VHDL-AMS MODELING OF AN AUTOMOTIVE VIBRATION ISOLATION SEATING SYSTEM

Leran Wang and Tom J. Kazmierski School of Electronics and Computer Science University of Southampton Southampton SO17 1BJ, UK {lw04r,tjk}@ecs.soton.ac.uk ABSTRACT This paper presents VHDL-AMS model of an automotive vibration isolation seating system with an active electromechanical actuator. Five control algorithms for the actuator are implemented and their efciencies are investigated by subjecting the system to a number of stimuli, such as a single jolt or noisy harmonic excitations. Simulations were carried out using the SystemVision simulator and results are shown to compare the relative performance merits of the control methods. KEYWORDS VHDL-AMS, Automotive systems, Mixed-domain modeling, Control algorithms. Five control algorithms for the actuator are investigated here: Relative Control (RC) [8], Proportional-Integral Control (PIC) [9] , Fuzzy Logic Control (FLC) [10], Variable Structure Control (VSC) [11] and Optimization Control (OC) [12]. The model consists of three parts: the plant (i.e. the passenger seat and the vehicle chassis), actuator and control algorithm. The newly-developed IEEE VHDL 1076.1.1 standard packages [13] for multiple energy domain support are incorporated in the model.

1 Introduction
In automotive systems, the electronics with embedded software typically controls some mechanical components like airbags, anti-blocking systems for the brakes or active wheel suspensions [1]. These systems usually cover at least two physical domains, electrical and mechanical. Sometimes hydraulic, magnetic or thermal components are also included. This makes the design and test of the whole integrated system difcult to perform since typically each of these domains uses a different design language and simulation environment. The appearance of VHDL-AMS [2] has made it possible to incorporate components of different energy domains in a single model, because VHDL-AMS supports modeling of electrical and non-electrical systems at various abstraction levels [3]. Several sources have already recommended VHDL-AMS to be the unied modeling language for the automotive industry [4, 5]. However, its use in actual automobile design has so far been limited, partly because mature simulators supporting VHDL-AMS have only recently begun to appear. The VHDL-AMS model presented here is an automotive vibration isolation seating system with an active electro-mechanical actuator [6, 7] (see Fig.1). It reduces vibrations between the vehicle oor and the passenger seat so that the drivers ride quality could be improved. The actuator is a force generator that is acting between the seat and chassis. It is the main component of the system and its performance is much dependent on the control algorithm.

Figure 1. Vibration isolation seating system.

2 Chassis and Seating System

Fig.1 shows the model of the chassis and seating system. The passenger seat motion equation is written as: d2 xs Cs = dt2 Ms dxs dxc dt dt Ks (xs xc ) Ms Ap (P1 P2 ) Ms

(1)

Similarly, the chassis motion equation can be presented as: d2 xc dxs Cs dxc Ks = (xc xs ) 2 dt Mc dt dt Mc dxd Cc dxc Kc Ap (xc xd ) (P1 P2 ) (2) + Mc Mc dt dt Mc

The actuator force, F a , is expressed as Ap (P1 P2 ). The dynamic state and output vectors are selected as x = [xs , xc , dxs , dxc ]T and y = [xrel , Vrel ]T , where xrel = dt dt xs xc and Vrel = dxs dxc . Two sensors monitoring the dt dt seat and chassis generate input signals to the controller.

Newtons second law is applied to the input dynamics (J l1 ) as follows: d2 g1 dg1 + Tg1 = Tm + bl1 (6) 2 dt dt The gear train is considered ideal so that the relationship between the input and output angular velocity ( g1 and g2 ) and torque (T g1 and Tg2 ) can be written as: g2 rg1 ng1 = = and Tg1 g1 = Tg2 g2 (7) g1 rg2 ng2 Jl1 Application of Newtons second law to the load shaft (J l2 ) yields: d2 g2 dg2 + TL = Tg2 Jl2 + bl2 (8) dt2 dt The rack linear velocity (V r2 ) can be determined from: Vr2 = g2 rg2 and Vr2 = Vr1 (9)

3 Actuator Design
The actuator is an electro-mechanical hydraulic design, which operates in parallel with the passive springs and dampers. It consists of a DC motor, some mechanical parts (such as the gear train and rack) and a hydraulic vibration absorber (see Fig.2). The actuator input from the controller is a DC voltage (ea ), which drives the motor to output a rotational torque (T m ). The gear train converts the rotational torque into the translational velocity of the rack, whose displacement impacts the pressures of the upper and lower chambers of the hydraulic piston (P 1 and P2 , respectively). The vehicle oor is attached to the hydraulic cylinders piston rod and the seat sits on the cylinder cap. The actuator force is dependent on the pressure difference between the upper and lower chambers. The generated force attenuates the vibrations by acting on the vehicle oor and passenger seat.
Passenger seat Vs

The load torque (T L ) acting on the load shaft is: TL = Fr rg2 where Fr = Ar2 P2 Ar1 P1 + Mr ar Mr g (10) The hydraulic pressure in the upper and lower actuator chambers are described as: 1 dP1 = (+Ap Vrel Ar1 Vr1 ) dt U1 2 dP2 = (Ap Vrel + Ar2 Vr2 ) dt U2 (11) (12)

ia Ra ea La eb

DC motor

Vr

Ar1 P1, U1 Ap Vrel

g1 Tm Gear train Rack Ar2 P2, U2

The above differential algebraic equations (DAEs) describe the behavior of the actuator components. In VHDL-AMS, DAEs can be dened by simultaneous statements. The above equations cover electrical, mechanical and hydraulic domains.

4 Control Algorithms
Vc Vehicle chassis

Figure 2. Actuator model for the vibration isolation seating system.

The motor develops a torque proportional to the armature current (i a ): T m = K t ia (3)

Five control algorithms were implemented in VHDL-AMS to investigate the efciency of the actuator drive. They are briey described as follows. The inputs to the controller are the dynamic seat and chassis motions, i.e. the displacement, velocity and acceleration of the passenger seat (x s , vs and as ), the displacement and velocity of the vehicle chassis (xc and vc ). Different methods may use different numbers of inputs. The output is the voltage sent to the DC motor, ea .

4.1 Relative Control

The relative control strategy offers an elegant approach with minimal algorithm and hardware demands [8]. The equations for this algorithm are: if|as |<asmin ; 0 ea= sgn(as)(k[|as |asmin ]+eamin) if asmin<|as |<asmax ; sgn(as)eamax if|as |>asmax ; (13) where k = 20(eamax eamin )/(asmax asmin ).

When the armature rotates, the back EMF induces a voltage proportional to the angular velocity: eb = K b dg1 dt (4)

The circuit can be analyzed using Kirchoffs voltage law: La dia + Ra ia + eb = ea dt (5)

4.2 Proportional-Integral Control

The PI controller offers good stability and a low steadystate error [9]. The equation for this controller is: ea = K p e + K I e dt (14)

quadratic performance index [12]: J=

0

(xT F x + uT Ru)dt

(16)

where x and u = Kx are the state vector and feedback respectively. The feedback gain vector K is selected to minimize the performance index.

where e = as is the error signal, K p and KI are proportional and integral gains, respectively.

5 VHDL-AMS System

Implementation

of

the

4.3 Fuzzy Logic Control

Fuzzy logic controllers, due to their ability of handling uncertain and complex systems, are also used in the vibration isolation systems [10]. The inputs to the FLC are the velocity (vs ) and acceleration (a s ) of the seat. The fuzzy relationship of the input and output variables are represented by three linguistic variables: positive (P), zero (Z) and negative (N). With these linguistic variables, a set of nine fuzzy rules is developed, as shown in Table 1. Acceleration Z N Z P Fig.3 shows the system diagram of integrated chassis, seat, actuator and controller. The VHDL-AMS model contains three entities: the plant, actuator and controller. The quantities between the blocks show how different components from electrical, mechanical and hydraulic domains are connected.

Velocity

P Z N

P N N Z

N Z P P

Table 1. Fuzzy rules base.

The fuzzy inference procedure used is the maxproduct composition and the defuzzication method employed is the center of gravity approach. Figure 3. System diagram.

4.4 Variable Structure Control

Variable structure control uses a high-speed switching feedback strategy to realize robust control authority for uncertain plant descriptions and modeling errors [11]. A linear sliding surface (x) may be dened as (x) = Sx = 0 (15)

In the actuator part, the quantities in the upper row are through quantities and the ones in the lower row are across quantities. The VHDL-AMS code of the actuator is given below. It models the behavior of the actuator described above in Section 3.
library ieee 1076 1 1; use ieee 1076 1 1.mechanical systems.all; use ieee 1076 1 1.electrical systems.all; use ieee 1076 1 1.fluidic systems.all; use work.VibIso System.all; entity Actuator is port(quantity ea: in VOLTAGE; --applied armature voltage(V) quantity Vrel: in VELOCITY; --relative velocity of piston(m/s) quantity deltaP: out PRESSURE --pressure difference(N/m2)); end entity Actuator; architecture Behaviour of Actuator is quantity Tm: quantity ia: quantity eb: TORQUE;--motor torque(N*m) CURRENT;--armature current(amp) VOLTAGE;--back emf(V)

where S is the sliding surface coefcient vector and x is the state vector dened in Section 2. The switching control algorithm drives the plants state trajectory to a user selected sliding surface and then maintains the trajectory at that surface. The sliding surface is selected such that system motion, when restricted to this surface, exhibits the desired stability and tracking characteristics.

4.5 Optimization Control

The optimization-based controller features a full state feedback control algorithm that minimizes a specied linear

quantity omegag1,omegag2:

ANGULAR VELOCITY;

--angular velocity of gear(rad/s) quantity Tg1,Tg2: TORQUE; --torque applied on gear(N*m) quantity TL: TORQUE; --load torque back to shaft(N*m) quantity Vr: VELOCITY; --linear velocity of rack(m/s) quantity P1,P2: PRESSURE; --chamber pressure(N/m2) quantity U1,U2: VOLUME; --volume of chamber(m3) begin Tm == Kt*ia; eb == Kb*omegag1; ea == eb + Ra*ia + La*iaDOT; Jl1*omegag1DOT + bl1*omegag1 + Tg1 == Tm; omegag2 == N*omegag1; Tg1 == Tg2*N; Jl2*omegag2DOT + bl2*omegag2 + TL == Tg2; TL == (Ar2*P2 - Ar1*P1)*rg2; Vr == omegag2*rg2; P1DOT == (Ap*Vrel - Ar1*Vr)*(beta1/U1); P2DOT == (-Ap*Vrel + Ar2*Vr)*(beta2/U2); deltaP == P1 - P2; U1DOT == Ap*Vrel - Ar1*Vr; U2DOT == -Ap*Vrel + Ar2*Vr; end architecture Behaviour;

uate the peak-to-peak value of the displacement X pp from 10cm to 3.2cm, while all the active methods can achieve at least 50% improvement (i.e. X pp less than 1.6cm). Among these ve algorithms, the OC method gives the best result.

The actuator covers three different physical domains: electrical, mechanical and hydraulic. The VHDL-AMS code presented here incorporates components from these domains into one single model. Using the IEEE 1076.1.1 multiple energy domain standard packages [13], each quantity is dened by its physical name, such as torque, voltage, pressure, etc. These names are connected with their corresponding physical natures as shown in the code above.

Figure 4. Responses to a single jolt stimulus. Control algorithm Passive RC PIC FLC VSC OC P-to-P value(mm) 30.720 12.926 15.163 12.642 6.8608 5.7807 RMS(mm) 4.9304 3.3913 2.8669 2.0718 1.1236 0.96846

6 Simulation Results
Simulations were carried out using the SystemVision [14] VHDL-AMS simulator from Mentor Graphics. The system input dened in the testbench is the road displacement x d and the main response of the system is the seat displacement xs which displays the vibration isolation effect of the system. Two types of stimuli were applied to the model, a single sine wave jolt and multiple sine waves with a WGN (white Gaussian noise). In both cases, the sine wave has the amplitude of 5cm, and its wavelength is 240cm. If the vehicle maintains a constant forward velocity of 12.3m/s, the disturbance frequency is 5.1Hz. A single jolt Fig.4 shows the road displacement, the response from the passive system (i.e. without the actuator) and the responses of ve active control algorithms. Table 2 gives the peak-to-peak value and RMS (root mean square) of each response. The passive spring and damper can atten-

Table 2. Responses to a single jolt stimulus. Multiple sine waves with WGN The standard deviation of the added WGN is 1cm. The simulation results of passive and active systems are shown in Fig.5 and Table 3. The RC, PIC and FLC algorithms show relatively poor performance while the VSC and OC results are satisfactory. In both of the test cases, VSC and OC give superior performance to the other three, while FLC has better results than those of RC and PIC. The cause of the difference may be the number of inputs into the controller: RC and PIC only use as as input, FLC uses as and vs , VSC and OC use xs , vs , xc and vc . Basically speaking, the more information given to the controller, the better performance it generates. However, using more inputs to a controller means a more complex implementation.

References
[1] G. Pelz, J. Bielefeld, and G. Zimmer. Modeling of embedded software for automotive applications. In IEEE/VIUF Workshop on Behavioral Modeling and Simulation 1997, pages 123 128, 1997. [2] E. Christen and K. Bakalar. VHDL-AMSa hardware description language for analog and mixed-signal applications. In Circuits and Systems II: Analog and Digital Signal Processing, IEEE Transactions on, volume 46, pages 1263 1272, 1999. [3] Y. Herve. Virtual prototyping with VHDL-AMS. In Industrial Technology, 2003 IEEE International Conference on, volume 2, pages 761 765, 2003. [4] E. Moser. Automotive model exchange using VHDLAMS: a benchmark. In Computer-Aided Control System Design, 1996., Proceedings of the 1996 IEEE International Symposium on, pages 276 281, 1996. [5] E. Moser and N. Mittwollen. VHDL-AMS: the missing link in system design. experiments with unied modelling in automotive engineering. In Design, Automation and Test in Europe, 1998., Proceedings, pages 59 63, 1998. [6] I.M. Ibrahim and S.M. El-Demerdash. Investigation of wheeled tractors ride comfort using hydraulic semiactive suspension system. SAE Paper, NO. 1999-013727, 1999. [7] X. Liu and J. Wagner. Design of a vibration isolation actuator for automotive seating systems. Int. J. of Vehicle Design, 29(4):335 375, 2002. [8] M.R. Jolly and L.R. Miller. The control of semi-active dampers using relative feedback signals. SAE Paper, NO. 892483, 1989. [9] Y.M. Sam, J.H.S. Osman, and R.A. Ghani. Proportional-integral sliding mode control of a quarter car active suspension. In TENCON 02. Proceedings. 2002 IEEE Region 10 Conference on Computers, Communications, Control and Power Engineering, volume 3, pages 1630 1633, 2002. [10] A.J. Barr and J.I. Ray. Control of an active suspension using fuzzy logic. In Fuzzy Systems, 1996. Proceedings of the Fifth IEEE International Conference on, volume 1, pages 42 48, 1996. [11] R.A. DeCarlo, S.H. Zak, and G.P. Matthews. Variable structure control of nonlinear multivariable systems: A turtorial. Proceedings of the IEEE, 76(3):212 232, 1988. [12] D. Hrovat, D.L. Marglois, and M. Hubbard. Approach toward the optimal semi-active suspension. Journal of dynamic systems, Measurement and control, 110(3):288 296, 1988. [13] P.R. Wilson, A.D. Brown, and H.A. Mantooth. Standard vhdl 1076.1.1 packages for multiple energy domain support. In Behavioral Modeling and Simulation, 2003. Proceedings of the 2003 International Workshop on, pages 70 75, 2003. [14] Mentor Graphics Corporation. SystemVision Users Manual. Version 3.2, Release 2004.3. July 2004.

Figure 5. Responses to a noisy sine wave stimulus. Control algorithm Passive RC PIC FLC VSC OC P-to-P value(mm) 26.500 14.987 13.702 15.688 7.3798 6.0416 RMS(mm) 6.5085 5.1171 4.2012 3.9063 1.8710 1.5655

Table 3. Responses to a noisy sine wave stimulus. Relative complexity of the control algorithms can be compared by their simulation CPU times. To simulate a 1 second run of the system, the corresponding CPU times on a Pentium 4 PC for these methods were: RC2.016s, PIC1.609s, FLC2.547s, VSC3.531s, OC1.891s.

7 Conclusion
An efcient VHDL-AMS model of a vibration isolation seating system with an active electro-mechanical actuator has been developed and investigated. Several control algorithms were implemented and tested with two types of stimuli, a single jolt and a harmonic wave with WGN. Simulation results show the superior performance of the Variable Structure and Optimization based algorithms compared with the Relative, Proportional-Integral and Fuzzy Logic ones. Among the ve methods, VSC is the most complex and PIC is the simplest, judged by the CPU time.