A

nttacsppa©

K

1

tevtp[r

ot

CT

1d

Medical Engineering & Physics 29 (2007) 367–374

Hardware-in-the-loop-simulation of the cardiovascular system,with assist device testing application

B.M. Hanson a,∗, M.C. Levesley a, K. Watterson b, P.G. Walker a

a School of Mechanical Engineering, University of Leeds, Leeds LS2 9JT, UKb Yorkshire Heart Centre, Leeds General Infirmary, Leeds, UK

Received 29 November 2005; received in revised form 13 March 2006; accepted 9 May 2006

bstract

This paper presents a technique for evaluating the performance of biomedical devices by combining physical (mechanical) testing with aumerical, computerised model of a biological system. This technique is developed for evaluation of a cardiac assist device prior to in vivorials. This device will wrap around a failing heart and provide physical beating assistance (dynamic cardiac compression). In vitro, the deviceo be tested is placed around a simulator comprising a mechanical simulation of the beating ventricles. This hardware model interfaces withcomputerised (software) model of the cardiovascular system. In real time the software model calculates the effect of the assistance on the

ardiovascular system and controls the beating motion of the hardware heart simulator appropriately. The software model of the cardiovascularystem can represent ventricles in various stages of heart failure, and/or hardened or congested blood vessels as required. The software displays

hysiological traces showing the cardiac output, depending on the natural function of the modelled heart together with the physical assistower provided. This system was used to evaluate the effectiveness of control techniques applied to the assist device. Experimental resultsre presented showing the efficacy of prototype assist on healthy and weakened hearts, and the effect of asynchronous assist.

2006 IPEM. Published by Elsevier Ltd. All rights reserved.

Simulat

suwtiwP

s

eywords: Hardware-in-the-loop (HIL); Cardiac assist device; Modelling;

. Introduction

Cardiac assist devices are currently being developed withhe aim of providing physical pumping assistance to a weak-ned or failing heart. Implantable impeller pump-based leftentricular assist devices (LVADs) are emerging. Alterna-ively, dynamic cardiac compression (DCC) can assist byroviding compression to the surface of the ventricle(s)1]—thereby avoiding some problems of immune-system

ejection and thromboses [2].

In the early development of LVADs, numerical simulationsf circulatory systems have been valuable tools when usedo simulate the effect of assist devices on the cardiovascular

∗ Correspondence to: Department of Mechanical Engineering, Universityollege London, Torrington Place, London WC1E 7JE, UK.el.: +44 7879 415 504.

E-mail address: ben@benhanson.com (B.M. Hanson).

[lLtc

aiat

350-4533/$ – see front matter © 2006 IPEM. Published by Elsevier Ltd. All rightsoi:10.1016/j.medengphy.2006.05.010

ion; Cardiovascular system; LVAD

ystem (CVS) [3,4]. These models have a long history ofse and some are highly detailed (e.g. [5]). However, whenorking prototypes have been constructed purely numerical

echniques become less attractive; it can be inconvenient andnaccurate to create numerical models of prototype devices,hose physical behaviour may not be fully understood yet.hysical testing is therefore required.

The actual hydraulic performance of prototype LVADystems has been tested on electro-hydraulic servo-systems6–8]. Investigators have used simple models of the circu-atory system to present a realistic hydraulic load to theVAD, however these testing models have not shown howhe mechanical performance of an LVAD directly affects theomplete circulatory system.

With a DCC assist device, the interaction between the

ssist device and the surface of the heart is crucial. Thisnteraction is likely to depend on physical features whichre particularly difficult to model, such as non-linear fric-ion and backlash. Physical testing of DCC devices requires

reserved.

3 ineerin

aaIbv(taukcma

2

(Hhhts

ipuuo

••

ueb

•

••

•

3

fthrla

68 B.M. Hanson et al. / Medical Eng

physical heart (or model) on which to apply compression,nd a means of measuring the compressive effort applied.n vitro studies on dead hearts are unfortunately not feasi-le due to non-function of papillary muscles, collapse of theentricular outflow tracts and increased myocardial stiffnessunpublished results). However, excised hearts have been sus-ained for in vitro tests using a blood supply from a “support”nimal [1,9,10]. In these studies, a hydraulic servo pump wassed to present a realistic outflow impedance using a wind-essel model, as with the LVAD tests. Again, the effects on thelosed-loop circulatory system were not studied. These ani-al studies have produced invaluable results, however they

re costly in terms of time, resources and animal lives.

. Hardware-in-the-loop concept

This paper describes the use of hardware-in-the-loopHIL) simulation to test a DCC assist device in vitro. ThisIL simulation combines a realistic numerical model of theeart and cardiovascular system with a controllable physicaleart simulator, and this interacts in real time with a proto-ype DCC assist device. The nature of these interactions ishown in Fig. 1.

Hardware-in-the-loop-simulation has been developing inndustrial control for testing of systems comprising somehysical and some simulated components [11–13]. Sim-lation is used to represent processes that are physicallynavailable, or whose use would be too costly, dangerous,r time-consuming. Proven benefits of HIL include:

Reproducibility of experiments.The ability to perform tests which would otherwise beimpossible, impractical or unsafe:

ftti

Fig. 1. Structure of the hardware-in-the-loop-si

g & Physics 29 (2007) 367–374

◦ testing a component under extreme or dangerous oper-ating or environmental conditions (e.g. extremes of tem-perature, pressure, acceleration);

◦ testing effects of sensor and/or actuator faults;◦ long-term durability testing—until failure.

In this investigation, the cardiovascular system is the sim-lated component, and the benefits described above applyqually to the biomedical field. Using HIL simulation foriological systems could provide:

The possibility to test on a wide range of simulated patientgeometries and pathologies.Repeated testing on a consistent model.Facilitated numerical quantification of performance byrecording simulated physiological parameters.Replacement of human and/or animal subjects:◦ sterile, clinical environment is not required;◦ ethical issues are removed;◦ cost and development time are reduced.

. Hardware-in-the-loop simulation

Fig. 1 shows the overall structure of the HIL simulationor the assist device application. The system involves a posi-ion control loop (indicated) whereby the diameter of theeart simulator is controlled by computer, such that it is aeal-time physical “display” of the diameter of the simu-ated heart. The assist device contracts around this simulatornd a sensor records the assist force at the physical inter-

ace between assist device and simulator. The HIL aspect ishat this physical force signal forms part of the control loop:he force signal is fed into the CVS model, which calculatests effect on pressure within the heart and therefore blood

mulation of the cardiovascular system.

ineerin

flstaorsvF

3

mslmDshodi

sihcebt

m

P

Fet

wrip

dtttdaa

σ

mhbs

pnpeiNgmpf

B.M. Hanson et al. / Medical Eng

ow into and out of the heart, therefore the diameter of theimulated heart. Thus, the motion of the hardware simula-or depends on the effect that hardware interaction with thessist device has on the software CVS model. From the pointf view of the assist band, the heart’s physical motion andesponse to assistance appear realistic, as governed by theoftware CVS model. The components of the HIL cardio-ascular simulation will now be described, with reference toig. 1.

.1. Numerical model of cardiovascular system

The human cardiovascular system has been modelledany times at various levels of complexity (for a review

ee, e.g. [14], and the state-of-the-art [15,16]). This particu-ar model is based on some elements of previously-reported

odels, selected as appropriate to the requirements of theCC assist device application. The structure of the model is

hown in Fig. 2; it is an important feature that the model isaemodynamically closed-loop in order to assess the effectsf applying assistance. The function of the software model isescribed in detail in [17], however an overview of the models provided here.

In a HIL simulation, the interface between hardware andoftware is crucial—in this case that is the heart. The models therefore biased with more detail being used to describe theeart than the rest of the circulatory system. The four hearthambers are modelled separately, allowing assessment of theffect of disease or incompetence in any or all of the cham-ers. Two further passive, compliant compartments representhe aorta and pulmonary artery.

The equation relating flow and pressure in each compart-ent is of the form:

x = Zxφx + Px additional (1)

ig. 2. Representation of the circulatory system model using an electricalquivalent, indicating six compartments in which blood can be stored (ver-ical branches).

act

t

Fp

g & Physics 29 (2007) 367–374 369

here Padditional is applied to the heart chambers only andepresents the sum of pressures generated by passive stretch-ng of the pericardium, natural systolic function, and assistressure, where appropriate.

For the atria we apply a pressure–time curve; this is notependent on atrial volume. For the ventricles we use a func-ion, f(t), to generate a time-varying myocardial wall stress;he instantaneous active systolic pressure within the ven-ricles is then calculated from the wall stress and ventricleimensions. This stress, σ, depends on the volume, V, to giverepresentation of the Frank–Starling relationship, and is

lso rate-dependent [5]:

(t) = σmaxf (t)KV (V (t) − V0)

[1 − DV

(dV

dt

)](2)

Since research has concentrated on modelling the primaryechanical effects of assistance, the model does not include

ormonal effects, vasomotor control, orthostatic stress orreathing, although these could of course be added at a latertage.

The model has been implemented using LabVIEWTM toroduce not only the software/hardware interface but also theumerical circulatory model itself, in LabVIEWTM’s visualrogramming language. The use of a graphical programmingnvironment makes it easy to incorporate physiological datanto the model, and to manipulate it into a suitable form.on-linear circulatory elements and functions are shownraphically and can therefore be consulted, verified, andanipulated more easily than tabulated data. As an exam-

le Fig. 3 shows the shape of a typical activation functionor myocardial stress generation. This is scaled in the Xnd Y directions and used in the heart chamber models to

reate the function f(t) which is a stress, Y, acting over aime, X.

The graphical code is compiled efficiently to take advan-age of the hardware computation processes available on a

ig. 3. A normalised activation function curve: used to generate myocardialressures within the CVS model.

3 ineerin

ta

3

imh

•

••

•

totr

iebcbehfit

c

F

a

p

s

ufitc

3

svwF

umappbcco[

3

ts

70 B.M. Hanson et al. / Medical Eng

ypical modern processor, thus in practice may be computeds fast as code written in, e.g. C.

.2. Heart simulator

An interface is required to communicate between the phys-cal (hardware) assist device and the software cardiovascularodel. The requirements of the physical simulator of the

eart for this assist device application are:

it must be possible to wrap a contractile band-type heartassist device around the simulator, andsense the assist pressure produced,the device must represent the motion of one “slice” throughthe ventricles—the volume of the heart encircled by oneassist belt,the device must be able to simulate normal and patholog-ical heart motion at rates of up to 150 bpm.

In use, when combined with the software model and con-roller, the device is required to “beat” in a real-time displayf the changing volume of the ventricles. The combined sys-em must respond to assistance compression in a physicallyealistic manner.

This interface takes the form of a heart simulator, asllustrated in Fig. 4. The simulator was constructed usinglectromagnetic swing-arm actuators that can be controlledy computer easily and accurately. These are arranged in aircular array as shown in Fig. 4—the swing arms are shown,ut the actuation method is omitted for clarity. On the end ofach of the six arms is mounted a vertical post; these form aexagon around which to wrap the assist device. With the con-

guration used, external diameters from 24 to 96 mm (vertex

o vertex) can be simulated.Though independently actuated, the six actuators are

urrently all linked with coupling rods, giving the sim-

ig. 4. Plan view diagram of the heart simulator, constructed from six swing-

rm actuators. Key: , 6× pivot points of swing-arm actuators; , 6×osts, around which is wrapped; , flexible belt of assist device; , force

ensor; , monitor unit of assist device.

ft5Tv(tsw

P

wfrat

ooul

fi

g & Physics 29 (2007) 367–374

lator just one degree of freedom: the diameter. There-ore, it is just the gross change in ventricle volume thats represented; this is sufficient since the assist band to beested also has just one degree of freedom (circumferentialontraction).

.3. Assist device

The prototype DCC assist device being tested consists ofeveral contractile belts which are to be placed around bothentricles of the heart to form a contractile blanket. One bandas tested in isolation on the heart simulator as indicated inig. 4.

The circumference of these flexible belts is controlledsing direct-drive, miniature dc motors. Compared to pneu-atic actuation used in other DCC devices, the torque

nd position of these motors are easily controlled by com-uter, and are suitable for use with an implantable batteryower supply. Long-term device life is being determinedy endurance trials. If necessary, brushless commutationould be used to increase motor life; brushless motors areurrently in use within implanted LVADs. Further detailsf the form and control of the device can be found in18].

.4. Interfacing between hardware and software

The key factor in an HIL simulation is the interfacing;his must be designed to suit both the hardware and softwareystems.

A custom-made force sensor was used to record theorce produced by the assist device. This comprises ahin aluminium cantilever beam structure (dimensions:.5 mm × 12 mm) with strain gauges on both beam surfaces.he force signal from this sensor is read into the softwareia analogue-to-digital conversion using an interfacing cardNational Instruments PCI-MIO-16E). The increase in ven-ricular pressure created by this force was calculated using aimple model, Eq. (3), and added to the ventricular pressuresithin the software model.

assist = T

rh(3)

here T is the circumferential tension in the assist belt (=orce recorded, with the current geometry), r the externaladius of the ventricles, and h is the effective width of thessist band. The same assist pressure was added to both ven-ricles.

A servo-potentiometer is used to measure the positionf one swing-arm actuator, and from this, the diameterf the heart simulator is calculated. This is the feedbacksed to control the instantaneous diameter of the simu-

ator.

The model is paced using an ex-planted pacemaker inter-aced to the computerised model. This allows synchronousn vitro/in vivo comparison to be performed in future, where

B.M. Hanson et al. / Medical Engineerin

F1

atmh

3

valwaairinri

ptutwaft

4

eta

oos

acdsttuit

sddcoppsc

icpmcmeas

ionicpOoab

5

tw

ig. 5. Position tracking performance of the heart simulator at 50, 100, and50 beats per minute (bpm).

pacemaker would relay real physiological pacing signals tohe computer model. For the data presented herein, the pace-

aker was set to a constant rate and the model could equallyave been paced numerically.

.5. Position control loop

The physical simulator’s task is to display the exact time-arying dimensions of the modelled “slice” through the heart,s shown in Fig. 4. The instantaneous position of the simu-ator is controlled in a feedback loop operating concurrentlyith the circulatory system model—data is passed along each

rrow in Fig. 1 at a loop rate of 500 Hz. This rate provideshigh resolution of the CVS simulation that allows detailed

dentification of the effects that assistance might have. A fastate is also desirable to reduce the delay associated with dig-tal filters that are used to remove high-frequency electricaloise from analogue input signals. The upper limit on loopate is in practice governed by the time required for analoguenterfacing rather than model computation.

A non-linear PID control algorithm is used for feedbackosition control of the hexagonal array of swing-arm actua-ors. This gives good positional accuracy in the presence ofnpredictable disturbance forces from the assist device. Theracking performance of the heart simulator is shown in Fig. 5,here the simulator replicated the motion of a heart beating

t 50, 100, and 150 bpm. The actual position of the simulatorollowed the desired position to within 0.5 mm diameter, andhis was deemed sufficient accuracy for the application.

. Experimental methods

The HIL simulator allows quantification of the circulatoryffects of real, physical assist. In this paper we present resultshat demonstrate the efficacy of the HIL testing environment,nd its specific benefits.

cth

g & Physics 29 (2007) 367–374 371

The HIL simulator was used to evaluate the performancef a prototype assist band, demonstrating the direct effectf its assistance on a model of a weakened cardiovascularystem.

A study of assist synchronisation was performed tottempt to determine the effect of assistance that is poorly syn-hronised with the heart’s own efforts: when used in vivo, theevice will use the heart’s natural pacing signal, if available,ensed from a pacemaker. This will be exhibit beat-to-beatiming variations, and without careful control it is possiblehat an assist device may lose synchronisation with the nat-ral heart. To assess this effect, a delay of up to 200 ms wasmposed between the pacing signal at the start of natural sys-ole and the onset of assist compression.

The assist band was mounted to the heart simulator, ashown in Fig. 4. A weakened heart model was used, asescribed below. With the model operating in a steady haemo-ynamic state, the assist device was switched on, applyingompression every beat. The CVS reached a new steadyperating state after approximately 8–10 beats. Simulatedhysiological traces were recorded from the model over thiseriod; the pressure within ventricles and main arteries wastudied in each case, along with the ventricle volumes andardiac output.

The energy efficiency of the assist device was measured:nstantaneous electrical power in to the assist device was cal-ulated as the product of voltage and current, and mechanicalower applied to the simulated ventricle was measured byultiplying the applied belt tension by the rate of change of

ircumference. The energy efficiency of the assist device waseasured in each delayed case as the ratio of total mechanical

nergy (out) to electrical energy (in). This was averaged overperiod of three cardiac cycles, once the CVS had reached a

teady state.To simulate acute ischaemic heart disease, the contractil-

ty of both left and right ventricles was scaled down to 50%f their nominal healthy values [4]. Although autonomouservous system (ANS) control of peripheral resistance is notncluded in this current model, the model’s values of vas-ular resistances were increased manually to maintain bloodressure in the weakened condition (values in Appendix A).ther parameters, including heart rate, were unchanged; usef a software model ensures that the experimental conditionsre identical for each repeated test—something that woulde impossible on a biological model.

. Results

Fig. 6 shows some traces from the numerical CVS simula-ion while undergoing testing. The assistance in this exampleas synchronised with the natural systolic effort.

The closed-loop CVS model has shown that when one

ompression band is applied around both ventricles, assis-ance affects the systemic (left heart) and pulmonary (righteart) circulation in different ways. These effects are dis-

372 B.M. Hanson et al. / Medical Engineering & Physics 29 (2007) 367–374

F hown, ad

caparmdar

bitooa

tehacd

c

opdotabfcrincrease the risk of further ischaemia.

The efficiency of the device in converting electrical powerinto mechanical power is indicated by the relative magnitudes

Table 1Effects of prototype assist device with delayed onset on HIL simulation ofthe CVS

Condition Assist Delay(ms)

L.V.(E.D.V.)

C.O.(l/min)

Assistefficiency (%)

Healthy None 140 5.02Weak None 142 2.45Weak Assisted 0 158 3.08 9.0

ig. 6. Simulated physiological traces from HIL simulation. Healthy CVS sevice: LV, left ventricle; RV, right ventricle; PA, pulmonary artery.

ussed in more detail in [17]. The model suggests that thessistance acts to empty the right ventricle, and the increasedulmonary pressure would then tend to increase the oper-ting volume of the left ventricle. These effects would beeduced in vivo by the body’s ANS applying compensatoryechanisms. Therefore, this model is valuable in showing the

irect, mechanical effects, as this allows the development ofcontrol scheme to maintain both ventricle volumes without

elying on ANS control.Assistance is seen to immediately increase the modelled

lood pressure within ventricles and major arteries. Again,n vivo, ANS control would act to reduce vascular resistanceo decrease the aortic blood pressure and increase the cardiacutput. This model only shows the direct, mechanical effectsf compression, which is sufficient for evaluating prototypessist devices.

The increase in cardiac output (C.O.) as a result of assis-ance is shown in Table 1. The C.O. of the simulated weak-ned state is dramatically reduced in comparison to theealthy state, however the C.O. was then increased withssistance. The beneficial increase in C.O. is highest for syn-

hronous assist, however a delay of up to 50 ms in assist actionid not indicate a significant effect on performance.

Fig. 7 shows a comparison between synchronous and asyn-hronous assist. In Fig. 7(a), assist is applied at beginning

WWWW

nd weakened CVS with effect of direct compression from a prototype assist

f systole, and a positive assist force is recorded over theeriod of ventricle contraction. In Fig. 7(b), the assist iselayed (by 150 ms), and a force is only recorded over partf the contraction period. The assist force continues intohe isovolumetric relaxation period, which is extended as

result—diastole only begins once the assist pressure haseen removed. Fig. 7(b) also shows a significant peak inorce during the isovolumetric period, which could be clini-ally important—the increased contact force in diastole couldestrict blood flow over the surface of the myocardium and

eak Assisted 50 158 3.08 9.0eak Assisted 100 154 3.01 6.7eak Assisted 150 152 2.87 4.0eak Assisted 200 144 2.59 1.5

B.M. Hanson et al. / Medical Engineering & Physics 29 (2007) 367–374 373

with ve

odimtae

vmtTa

aaPcmIc

6

msw

dtiesaa

u

peofac

fsnfipMaac

–htTipdr

techIunse

Fig. 7. Force and power traces recorded from heart simulator, shown

f the electrical and mechanical traces in Fig. 7. With noelay, the efficiency recorded was approximately 9%. Thiss below the theoretical maximum efficiency of dc electric

otors (up to 70%), however that occurs at much higher rota-ional speeds than used in this application. Friction in the beltnd the motor’s pulley system will have reduced the potentialfficiency of the device.

The electrical power used by the device does not changeery significantly between Fig. 7(a and b), however the usefulechanical power out from the device is reduced if the assis-

ance is applied to ventricles that have finished contracting.able 1 shows how the efficiency reduces with increasinglysynchronous assist.

Further investigation of the conversion of mechanicalssist power into fluid power is recommended, taking intoccount the work done by the myocardium (in simulation).reliminary investigations, as yet unpublished, have indi-ated that when assisted through systolic contraction, theyocardium generates a lower active component of stress.

t is hoped that this could promote remodelling of the mus-le.

. Discussion and conclusions

The results of HIL testing have demonstrated that applyingechanical assistance in the form of direct cardiac compres-

ion can increase blood pressure and cardiac output from aeakened heart.The experimental testing described in Sections 4 and 5 has

emonstrated several of the benefits of HIL simulation iden-ified in Section 2: compared to testing on an animal model,t would not have been possible to perform these repeatedxperiments, all on an identical patient model, in a shortpace of time, in a non-clinical setting. The HIL environment

lso facilitated numerical evaluation of the experiments andssessment of the device’s efficiency.

Compared with a purely numerical simulation, HIL sim-lation enabled evaluation of the effect of the real, physical

TtLL

ntricle diameter: (a) synchronous assist and (b) asynchronous assist.

erformance of the prototype assist device. This included thelectromechanical properties of the motor and the mechanicsf tension transmission via the flexible band to the heart sur-ace. The efficacy of the prototype assist device was assesseds the dimensions of the ventricles changed over the cardiacycle.

The numerical CVS model used in HIL simulation can beurther developed as required by future investigations. Thehort computational time required for the current model didot suggest that future models will have to be greatly simpli-ed, especially given the increasingly available computingower. Ferrari et al. [4] found that LabVIEW running undericrosoft Windows provided limited time for computation,

nd suggested LabVIEW Real Time. For this apparatus were also investigating LabVIEW Real Time, installed on aonventional PC.

The assist device bands each have one degree of freedomto contract circumferentially – however, this form of assistas been shown to produce differing effects on the two ven-ricles. Future work will consider ventricle-specific assist.he simulator can be enhanced by removing the mechan-

cal links between the swing-arms, and providing separateosition controllers for each of the six actuators. Then theifferent compliances of the right and left ventricles can beepresented, as can regional wall motion abnormalities.

Given the proven benefits of this technique, it is likelyhat this hardware-in-the-loop technique will be suitable forvaluation of prostheses and interaction with other biologi-al systems. In such applications, it is the interface betweenardware and software that will require the most attention.n this instance that interface took the form of a heart sim-lator; taking the example of an LVAD, the interface wouldecessarily involve fluid and may take the form of a precisionervo-controlled displacement pump, with pressure transduc-rs to measure the instantaneous pressure rise over the LVAD.

his pressure would be fed into a circulation model similar

o that of Fig. 2, with the addition of a branch through theVAD. That flow would then be presented physically to theVAD via the servo-controlled displacement pump. Arterial

3 ineerin

gimf

A

P

PPLRH

R

[

[

[

[

[

[

[

[

IEEE EMBS, (EMBC ’05). 2005.

74 B.M. Hanson et al. / Medical Eng

rafts, stents, and valves could be evaluated in a very sim-lar manner, and the HIL technique may also be applied to

usculoskeletal prostheses using software models of muscleunction.

ppendix A

Some cardiovascular system (CVS) parameters used:

arameter Healthycondition

Ischaemicheartdisease

eripheral venous resistance (mmHg/ml/s) 1.0 2.4ulmonary resistance (mmHg/ml/s) 0.07 0.15V contractility scaling (dimensionless) 28 14V contractility scaling (dimensionless) 4 2eart rate (bpm) 70 70

eferences

[1] Oz MC, Artrip JH, Burkhoff D. Direct cardiac compression devices. JHeart Lung Transplant 2002;21(10):1049–55.

[2] Macnair R, Underwood MJ, Angelini GD. Biomaterials and cardiovas-cular devices. Proc Inst Mech Eng 1998;212(Part II):465–71.

[3] Wu Y, Allaire P, Tao G, Wood H, Olsen D, Tribble C. An advancedphysiological controller design for a left ventricular assist device toprevent left ventricular collapse. Artif Organs 2003;27:926–30.

[4] Ferrari G, Kozarski M, De Lazzari C, Gorczynska K, Mimmo R,Guaragno M, et al. Modelling of cardiovascular system: develop-ment of a hybrid (numerical-physical) model. Int J Artif Organs2003;26(12):1104–14.

[5] Urbaszek A, Schaldach M. A numerical heart and circulation model tosimulate hemodynamics for rate-responsive pacing. In: Power H, editor.Bio-fluid mechanics. Southampton/Boston: Computational Mechanics

Publications; 1995. p. 129–60.

[6] Pillon M, Duffour H, Jufer M. In vitro experiments: circulatory assistdevice interaction with a virtual cardiovascular system. In: Proceed-ings of Annual International Conference IEEE-EMBS, vol. 14. 1992.p. 740–1.

[

g & Physics 29 (2007) 367–374

[7] Wu Y, Allaire P, Tao G, Liu Y. In-vitro test of an adaptive flow controllerfor a continuous flow LVAD. In: Proceedings of American control con-ference. 2004. p. 1647–8.

[8] Kozarski M, Ferrari G, Clemente F, Gorczynska K, De Lazzari C,Darowski M, et al. A hybrid mock circulatory system: developmentand testing of an electro-hydraulic impedance simulator. Int J ArtifOrgans 2003;26(1):53–63.

[9] Kawaguchi O, Goto Y, Ohgoshi Y, Yaku H, Murase M, Suga H.Dynamic cardiac compression improves contractile efficiency of theheart. J Thorac Cardiovasc Surg 1997;113(5):923–31.

10] Sunagawa K, Burkhoff D, Lim KO, Sagawa K. Impedance load-ing servo pump system for excised canine ventricle. Am J Physiol1982;243(2):H346–50.

11] Isermann R, Schaffnit J, Sinsel S. Hardware-in-the-loop simulation forthe design and testing of engine-control systems. Control Eng Pract1999;7:643–53.

12] Ferreira JA, Gomes Almeida F, Quintas MR, Estima de OliveiraJP. Hybrid models for hardware-in-the-loop simulation of hydraulicsystems. Proc Instn Mech Engrs Part 1: J Syst Control Eng2004;218:465–74.

13] Hanselmann H. Hardware-in-the-loop simulation testing and its inte-gration into a CACSD toolset. In: IEEE international symposium ofcomputer-aided control system design. 1996.

14] Melchior FM, Srinivasan RS, Charles JB. Mathematical mod-elling of human cardiovascular system for simulation of orthostaticresponse. Am J Physiol (Heart Circ Physiol) 1992;262(31):1920–33.

15] Ha R, Qian J, Wang D, Zwischenberger JB, Bidani A, Clark JW. Aclosed-loop model of the ovine cardiovascular system. In: Proceedingsof 26th annual international conference of the IEEE EMBS. 2004. p.3781–4.

16] Olansen JB, Clark JW, Khoury D, Ghorbel F, Bidani A. Aclosed-loop model of the canine cardiovascular system thatincludes ventricular interaction. Comput Biomed Res 2000;33:260–95.

17] Hanson BM, Levesley MC, Watterson K, Walker PG. Simulation of thehuman cardiovascular system for real-time interaction with an assistdevice. In: Proceedings of 27th annual international conference of the

18] Hanson BM, Richardson RC, Davies GLJ, Watterson K, Levesley MC,Walker PG. Control of a non-blood contacting cardiac assist device.In: Proceedings of IASTED international conference on biomedicalengineering BioMED2005. 2005. p. 679–84.