INTRODUCTION
Steering system is an important control mechanism that human driver can be
used to change or maintain the heading of the vehicle. By manipulating the steering
wheel around its longitudinal axis, driving direction of vehicles can be controlled.
Generally, steering system has been divided into two categories: The mechanical
steering system and the power steering system. The mechanical steering system
bases on the physical energy of the driver and its manipulation inconvenient,
so, it has been replaced by the power steering system gradually. Electric Power
Steering system (EPS) is a class of power steering technology that can provide
steering torque by means of assistant motor. To make steering process easier
at low speed, and to make driver obtain a clearer road feeling at high speed,
EPS provides a feasible assistant torque according to driving speed. The EPS
has solved the contradiction between vehicle driving agility and the road feeling
effectively, so, it becomes focus of power steering technology (Ken
and William, 2000).
Assistant power characteristics and control strategy are critical steps of
the EPS design process. Repeated optimization is required and finalize the design
must be obtained by combining test data and correcting. This process not only
seriously affects the design efficiency, but also hinders the development of
system potential. If a measurement of simulation can be provided, the EPS can
then be reasonably assessed in low cost designed phase and the design efficiency
can be effectively improved as well. In addition, existing EPS evaluation is
mainly based on specific evaluation indicators, which can only examine its response
characteristics, or through some optimization design algorithms to optimize
the conflicting parameters (Parmar and Hung, 2004;
Chen et al., 2008). In nature, these approaches
attribute to openloop evaluation methods. Driver behavior, however, should
also be incorporated into the design process to perform a more comprehensive
performance evaluation.
With the development of research on driving behavior modeling, people have
proposed models such as the optimal curvature (Macadam,
2003), previewfollower model (Guo et al., 2004)
and other wellknown modeling theories. Some reports have also addressed through
parameter identification (Ungoren and Peng, 2004),
artificial neural networks (Plochl and Edelmann, 2007)
and other intelligent methods to study the driver modeling methods. However,
these studies are mainly for understanding and evaluating the characteristics
of the vehicle, the driver is generally idealized as an optimal controller to
generate steering commands, while the driver steering action are neglected.
That is, the steering commands are directly inputted into the vehicle dynamics
model. So, it can only be used to evaluate the manipulating stability of the
vehicle, but not to the assistant power performance of the EPS.
The purpose of this study is considering the dynamic interaction between the
driver and steering wheel, so as to establish a new driver steering control
model based on the torque feedback from the steering wheel. This model can expected
achieve a good path tracking performance and more important, it can work with
EPS integrated steering system when evaluating assistant power characteristics
in the fashion of closedloop simulation is needed.
MATERIALS AND METHODS
Vehicle model: Assuming that tires are working in the linear range,
the frontwheel steering angle is regarded as input, lateral speed, lateral
displacement, yaw angle and yaw rate are regarded as outputs, the linear model
is given by:
where, m is the vehicle mass and I_{z} is the yaw moment of inertia
around the vertical axis; v is the vehicle velocity; β and ω_{r}
are the sideslip angle and yaw rate of vehicle; δ is the steering angle
of frontwheel; c_{f} and c_{r} stand for the front and rear
axle cornering stiffness; l_{f }and l_{r} are the distance from
centre of gravity to the front and rear wheels, respectively.
From Eq. 1 and 2, the transfer function
from frontwheel angle to yaw rate can be calculated as follows:
Where:
Equation 3 describes the yaw rate response characteristics
with the input of the frontwheel steering angle.
EPS model: Typical structure of EPS is shown in Fig. 1,
including the steering wheel input, power motor and frontwheel angle output.

Fig. 1: 
Typical structure of the electric power steering system 
Through the dynamics analysis of the steering wheel and steering column, the
equations can be expressed by:
where, J_{s} is the moment of inertia and C_{s} is the viscous
damping coefficient; θ_{s} is the steering wheel angle; T_{h}
and T_{s} are the input torque and torque sensor output torque, respectively.
The symbol k_{s} is the stiffness coefficients; x_{r} and r_{p}
are the displacement of the gear rack and radius of gear, respectively.
The model of the motor can be written as:
where, J_{m} is the moment of inertia and C_{m} is the motor
viscous damping coefficient; θ_{m} is the motor rotation; T_{m}
and T_{a} are the motor electromagnetic torque and output torque, respectively.
The symbol k_{m} denotes the output shaft stiffness coefficient; k_{t}
and k_{f} are the electromagnetic torque coefficient and back EMF coefficient
of the motor, respectively. The symbol U is the motor control voltage; I is
the motor current; R is the motor armature resistance; L is the motor inductance
(the inductance value is small, generally given ignored); G_{1} is the
reduction ratio ofretarding mechanism.
Front angle output contains the output shaft and the steering rack two aspects,
the first based on the output shaft of the stress analysis, giving rise to the
following equation:
where, J_{e} denotes the moment of inertia, C_{e} the damping
coefficient, θ_{e} = x_{r}/r_{p} the steering pinion
angle, T_{l} the reverse acting output shaft torque.
According to the steering rack model, it can be expressed by:
where, m_{r} is the gear and rack equivalent mass; c_{r} is
the rack damping coefficient; k_{r} is the equivalent spring constant;
T_{r} is the resistance of the tire that road offer.
From Eq. 6 and 7, the model equation can
be obtained as:
where, M_{r} = m_{r}+J_{e}/r_{p}^{2}
denotes the equivalent mass of rack and pinion gear mechanism; C_{r}
= c_{r}+C_{e}/r_{p}^{2} the damping coefficient
of rack and pinion gear mechanism.
From Eq. 48, EPS model can be expressed
as the following statespace equation:
Where:
Driver model
Direction preview: Visual information is an important cue when the driver
controls the vehicle following a certain road. Experienced drivers are capable
of binding the road information and the vehicle state information to make an
expected steering angle. Meanwhile, due to the inherent physiological limitations
of driver, control instruction always has some delays before actually executed,
this preview kind of visual information can effectively compensate the driver’s
delay. As shown in Fig. 2, if one needs lane change, he will
shift his attention to the target track firstly. In this case, the preview direction
of the driver gaze and the heading of the vehicle form a deviation angle, referred
to herein as preview deviation angle θ. Studies have shown that, the driver
can gain the desired vehicle yaw rate according to the linear combination of
preview deviation angle and its rate of change (Shen et
al., 2012). Accordingly, this study defines yaw rate as:
where, ω_{d} denotes the desired yaw rate, θ the deviation
angle for the preview, α_{1}(α_{2}) the desired yaw
angle adjustment parameters.
Feedforward correction: Figure 3 shows the closedloop
system of driver and vehicle. Driver firstly obtains the desired yaw rate ω_{d}
through the preview and then passes it to the feedforward control model G_{d}
(s) to obtain the desired frontwheel steering angle δ_{d}. It
is noted that one cannot obtain the actually frontwheel steering angle δ^{*}
unless the driver latency delay has been considered. In this study, the θ^{τs}
item denotes the human nervous system response delay, while 1/(t_{d}s+1)
the inertia response lag of the arm (shen et al.,
2013).
Besides, experimental studies of McRuer et al.
(1977) and McRuer and Weir (1969) show that openloop
transfer characteristic of the drivervehicle closedloop system has a slope
about 20 dB dec^{1} near the crossover frequencies.

Fig. 2: 
Diagram of preview deviation angle θ 

Fig. 3: 
Structure of drivervehicle closed loop system 

Fig. 4: 
Scheme of the steering control model 
To meet this conclusion, the nerve latency influence should be taken into account,
which makes drivervehicle closedloop system open loop transfer function holds:
Aim to obtain G_{d}(s), Eq. 11 need to be do some
transforms and then developed with Taylor expansion, ignoring terms of fourth
and higher order, giving rise to the following equation:
where, ω_{c} denotes cutoff frequency of the closed loop system
and:
Feedback control: Expected steering wheel angle is not directly passed
to the front wheels, but applied torque instead to the steering wheel by driver’s
arm and then it will be force the steering column running gradually up to the
expected frontwheel angle. In order to modeling such relation, the transfer
delay of the driver neural system is indicated as τ and the inertia delay
of the driver arm is defined as t_{d.}. Then, the relationship between
the actual angle steering wheel and expected angle of the driver can be described
by Eq. 13:
where, δ_{d} and δ are the desired and actual angle of the
steering wheel, respectively.
In order to achieve the turning process described in Eq. 13,
drivers must applied proper torque continuingly to the steering wheel through
the arm muscles, until the completion of steering movements. In the steering
process, due to the contact of the tire with the road surface, a torque will
feedback to the driver, besides, this feedback relation change with the steering
states. Therefore, the actually procedure of driver adjusting the steering wheel
involved a dynamic force interaction. For the establishment of such a dynamic
interaction, a second order massspringdamper system can be used to describe
it, that means the movement rule of the steering wheel and torque interaction
satisfy the following equation:
where, m_{s} denotes the expectations mass, c_{s} (k_{s})
the damping (stiffness) coefficients in constrained driver interaction with
the steering wheel process, θ_{sr} desired steering wheel angle,
θ_{s} actual steering output angle, T_{h} the torque drivers
applied to the steering wheel, T_{r} the steering wheel feedback torque.
Take the Laplace transform on the Eq. 13 and 14,
obtained after proper deformation that:
Equation 15 indicates the feedback control strategy that
used by the driver when steering.
Based on the analysis above, the drivervehicle closedloop system can be built,
as shown in Fig. 4. Through preview the upcoming road, both
the direction deviation and its changing rate can be obtain, which driver can
be used to make decide the desired yaw rate ω_{d} and then, get
the desired steering wheel angle δ through feedforward correction, following
convert the angle signal into a steering torque signal T_{h} through
the arm system H (s), together with EPS achieve steering, output frontwheel
angle to the vehicle, so steering process can be achieved.
Simulation analyses: In order to assess the EPS assistant power characteristic,
choose the serpentine line conditions as driving manipulation condition.

Fig. 5(ab): 
Profile of the assistant characteristic curves with the input
of current I and torque T_{h}, (a)Weak assistant power and (b) Strong
assistant power 
Reference GB/T6232.194 serpentine curves trial design roads as the input
signal. Vehicle simulation parameters are m = 1715 kg, I_{z} = 2697
kgm^{2}, L = 2.54 m, l_{f} = 1.07 m, l_{r} = 1.47 m,
the angular steering transmission ratio G = 20, front and rear equivalent cornering
stiffness were c_{f} = 89733 N rad^{1}, c_{r} = 114100
N rad^{1}. Arm inertial delay is, according to previous studies, t_{d}
= 0.1 and set the speed as v = 50 km h^{1}.
To analyze the assistant effect, two different types of linear power characteristic
curve shown in Fig. 5 are designed (Shi
et al., 2007). The simulation results show in Fig.
6 and 7.
Figure 6 shows the results in the condition of the serpentine
line tracking. It can be seen from the figure that with or without assistant
in the case, drivervehicle closedloop system model built in this study can
complete steering tasks quite well, indicating that the driver model has a good
path tracking capability.

Fig. 6: 
Comparsion of simulation result of trajectory tracking with
different assistant power 

Fig. 7: 
Comparsion of simulation result of driver’s steering
control torque with different assistant power 
Figure 7 shows the condition when the steering torque is
applied to the steering wheel. Intuitively, as the EPS power increases, the
torque that the driver’s arms need to be applied to the steering wheel
decreases. In case of no power assistant, the maximum torque required by driver
is 6 N m. In the small power assistant case (Fig. 5a), the
maximum torque need to be applied reduces to 4.9 N m; while in the large power
assistant case (Fig. 5b) the driver only needs to apply 3.8
N m. It can be concluded that with the assistant of EPS, the driver just need
to apply relatively small torque to accomplish the same steering task, indicating
that the easiness and convenience performance are improved, which is designed
to be coincided with the goal, (Parmar and Hung, 2004;
Shi et al., 2007). In addition, other observation
in Fig. 6 shows that not bigger EPS power is always better.
Taking a further look at Fig. 7, when in the absence of steering
power assistant, the overshoot occurs and the delay is also present; in the
case of a small power assistant, the overshoot is reduced and the delay time
is also reduced. This indicates that with the assistant of EPS power, the driver
are free from heavy steering movements and can devote more time to focusing
on the task of steering control, so that the control performance have been improved.
However, in the case of large power assistant, although, a further delay time
is reduced, its tracking gain is also reduced and the driver steering torque
jitter occurs, which indicates that when EPS power increases, the road feeling
of the driver sensed would be reduced. Therefore, bigger EPS power is not always
better, one need to balance the performance between steering agility and road
feel, so as to optimization can be achieved. Simulation result shows that the
model in this study is valuable since there is no attempt has been conducted
for evaluate EPS with a driver model into considered.
CONCLUSION
A driver steering control model is established. This model has three components,
including direction preview, feedforward correction and feedback control. Because
the driver’s visual cues, physiological
characteristics and the torque interaction between driver and the steering wheel
has been considered comprehensively, so the steering behavior can be described
more reasonably. The simulation has been conducted and different EPS assistant
power characteristic demonstrated that although the linear assistant power steering
system can meet the performance of agility, too much assistant power will cause
the loss of information of road feel. To achieve better assistant performance,
there still needs further optimization design. For example, exploring nonlinear
assistant power strategy or developing adaptive control methods. In addition,
due to active intervene of the EPS assistant torque; the driver needs a process
of adaptation. Therefore, active interventions also need to be carefully processing
and combined with driver’s steer
characteristics to finish integrated design, so as to reflect the modernization
design of humancentered.
ACKNOWLEDGEMENT
This project was supported by the National Natural Science Foundation of China
under Grant No. 11202096.