Downloaded from SAE International by Univ of California Berkeley, Saturday, July 28, 2018

Published 03/28/2017
Copyright © 2017 SAE International
doi:10.4271/2017-01-1574

Longitudinal Vehicle Dynamics Modeling and Parameter Estimation for Plug-in

Hybrid Electric Vehicle
Sindhura Buggaveeti, Mohit Batra, John McPhee, and Nasser Azad
University of Waterloo

ABSTRACT
System identification is an important aspect in model-based control design which is proven to be a cost-effective and time saving
approach to improve the performance of hybrid electric vehicles (HEVs). This study focuses on modeling and parameter estimation of
the longitudinal vehicle dynamics for Toyota Prius Plug-in Hybrid (PHEV) with power-split architecture. This model is needed to
develop and evaluate various controllers, such as energy management system, adaptive cruise control, traction and driveline oscillation
control. Particular emphasis is given to the driveline oscillations caused due to low damping present in PHEVs by incorporating
flexibility in the half shaft and time lag in the tire model. Accurate and reliable vehicle dynamics parameters that control the vehicle
motion are estimated by acquiring experimental data from longitudinal maneuvers of the PHEV equipped with a vehicle measurement
system (VMS), global positioning system (GPS) and inertial measurement unit (IMU). The simulated model with estimated parameters
is analyzed for longitudinal dynamics by comparing with experimental data from on-road testing.

CITATION: Buggaveeti, S., Batra, M., McPhee, J., and Azad, N., "Longitudinal Vehicle Dynamics Modeling and Parameter Estimation for
Plug-in Hybrid Electric Vehicle," SAE Int. J. Veh. Dyn., Stab., and NVH 1(2):2017, doi:10.4271/2017-01-1574.

INTRODUCTION to be specified with accurate parameters to replicate the real vehicle

behavior, making system identification a crucial part in model based
Over the past few years, the automotive industry has witnessed a
control design.
major shift from conventional vehicles (CV) to plug-in hybrid electric
vehicles and battery electric vehicles (BEV) to address the growing
Advanced test facilities such as vehicle inertia measurement rigs,
concern for gasoline prices and carbon pollution. Although EVs have
wind tunnels, drum testing machines and 8-post rigs are capable of
demonstrated significant energy efficiency over CVs, limited driving
estimating location of vehicle center of gravity, drag coefficient, tire
range, higher upfront costs and limited availability of public charging
and suspension parameters respectively with high accuracy. For
stations made PHEVs a viable option over EVs for the short term.
example, Vehicle Inertia Measurement Facility (VIMF) [3],
PHEVs are equipped with complex power train architecture and use
commonly used by automotive manufacturers could measure center
advanced control strategies to improve their performance which led
of gravity height with an accuracy of 0.5%. However, utilizing these
the automotive industry to research solutions to reduce the
facilities is not always feasible for research purposes and requires
development procedure.
skilled technicians to conduct tests. Also, tire parameters that are
determined through tire test rig and adapted to real road conditions
Model-based control design is proven to be a time-saving and cost
requires relevant expertise of the user and compromise on the fact
effective approach that allows the designer to move directly from
that the results of drum testing can differ significantly from the
model creation to simulation, code generation and hardware-in-the-
behavior of real tire on the track.
loop (HIL) test in a systematic fashion. It enables the designer to
optimize and validate the controllers before deploying on hardware
Therefore, this paper focuses on identification of parameters of Prius
platform. Model based design of trip planning module, energy
PHEV through the data collected from real road driving maneuvers.
management system(EMS) and adaptive cruise control system whose
Data is acquired from a vehicle measurement system (VMS)
ultimate goal is to improve fuel efficiency of Prius PHEV requires a
developed by AnD Technology [4] that includes embedded
longitudinal dynamics model of the vehicle which is an important
controllers and high-accuracy sensors. It also comes with on-board
degree of freedom for assessing fuel consumption [1]. In addition,
filters that remove high frequency noise from the recorded data,
power-split HEVs are prone to lower damping in the driveline due to
thereby reducing the need to use additional filters during the
the absence of clutch or torque converter [2] making it necessary to
estimation process. This paper presents improved estimation methods
have a vehicle model that could be used in implementing and
over a similar work done on an EV[5].
evaluating driveline oscillations controller. The developed model has

289
 Downloaded from SAE International by Univ of California Berkeley, Saturday, July 28, 2018

290 Buggaveeti et al / SAE Int. J. Veh. Dyn., Stab., and NVH / Volume 1, Issue 2 (July 2017)

TEST VEHICLE
The test vehicle chosen is the Toyota Prius Plug-in Hybrid with front
engine, two axles and front wheel drive (FWD) layout. Detailed tests
were conducted on the Prius at Toyota Motor Manufacturing Canada
(TMMC) test tracks. Data was recorded by integrating the
measurements from a system of sensors using VMS manufactured by
AnD, GPS and IMU. Fig. 1 shows each of the four wheels fitted with
sensing equipment. The sensor set mounted on each of the wheels
collates required data to measure vehicle parameters from a group of
five sensors listed in Table 1.

While the GPS was used to record vehicle speed and acceleration, the
triggers for acceleration and braking are recorded in order to evaluate
the vehicle response during longitudinal maneuvers. The data Figure 2. System architecture for data collection
recorded by VMS, GPS and IMU was integrated with the help of an
integration device from ‘Vector Informatik GmbH’ so as to have a
common time stamp for the collected data. Few sets of measurements
Longitudinal Vehicle Dynamics Model
from VMS, GPS and IMU are compared to cross validate the Forces acting on the vehicle during longitudinal motion are shown in
correctness of data. The system architecture for integrating the three Fig. 3. The dynamic equation for the longitudinal motion of the
devices using the Vector is shown in Fig. 2. vehicle is:

(1)

where Fxf and Fxr are the longitudinal traction/braking forces acting
on the front and rear wheels, Frf and Frr are rolling resistance forces
on the front and rear wheels and Fd is the aerodynamic drag on the
vehicle acting at center of pressure (CP) located at a height of hd from
ground. Also Fzf is the sum of normal forces on the front left and right
wheels and Fzr is the sum of normal forces on the rear left and right
wheels. The location of CG from the front and rear wheels is lf and lr
Figure 1. Prius fitted with VMS respectively, the height of the CG is indicated by h, and mass of the
vehicle is represented by M.
Table 1. List of sensors in vehicle measurement system (VMS)

Figure 3. Forces acting in longitudinal direction on Prius PHEV

MapleSim Model
The vehicle dynamics model of Toyota Prius PHEV is developed in
Coast down tests, acceleration and braking tests that excite the
MapleSim due to its symbolic calculation and optimized code
vehicle in longitudinal direction, and speed bump tests that excite the
generation capability which makes the model very suitable for HIL
vehicle in vertical direction were done for the parameter estimation of
and model-in-the-loop (MIL) tests. The model consists of 18 degrees
the vehicle dynamics model.
of freedom (DOF) with chassis modeled as 6DOF rigid body, 4DOF
from suspensions, 4DOF from wheels and another 4DOF from
torsional deformation of the half shaft. Fig. 4 shows the three
dimensional representation of longitudinal vehicle dynamics model in
MapleSim.
 Downloaded from SAE International by Univ of California Berkeley, Saturday, July 28, 2018

Buggaveeti et al / SAE Int. J. Veh. Dyn., Stab., and NVH / Volume 1, Issue 2 (July 2017) 291

The model accepts motor torque as a model input and half-shaft

torque, vehicle speed, wheel speed and jerk are the model outputs. All
parameters identified based on experimental testing of Prius PHEV
are specified in Table 2.

PARAMETER ESTIMATION
MapleSim model should be supplied with accurate parameters for
vehicle dynamic simulation. A few parameters such as wheelbase,
front and rear track widths can be measured directly. However, these
are insufficient to describe the vehicle longitudinal motion. More
calculations are necessary to compute the vehicle parameters such as:

Frontal area
Rolling resistance coefficient
Figure 4. Chassis model of Prius PHEV in MapleSim
Center of gravity location
Table 2. Few parameters of MapleSim model: Prius PHEV Suspension stiffness and damping
Wheel inertia
Tire model coefficients
Half shaft stiffness

In this section, the methods used to calculate the vehicle parameters

are discussed. A MATLAB/Simulink based non-linear least squares
estimator [7] is chosen for parameter identification.

Frontal Area Estimation

Frontal area of a vehicle is defined by the Environmental Protection
Agency (EPA) as, “the area of the orthogonal projection of the
vehicle on a plane perpendicular to the longitudinal axis of vehicle”.
It is an important parameter used in the estimation of drag force
acting on a vehicle. At first, the frontal image of Prius PHEV was
processed to fill the region of the car with black and the back ground
with white color [8]. Then ‘im2bw’ function in MATLAB is used to
convert the grey scale image to binary image which outputs an image
as shown in Fig. 5 in which black pixels are replaced with the value 0
and white pixels are replaced by value 1. Percentage of area with
black pixels is determined and scaled back to the actual size of Prius
which gives the frontal area of the Prius as 2.19m2 which is 84.2% of
the total area of the box (2.60m2).

The model consists of a rigid body connected to tires through four

suspensions. One end of the half shaft is connected to the wheel while
the other end is connected to the car body through universal joints.
This configuration ensures that drive train components are mounted
on the chassis. Flexibility is introduced in the half shaft by modeling
it is as a torsional spring, damper and mass system. Pacejka tire
model is used in the vehicle dynamics model due to its advantages
over other tire models [6]. Suspensions are modeled as four vertical
springs in order to maintain the simplicity of the longitudinal vehicle Figure 5. Frontal image of Prius PHEV in B and W.
dynamics model.
 Downloaded from SAE International by Univ of California Berkeley, Saturday, July 28, 2018

292 Buggaveeti et al / SAE Int. J. Veh. Dyn., Stab., and NVH / Volume 1, Issue 2 (July 2017)

Estimation of Coefficient of Drag and Rolling Coefficient of drag of Prius PHEV 2015 was chosen to be 0.25[9].
Resistance The non-linear differential equation (3) is solved simultaneously
The aerodynamic force acting on a vehicle is given as: using Matlab’s ODE 45 routine as well as optimized for frr through
least squares minimization of velocity errors. Fig. 7 shows the curve
fit of simulated and experimental velocities in a particular run. It can
be seen from Fig. 8 that the average coefficient of rolling resistance
(2) (frr) is estimated as 0.014.

Where:

ρ is the density of air (1.225 kg/m3)

Cd is the coefficient of drag of car

A is the frontal area of car (2.19 m2)

V is the velocity of car

Vw is the velocity of head wind

Tire rolling resistance coefficient (frr) is obtained from the data

acquired during vehicle coast down testing done on a flat road.
During the test, the Prius is accelerated to a speed of 70 km/hr and
allowed to decelerate to a speed of 25 km/hr by road load forces
consisting of aerodynamic drag and rolling resistance. Tests were
conducted on a calm day to eliminate the effect of fluctuating wind Figure 7. Plot of rolling resistance coefficients (frr) for 6 different runs.
speeds. This procedure is repeated for 6 runs. As the throttle is
released during coasting while maintaining the vehicle in neutral, Fxf
and Fxr can be neglected.
Estimation of Location of Center of Gravity (CG)
Redistributed vertical tire loads during an acceleration/braking
The longitudinal dynamics equation of vehicle during coast down maneuver contain the major information necessary to determine the
testing is: height of the center of gravity. Hard acceleration and hard braking
tests are done so that the vehicle is excited only in longitudinal
direction thereby avoiding dangerous lateral, yaw or roll motions.
The vertical force on front wheels Fzf is expressed as:
(3)

(4) (5)

Where, center of gravity height (h), longitudinal location of center of

gravity from rear wheels (Lr) and pitch inertia (I) are the parameters
to be determined. Due to considerable amount of pitching during a
rapid acceleration and hard braking maneuver, longitudinal
acceleration at the center of gravity (aCG) of the vehicle is slightly
different from the longitudinal acceleration measured by GPS sensor
(aGPS) located at the top of the vehicle. Using acceleration
transformation equation for a rigid body and neglecting the terms
associated with roll and yaw motion, aCG is obtained as:

(6)

Figure 6. Plot of speed versus time to estimate frr

 Downloaded from SAE International by Univ of California Berkeley, Saturday, July 28, 2018

Buggaveeti et al / SAE Int. J. Veh. Dyn., Stab., and NVH / Volume 1, Issue 2 (July 2017) 293

Where, H is the height at which GPS is mounted from the ground. reason behind this assumption is that it best fits when front/rear load
Substituting equation (6) into (5), we obtain: transfer is taken into consideration, i.e. during acceleration or braking
situations. Also, suspension is assumed to be symmetric at the front
and rear. Thus, the stiffness and damping coefficients of suspension
are identified through vertical dynamic analysis of a four degree of
freedom half-car model as shown in Fig. 9. This model is represented
by unsprung masses of front and rear wheels mtf and mtr respectively
(7) and sprung mass of the vehicle body Ms. Is is the pitch inertia of
sprung mass. Suspension elements are characterized by stiffness
Where, pitch acceleration is obtained by taking the derivative of
coefficients Kf and Kr and damping coefficients Cf and Cr. Elasticity
pitch rate measured by GPS. Also, load on the front wheels (Fzf) is
in the tire is described by spring with stiffness coefficient Kt.
measured by WFS sensors (VMS system). Parameters h, Lr and I are
estimated by the non-linear least squares minimization of the
difference between experimental and simulated values of Fzf from
equation (7).

Figure 9. Half car suspension model

(8)

(9)
Figure 8. Plot of normal force on front wheels to estimate CG height (h)

Fig. 8 shows the experimental and simulated data for Fzf while
accelerating the car from 0 to 100 km/hr and braking from 100 to 0 (10)
km/hr. Table 3 shows the estimated values of center of gravity height
for different acceleration and braking maneuvers. Road grade angle
(θ) is neglected during estimation as the tests are done on a flat road.
Hence, the height and longitudinal location of center of gravity are (11)
taken as 0.61 m and 1.45 m for the simulation of MapleSim model.

Table 3. Estimated CG height for 3 different maneuvers (12)

During the test vehicle is driven over a speed dump of known

geometry in order to excite pitch and heave motions. z, ztf and ztr are
the vertical displacements of individual masses and θ is the pitch
angle of sprung mass. Finally, uf and ur are the road profile inputs to
the half car model.

Estimation of Suspension Stiffness and Damping

Prius PHEV 2015 consists of independent MacPherson strut
suspension in the front and torsion beam suspension in the rear.
However, suspension is modelled using vertical linear spring and (13)
damper elements at each wheel in the MapleSim model. The main
 Downloaded from SAE International by Univ of California Berkeley, Saturday, July 28, 2018

294 Buggaveeti et al / SAE Int. J. Veh. Dyn., Stab., and NVH / Volume 1, Issue 2 (July 2017)

Df,r is the distance recorded by LGS sensors at the front and rear Table 4. Estimated values of suspension parameters
wheels before encountering the speed bump of height A and width λ.
Normal load on the front and rear tires is given by:

(14)

(15)

The objective is to minimize the difference between experimental and

simulated normal loads with Kf, Kr, Cf, Cr and Kt as the parameters to
be estimated.

(16)

(17) Estimation of Wheel Inertia

Equation of motion representing the dynamics of front wheel rotation
of Prius is given as:
(18)

Where, , are the normal loads measured by WFS sensors on

(19)
front and rear tires. Objective function is minimized using non-linear
least squares technique with weighting factor (w) as 1. Fig.10 shows Where,
the experimental and simulated values of load on front and rear
wheels measured while vehicle is traveling over a speed bump of
height 0.07m and width 0.48m at a constant velocity of 15km/hr. (20)
Table 3 contains the estimated values of Kf, Kr, Cf, Cr and Kt.
Also the longitudinal equation of motion of the wheel is given as:

(21)

Where, Jw is the polar moment of inertia of tire-wheel assembly about

the wheel rotation axis, T is the torque applied at wheel-hub, Mrr is
the rolling resistance moment, Fa is the tire-road friction force, is
the longitudinal acceleration of the tire, rw is the dynamic tire radius
and mw is the mass of the tire-wheel assembly.

Figure 10. Plot of normal force on front wheels in a speed bump test

Figure 11. Forces and moments on tire during longitudinal motion

 Downloaded from SAE International by Univ of California Berkeley, Saturday, July 28, 2018

Buggaveeti et al / SAE Int. J. Veh. Dyn., Stab., and NVH / Volume 1, Issue 2 (July 2017) 295

From equations (19), (20) and (21) we get: ω is the angular velocity of the wheel, rw is the effective radius of the
wheel and vx is the vehicle speed.

(22) The tire parameters B, C, D and E in equation (23) are explained as

follows:
Using the data recorded by the VMS system for T, ω, Fx and rw,
equation (22) was used to identify the wheel inertia. Non-linear least B: Stiffness factor
squares minimization was done with Fx, rw, T, Fz, mw as inputs and ω
as system output resulting in a Jw of 1.65 kgm2. Fig. 12 shows the fit C: Shape factor
of estimated data against experimentally measured data.
D: Peak value

E: Curvature factor

BCD: The product of BCD represents the longitudinal stiffness of tire

The longitudinal force Fx is directly measured by the WFS sensor

(VMS system). Wheel speed (ωrw) and vehicle speed (vx) are
measured using laser ground sensors (LGS) and laser doppler
velocimeter (LDV) sensors (VMS system). Wheel slip s can then be
derived from wheel speed and vehicle speed as in equation (24).
Parameters B, C, D, E obtained by curve fitting the simulated values
from equation (23) and experimental data for normalized longitudinal
force as shown in Fig. 13 are listed in Table 4. Longitudinal stiffness
of tire is calculated as:

Figure 12. Non-linear least squares fit to estimate the rotary wheel inertia
(25)
A simple pendulum test has been conducted to verify the estimated
wheel inertia. For this, Prius tire was suspended from a pivot point by
a rope passing through the tire center. The tire was given small initial
angular displacement about the pivot point and natural frequency of
the tire was calculated by recording the time period of oscillation. A
shift in the axis from pivot point to center of gravity of tire was
accounted for in the final calculation of wheel inertia using parallel
axis theorem. Wheel inertia was obtained as 1.25 kgm2 by taking the
average of multiple readings.

Tire Model Parameter Estimation

The longitudinal force arising from the interaction of tire with road is
modeled by the Pacejka’s magic formula [10], an empirical equation
based on four fitting coefficients. Neglecting the lateral force
generated during vehicle longitudinal motion, Pacejka tire model uses
equation (23) to calculate the longitudinal force based on percentage Figure 13. Plot of normalized longitudinal force versus slip for tire
slip (s). Table 4. Parameters for Pacejka tire model

(23)

According to the definition of SAE J670, slip-ratio s is represented

as:

(24)
 Downloaded from SAE International by Univ of California Berkeley, Saturday, July 28, 2018

296 Buggaveeti et al / SAE Int. J. Veh. Dyn., Stab., and NVH / Volume 1, Issue 2 (July 2017)

Half Shaft Parameter Estimation

Stiffness of the half shaft is derived from its physical and material
properties. Fig. 14 shows the front left and right solid half shafts of
the Prius PHEV 2015 with torsional dampers. Both shafts made of
structural steel are of unequal lengths as it is a front engine, FWD car.

Torsional stiffness (torque required for unit twist) can be expressed

as:

(26)

Where: Figure 15. (a) Torque input measured experimentally. (b) Vehicle speed
measured experimentally vs simulated by model in response to the applied
G is the modulus of rigidity torque.

Jhs is the polar moment of inertia The simulation results show that vehicle speed of MapleSim model is
slightly higher than experimentally determined speed which may be
L is the length of shaft due to inaccuracies in the estimation of wheel inertia. Also,
aerodynamic resistance force acting on the vehicle during testing is
considered to be higher due to the presence of sensors and cables on
the exterior surface which could alter the drag coefficient of the
vehicle. Whereas, the simulated model is given a drag coefficient of
0.25 as the primary interest of this research is to have a validated
model without the effects of sensors. Further work will be conducted
to improve the estimation methods.

CONCLUSIONS
In this paper, a longitudinal vehicle dynamics model of Prius PHEV
Figure 14. Front right and left half shafts of Prius was developed as part of the model based development of various
controllers that aim at minimizing the fuel consumption and driveline
Front left half shaft was modeled as a stepped shaft. Equivalent oscillations. The developed model needs to be specified with accurate
stiffness and polar moment of inertia were calculated by measuring parameters. Compared to the previous methods of estimation that
dimensions across different sections. Torsional stiffness (khs) of right required standardized test facilities and extensive filtering, this paper
and left shafts were obtained as 16576 and 29292 Nm/rad indicates how parameter identification of vehicles can be effectively
respectively. undertaken using the GPS and a set of measurement sensors from
VMS. Simulation results show that the proposed estimation methods
MODEL VALIDATION can provide a sufficient precision of parameters.

For validating the Toyota Prius plug-in hybrid vehicle dynamics

model in MapleSim, the parameters identified from experimental data REFERENCES
were supplied to the model. Torque output of the planetary gear set 1. Taghavipour, A., “Real-Time Optimal Energy Management System for
while accelerating the car from 0 to 90 km/hr is given as an input to Plug-in Hybrid Electric Vehicles.” PhD Thesis, University of Waterloo,
2014.
the longitudinal vehicle dynamics model. Aerodynamic drag force 2. Syed, F.U., Kuang, M.L., Ying, H., “Active Damping Wheel-Torque
calculated from the feedback signal of vehicle speed during the Control System to Reduce Driveline Oscillations in Power-Split Hybrid
simulation is another input to the model. The vehicle speed output Electric Vehicle,” IEEE Trans. Veh. Technol. 2009, 58, 4769–4875.
3. Andretta, D., Heydinger, G., Bixel, R., and Coovert, D., “Inertia
from the simulated model is compared with the vehicle speed Measurements of Large Military Vehicles,” SAE Technical Paper 2001-
measured by GPS on the vehicle as shown in Fig. 15. 01-0792, 2001, doi:10.4271/2001-01-0792.
4. Batra, M., Azad, N.L., McPhee, J., “Parameter Identification for
Longitudinal Dynamics Model Based on Road Tests of an Electric
Vehicle,” ASME 2016 International Design Engineering Technical
Conferences and Computers and Information in Engineering
Conference, doi:10.1115/DETC2016–59732.
5. A&D Brochure for Vehicle Measurement System, https://www.
aanddtech.com/VMS.html, Date of Access: Aug.20, 2016.
 Downloaded from SAE International by Univ of California Berkeley, Saturday, July 28, 2018

Buggaveeti et al / SAE Int. J. Veh. Dyn., Stab., and NVH / Volume 1, Issue 2 (July 2017) 297

6. Pacejka, H., “Tire and Vehicle Dynamics.” Elsevier, 2005. CONTACT INFORMATION
7. Simulink Design Optimization Guide, MATLAB Documentation (Ver
2.8), Aug. 2016, http://www.mathworks.com/help/sldo/parameter- Sindhura Buggaveeti
estimation.html. Motion Research Group
8. Emam, M., “A New Empirical Formula for Calculating Vehicles Frontal
Area.” SAE Technical Paper 2011-01-0763, 2011, doi:1010.4271/2011-
University of Waterloo
01-0763. Ontario, Canada
9. Automobile Drag Coefficient, https://en.wikipedia.org/w/index. N2L 3G1
php?title=Automobile_drag_coefficient&oldid=759634980, Date of
Access: Jan.13, 2017. ACKNOWLEDGMENTS
10. Pacejka, H. B., and Bakker, E., 1992. “The Magic Formula
Tyre Model”. Vehicle System Dynamics, 21(s1):1–18, We would like to thank Toyota and Automotive Partnership Canada
doi:10.1080/00423119208969994. as sponsors of this research. A special thanks to our research engineer,
Chris Shum, to enable synchronous recording of signals from the
VMS, GPS and IMU. We are also grateful to Toyota Motor
Manufacturing Canada (TMMC) engineers, Matthew Van Gennip and
Bijan Sakhdari for their support during track testing.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or
otherwise, without the prior written permission of SAE International.

Positions and opinions advanced in this paper are those of the author(s) and not necessarily those of SAE International. The author is solely responsible for the content of the paper.