively
small-scale so it can be solved by different quadratic
programing solvers.

6. SIMULATION RESULTS
6.1 Turn-in Ramp Steer
Fig. 8 Vehicle understeer characteristic, ramp steer
The controller performance is first tested in a ramp
response with mild acceleration.
steer maneuver at constant radius curve scenario, as
compared to without the function. The results are
obtained in the CarMaker® closed-loop simulation
environment with the calibrated nonlinear vehicle
dynamics and powertrain model. The driving starts at
85 km/h, while entering a curve of 80 m radius. In this
case, the driver model has full autonomy to control the
curve speed and the steering intensity during the
cornering maneuver. Here the acceleration pedal is Fig. 9 Steering wheel angle (SWA), ramp steer response
applied in a relative mild way and almost throughout the with mild acceleration.
whole curve, except right at the entry where the driver
slows down a little bit at the transient steering phase, see 6.2 Power on Curve
Fig. 7. This scenario is thus a typical highway entry case.
In Fig. 8, the yaw response is measured as the ratio Further, to demonstrate the complete control
of steer angle and lateral acceleration, i.e. vehicle contents here, we select a power on curve scenario with
understeer. It can be seen that with TV control, the actual considerable longitudinal and lateral acceleration, where
steer angle could follow the reference one (Eq. 6), which the tyre friction is well saturated. The intervention
is calculated based on the vehicle designed understeer therefore includes the preventive understeer and
gradient with online state and parameter estimation. oversteer torque vectoring control. The purpose of this
Directly before reaching the lateral acceleration 8 𝑚/𝑠 2 , scenario is also to compare the result with and without
the driver steers back to zero by the end of the curve, considering the indirect yaw torque in the control
therefore the reference steer angle is shown as higher allocation. In this case, the car starts at 60 km/h, and
than the actual one. In Fig. 9, it is shown that without TV enters a road of 90 m radius, with dry asphalt pavement.
control, driver obviously needed to steer faster and more, A number of accelerations are applied during the driving.
as compared to with TV control, which appears to be a Comparing the actual and reference steering wheel
smooth, precise and more confident turn-in. This mild angle (SWA) in Fig. 10 and Fig. 11, one can notice that
acceleration case does not need to trigger the understeer, considering indirect yaw torque in the control allocation
neither the oversteer control components, while only the will result in a more precise turn-in control. At Time = 5s,
feedforward control is tuned to achieve a predictable a sudden increase of gas pedal input at nearly saturated
controllable cornering experience. tyres introduce severe understeer of the car, where
understeer control intervention is observed. The vehicle
yaw rate and side slip angle is thus built up quickly
(Fig. 12), which further triggers an oversteer control
intervention (Fig. 10). Looking at the results for the
controller without considering indirect yaw torque, at the
critical moment when power is requested, the vehicle is
predicted by our controller to be understeered therefore a
sharp yaw torque is requested from VDCM. However,
this yaw correction torque is not fulfilled in time as
shown in Fig. 14 (Time = 5s ~ 6s). This inevitably causes
a larger uncertainty and oscillation in the following part
of the manoeuvre, where the driver was forced to apply
Fig. 7 Vehicle speed, ramp steer response with mild an unreasonably big steering input though helps
acceleration. marginally. While with the indirect yaw torque
allocation, the manoeuvre turned out to be much more
controllable (Fig. 10, Fig. 13) with a clean understeer and
oversteer control sequence. Although it is noticed that the
AVEC’18

allocation of yaw torque is still not perfect, due to certain


model error in tyre force estimation and possibly also the
measurement noise in the drive torque estimation. It is
also interesting to note that, with indirect yaw torque, the
allocator tends to find the solution where yaw is
prioritized than drive torque which is understandable. On
the other hand, the control without considering indirect
yaw torque prioritizes the drive torque instead.

Fig. 13 Total requested torque vs. total actual torque,


integrated control sequence, power on curve maneuver.

Fig. 10 Steering angle, integrated control sequence,


power on curve maneuver.

Fig. 14 Total requested torque vs. total actual torque,


integrated (not consider indirect yaw torque) control
sequence, power on curve maneuver.

Fig. 11 Steering angle, integrated (not consider indirect The first example is an intensive slalom manoeuvre
yaw torque) control sequence, power on curve maneuver. on a high road friction surface, at about 80 km/h with
power/gas pulse in between. Fig.15 shows the vehicle
states together with the reference yaw torque request
expressed at the wheel Mb. A close scrutinize exhibits a
nice follow-up of yaw intervention whenever the vehicle
is powered at the tyre limits specially. Steering angle is
matching with the reference fairly well so that the
understeer control did not kick in. On the high friction
road, it is desired to make the Torque Vectoring control
mainly by feedforward control, which is shown in
Fig. 17. The yaw damping control component nicely
jumps in due to the excessive yaw rated monitored by the
controller, Fig. 16.
The second example is a relatively low frequency
slalom manoeuvre with smoother drive torque inputs, at
an ice track with fluffy snow. There a completely
Fig. 12 Vehicle states, at control considering indirect yaw different control sequence is exhibited where the
torque (solid) vs. only direct yaw torque (dashed), power feedback and yaw damping control become more active
on curve maneuver. (Fig. 20), given the same matching of desired steering
angle and the actual one (Fig. 18). Fig. 19 shows the
7. EXPERIMENTS progressive yaw performance in a controlled manner.
Good to mention that none of the driving cases presented
Here we present the traces from the experiments in a
here have triggered intervention from the conventional
test car with preliminary control interfaces, where a rule-
vehicle stability control systems. This shows a great
based allocation method is used as a robust benchmark
potential in real-time implementation and further
before complicating the implementation. Hence, the
function development.
following results involve no online optimization.
AVEC’18

generated by the tyre lateral forces is considered. Hence


a more accurate cascading of yaw torque request is
observed for more predictable and progressive cornering
performance.

Fig. 15 Vehicle speed, lateral acceleration, reference


yaw torque at wheel, wheel steer angle (actual vs.
reference). High frequency slalom maneuver on dry
asphalt track.
Fig. 18 Vehicle speed, lateral acceleration, reference
yaw torque at wheel, wheel steer angle (actual vs.
reference). Low frequency slalom maneuver on ice track
with fluffy snow.

Fig. 16 Actual yaw rate vs. reference yaw rate for yaw
damping control. High frequency slalom maneuver on
dry asphalt track.

Fig. 19 Actual yaw rate vs. reference yaw rate for yaw
damping control. Low frequency slalom maneuver on ice
track with fluffy snow.

Fig. 17 Interventions of the three control components:


feedforward, feedback and yaw damping control. High
frequency slalom maneuver on dry asphalt track.

8. CONCLUSIONS
Fig. 20 Interventions of the three control components:
In this paper, a yaw-based integrated torque feedforward, feedback and yaw damping control. Low
vectoring control function is designed and implemented frequency slalom maneuver on ice track with fluffy
in a hybrid car with two electric motors at the rear axle. snow.
Reference yaw torque is calculated for three control
purposes: feedforward response, understeer correction, The integrated controller is preliminarily verified in
yaw damping. Furthermore, the algorithm integrates the the simulation and real-time vehicle testing environment,
eAWD and eTV function in a novel way using the at both low and high friction adhesion roads. Future work
reference yaw torque. This integration is achieved by is expected to improve the coordination and arbitration
control allocation where the indirect yaw torque with existing stability functions on board, robustness of
AVEC’18

the control allocator especially with respect to the


prioritization between the direct and indirect yaw torque,
as well as between the drive and yaw torque request.

REFERENCES
[1] Blogsadmin (2015-04-21), “Advantages of Toyota
Dynamic Torque Control All-Wheel Drive -
Technology, Toyota RAV4”,
www.whiterivertoyota.com. Retrieved 2017-12-11.
[2] Tang, Y., “Control System For an All-wheel Drive
Electric Vehicle”, US Patent: 9527406 B2, 2009-01-
29.
[3] Barlage, J., “Compact Torque Vectoring Technology
Opens Up New Possibilities”, BorgWarner
Knowledge Library, 2016, pp. 1–3.
[4] Severinsson, L., Nilsson, K., “A Device for Torque
Vectoring”, European Patent: 2404081 B1, 2009-03-
05.
[5] Juenemann, P., Kramer, F. et al., “Coupling
Assembly and Driveline Assembly For a Motor
Vehicle with Multiple Driven Axles”, US Patent:
8535194 B2, 2008-08-15.
[6] Porsche 911 Carrera Models (2017-11-10), “Porsche
Torque Vectoring (PTV) and Porsche Torque
Vectoring Plus (PTV Plus)”, www.porsche.com.
Retrieved 2017-12-10.
[7] Gruber, P. et al., “Energy Efficient Torque Vectoring
Control”, Proc. of AVEC’16, 2016, Munich,
Germany.
[8] Sun, P. et al., “Analysis of Camber Control and
Torque Vectoring to Improve Vehicle Energy
Efficiency”, Proc. Of IAVSD’17, 2017,
Rockhampton, Australia.
[9] Novellis, L. De. et al., “Direct Yaw Moment Control
Actuated through Electric Drivetrains and Friction
Brakes: Theoretical Design and Experimental
Assessment”, Mechatronics, No. 26, 2015, pp. 1–15.
[10] Sawase, K. et al., “Left-Right Torque Vectoring
Technology as the Core of Super All Wheel Control
(S-AWC)”, Mitsubishi Motors Technical Review,
No. 18, 2006, pp. 16–23.

View publication stats

You might also like