Fan 1 Half-Car Model Based LQR Control of Active Suspension Xingchen Fan ME 131 Final Project Professor Hedrick Spring 2013 1. INTRODUCTION Suspension is one of the key connections between road and vehicle body. It not only supports the static weight of the vehicle, but also influences the comfort of passengers, the road holding abilities of tires and vehicle’ s handling characters. Several criteria are usually adopted to evaluate the performance of suspension, namely ride quality, suspension deflection (also called rattle space) and tire deflection. For passive suspension, when damping and stiffness are tuned to improve one performance, the other two are likely to be penalized. The inevitable tradeoffs in passive suspension design motivate the implementation of active suspensions, in which a controlled actuator is able to provide force in addition to that from conventional spring/damper unit. Since the evaluation of suspension performance consists of several criteria, it is reasonable to weight each criterion and put them in a performance index. Linear quadratic regulator (LQR) is a good choice for such optimization based control. LQR control has already been successfully implemented on quarter-car model (Rajamai). This
30
Embed
Half-Car Model Based LQR Control of Active …xingchenfan.weebly.com/uploads/2/4/9/9/24992009/xingchen...Fan 1 Half-Car Model Based LQR Control of Active Suspension Xingchen Fan ME
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Fan 1
Half-Car Model Based LQR Control of
Active Suspension
Xingchen Fan
ME 131 Final Project
Professor Hedrick
Spring 2013
1. INTRODUCTION
Suspension is one of the key connections between road and vehicle body. It not only
supports the static weight of the vehicle, but also influences the comfort of passengers, the
road holding abilities of tires and vehicle’s handling characters. Several criteria are usually
adopted to evaluate the performance of suspension, namely ride quality, suspension
deflection (also called rattle space) and tire deflection. For passive suspension, when damping
and stiffness are tuned to improve one performance, the other two are likely to be penalized.
The inevitable tradeoffs in passive suspension design motivate the implementation of active
suspensions, in which a controlled actuator is able to provide force in addition to that from
conventional spring/damper unit. Since the evaluation of suspension performance consists of
several criteria, it is reasonable to weight each criterion and put them in a performance index.
Linear quadratic regulator (LQR) is a good choice for such optimization based control. LQR
control has already been successfully implemented on quarter-car model (Rajamai). This
Fan 2
paper is dedicated to see how LQR control works on a more complicated half-car model with
pitch movement. Therefore, in addition to vertical acceleration, angular acceleration about
center of gravity (CG) is also used to assess ride quality. In fact, the half-car model based
LQR controller for active suspension also works for roll control; however, this paper only
focuses on pitch control.
2. HALF-CAR MODEL
A two-degree-of-freedom quarter-car model is usually used for LQR control on active
suspension (Figure 1). Sprung mass (𝑚𝑠) represents the vehicle body (quarter weight in this
case) and unsprung mass (𝑚𝑢) represents components including drive axle, suspension,
wheel and brake. Damper has damping coefficient 𝑏𝑠. Suspension and tire has stiffness 𝑘2
and 𝑘1 respectively. Tires usually have negligible damping; otherwise, a large amount of
heat will be generated. 𝐹𝑎 is the output from the active actuator.
Figure 1 Quarter-car model
Fan 3
In our analysis, we consider the most common conditions where wheels are always in
contact with ground. Thus, all displacements are measured from the static equilibrium
position, allowing gravity to be neglected.
The half-car model combines two quarter-car models with a shared sprung mass, where
a human model could be attached. The model has four degrees of freedom: sprung mass’s
vertical translation, its rotation about center of gravity, front and rear unsprung masses’
vertical translations. By adopting the same assumptions in quarter-car model, the
gravitational effect is neglected.
Figure 2 Half-car model
The equations of motion for half-car model are shown in Appendix I. Notice the positive
pitch is defined when the vehicle body rotates clockwise, or the front end of the vehicle
moves down.
Fan 4
With four degrees of freedom and their first time-derivatives, an eight-state linear
time-invariant (LTI) system is developed to represent the plant. The state-space representation
of half-car model is shown below:
�̇� = 𝐴𝑥 + 𝐵𝑢 + 𝐿𝑟
𝑦 = 𝐶𝑥 + 𝐷𝑟 (1)
Matrices A, B, L, C and D are included in Appendix I. In this case, x is an 8 × 1 state vector;
u is a 2 × 1 input vector including front and rear actuator forces (𝐹𝑎𝑓 and 𝐹𝑎𝑟); r is a 2 ×
1 disturbance vector containing road inputs to front and rear tires (𝑧𝑟𝑓 and 𝑧𝑟𝑟); y is the
output vector with all criteria, which will be discussed in the next section. For road inputs, the
vehicle is assumed to travel at constant speed, so the rear tire experiences the same road input
as the front tire does, but with a constant time delay.
3. LQR CONTROL WITH VARIOUS CRITERIA
The design of the controller is associated with the criteria in which we are interested. For
suspension, we are interested in:
i. Ride quality
Passengers feel uncomfortable when they experience large vertical accelerations.
Hence the acceleration of center of gravity ( �̈�𝑐𝑔 ) is one of the criteria.
Furthermore, with only �̈�𝑐𝑔, it’s possible that CG is at constant height, while
front and rear ends of the vehicle are oscillating. Therefore, the pitch acceleration
about the center of gravity (�̈�𝑐𝑔) is also taken into account. Penalties on �̈�𝑐𝑔 and
Fan 5
�̈�𝑐𝑔 are able to reduce the acceleration of every point on sprung mass, even
though the acceleration of specific passenger point is not considered in this paper.
Moreover, people feel uncomfortable when acceleration itself is changing fast; so
the first time-derivative of acceleration, jerk (𝑧𝑐𝑔 and 𝜃𝑐𝑔), can also be used as a
criterion. However, to simplify the system, it is not included in the performance
index. Jerk data is extracted from simulation results through numerical
differentiation of �̈�𝑐𝑔 and �̈�𝑐𝑔.
ii. Suspension deflection (rattle space)
Suspension deforms under load so it can absorb some energy from the road.
Suspension deflection (𝑧𝑠𝑓 − 𝑧𝑢𝑓 and 𝑧𝑠𝑟 − 𝑧𝑢𝑟 ) refers to the change in distance
between sprung and unsprung mass. How much suspension is allowed to deflect
depends on the limited space between wheels and vehicle body. If suspension
bottoms out, there will be a rigid connection between wheels and vehicle body.
Passengers will feel really uncomfortable. Small vehicles usually have a low ride
height, so their suspensions must deflect within a smaller range. Therefore,
suspension deflection should be held under control.
iii. Tire deflection
Tires are generally stiffer than suspension but they still deflect under load. Tire
deflection (𝑧𝑢𝑓 − 𝑧𝑟𝑓 and 𝑧𝑢𝑟 − 𝑧𝑟𝑟 ) is defined as the change in distance
between unsprung mass and road. Tires’ road holding ability is sensitive to
change in normal force and the shape of the contact patch. Excessive tire
Fan 6
deflection can cause tires to lose grip and result in slipping. Thus tire deflection
has to be minimized.
iv. Actuator force
Actuators only provide limited forces (𝐹𝑎𝑓 and 𝐹𝑎𝑟). Typical hydraulic actuators
in the market today are able to supply as much as 20,000 N. In addition,
compared to passive and semi-active suspensions, higher power consumption is
one of the drawbacks of active suspension. Therefore, it is necessary to constrain
the actuator output both to save power and to ensure that the desired force is
achievable.
The resultant performance index is defined as below:
𝐽 = ∫ [𝜌0�̈�𝑐𝑔2 + 𝜌1�̈�𝑐𝑔
2+ 𝜌2(𝑧𝑠𝑓 − 𝑧𝑢𝑓)
2+ 𝜌3(𝑧𝑠𝑟 − 𝑧𝑢𝑟)2
∞
0
+ 𝜌4�̇�𝑠𝑓2 + 𝜌5�̇�𝑠𝑟
2
+𝜌6(𝑧𝑢𝑓 − 𝑧𝑟𝑓)2
+ 𝜌7(𝑧𝑢𝑟 − 𝑧𝑟𝑟)2 + 𝜌8 �̇�𝑢𝑓2 + 𝜌8�̇�𝑢𝑟
2 + 𝜌10𝐹𝑎𝑓2 + 𝜌11𝐹𝑎𝑟
2]𝑑𝑡 (2)
It is the cost function for a quadratic programming problem, which should be minimized. For
weights in the performance index, only relative values matter. Therefore, the weight for z̈cg2
is assumed to be 1 whenever this term is penalized. All other weights are referenced to it. The
performance index is rearranged to the standard form as below:
𝐽 = ∫ [∞
0𝑥𝑇𝑄𝑥 + 2𝑥𝑇𝑁𝑢 + 𝑢𝑇𝑅𝑢]𝑑𝑡 (3)
where Q, N and R are shown in Appendix II. These three matrices are large and have
complicated entries. Hence instead of putting them down in state-space representation
explicitly, they are defined elementwise in MATLAB code.
It is shown that the solution to this optimization problem is feedback control: 𝑢 = −𝐺𝑥,
Fan 7
where 𝐺 = 𝑅−1(𝐵𝑇𝑃 + 𝑁), and P is the solution to algebraic Riccati equation (Rajamani):
Import parametersl_r = P.l_r;l_f = P.l - P.l_r;m = P.m; % Half car sprung mass [kg]I = P.I;m_uf = P.m_uf; % Quarter car unsprung mass [kg]m_ur = P.m_ur; % Quarter car unsprung mass [kg]k_sf = P.k_sf; % Spring constant of sprung mass [N/m]b_sf = P.b_sf; % Damping constant of sprung mass [N*s/m]k_tf = P.k_tf; % Spring constant of tire [N/m]k_sr = P.k_sr; % Spring constant of sprung mass [N/m]b_sr = P.b_sr; % Damping constant of sprung mass [N*s/m]k_tr = P.k_tr; % Spring constant of tire [N/m]
rho_0 = P.rho_0; % weight of z_cg_ddotrho_1 = P.rho_1; % weight of theta_cg_ddotrho_2 = P.rho_2; % weight of z_s_dotrho_3 = P.rho_3; % weight of z_u-z_rrho_4 = P.rho_4; % weight of z_u_dotrho_5 = P.rho_5; % weight of F_arho_6 = P.rho_6; % weight of z_s-z_urho_7 = P.rho_7; % weight of z_s_dotrho_8 = P.rho_8; % weight of z_u-z_rrho_9 = P.rho_9; % weight of z_s_dotrho_10 = P.rho_10; % weight of z_u-z_rrho_11 = P.rho_11; % weight of z_u-z_r
Input argument "P" is undefined.
Error in ==> generate_QRN at 3l_r = P.l_r;
Generate Q R N% R for z_cg_ddotR0 = [1/m^2, 1/m^2;... 1/m^2, 1/m^2];