Sage Journals: Discover world-class research

Abstract

Medical advancements, particularly in ventricular assist devices (VADs), have notably advanced heart failure (HF) treatment, improving patient outcomes. However, challenges such as adverse events (strokes, bleeding and thrombosis) persist. Computational fluid dynamics (CFD) simulations are instrumental in understanding VAD flow dynamics and the associated flow-induced adverse events resulting from non-physiological flow conditions in the VAD.

This study aims to validate critical CFD simulation parameters for accurate VAD simulations interacting with the cardiovascular system, building upon the groundwork laid by Hahne et al. A bidirectional coupling technique was used to model dynamic (pulsatile) flow conditions of the VAD CFD interacting with the cardiovascular system. Mesh size, time steps and simulation method (URANS, LES) were systematically varied to evaluate their impact on the dynamic pump performance (dynamic $H - Q$ curve) of the HeartMate 3, aiming to find the optimal simulation configuration for accurately reproduce the dynamic $H - Q$ curve. The new Overlapping Ratio (OR) method was developed and applied to quantify dynamic $H - Q$ curves.

In particular, mesh and time step sizes were found to have the greatest influence on the calculated pump performance. Therefore, small time steps and large mesh sizes are recommended to obtain accurate dynamic $H - Q$ curves. On the other hand, the influence of the simulation method was not significant in this study. This study contributes to advancing VAD simulations, ultimately enhancing clinical efficacy and patient outcomes.

Keywords

Ventricular assist devices rotary blood pump computer simulation cardiovascular modelling computational fluid dynamics,lumped parameter model

Introduction

Advancements in medical technology have revolutionized the treatment of heart failure, prominently featuring ventricular assist devices (VADs) as life-saving interventions. These rotary blood pumps play a pivotal role in supporting heart function for patients with severe heart failure (HF), significantly enhancing quality of life and survival rates.¹ Nevertheless, despite therapeutic advances, patients continue to experience significant adverse events, including strokes, gastrointestinal bleeding and device thrombosis, associated with the use of VADs.^2,3

The aim of research in the field of VADs is to improve the therapy and to minimise the adverse events associated with VADs. Computational fluid dynamics (CFD) simulations have become invaluable tools for investigating the non-physiological flow dynamic within VADs, which leads to these adverse events. Factors such as high shear stress and exposure times of unfavourable flow conditions on the blood components are key contributors.^4–6 CFD simulations are therefore an essential part of the development of VADs.

CFD simulations are often the only way to gain a deeper insight into the flow of the VAD and to evaluate relevant factors to minimise adverse events.

In order to achieve the optimal VAD design and operational conditions for the sufficient HF patient support, it is essential that the CFD simulations accurately reflect the real-world conditions. Therefore, CFD simulation of the VAD should be performed in interaction with the cardiovascular system to account for the dynamic (pulsatile) operation environment of the VAD in the human body.⁷ Since the VAD operates together with the still beating heart, dynamic boundary conditions at the inlet and outlet of the CFD domain of the VAD simulation are desirable, as they consider the physiological behaviour.

CFD simulations under dynamic inflow conditions^8–14 are particularly challenging, as they differ significantly from static boundary conditions. The dynamic (pulsatile) flow conditions within a ventricular assist device (VAD) are significantly influenced by both the patient’s remaining heart function and the adjusted pump speed.⁷ Consequently, the flow rate and pressure head dynamically changes (in time) throughout the cardiac cycle, resulting in dynamic progression of the pump performance, indicating a hysteresis for the dynamic $H - Q$ curve.⁸ VADs operate in two distinct states: full support and partial support. During full support, the pump manages the entire cardiac output, with the aortic valve remaining closed. In contrast, the cardiac output is shared between the VAD and the native heart when the VAD operates within a partial support range. Both support modes exhibit this dynamic progression of the $H - Q$ curve,⁹ presenting a challenge for accurate simulation due to the intricate interaction between the pulsating heart and the operating VAD.

In previous studies from the literature, dynamic boundary conditions were already utilized to simulate VAD flow interacting with the cardiovascular system.^8–14 These dynamic boundary conditions were often defined through experiments, employing in vitro or in vivo measurements. In contrast, in a novel approach of us, introduced by Hahne et al.,¹⁰ the 3D-VAD simulation was bidirectionally coupled with a lumped parameter model, representing the dynamic blood flow in the cardiovascular system. This integration allows for the automatic generation of dynamic boundary conditions without experiments beforehand.

As with all CFD studies, it is also essential to verify and validate the accuracy of VAD simulations, as changes to the simulation setup in CFD can have a significant impact on the simulation outcome. When examining literature studies that perform VAD CFD simulations under pulsatile boundary conditions, it is evident that a proper validation of the simulation is often lacking.^11–13 Typically, validation is approached in one of three ways: (a) Validation at static operation points using ‘static CFD simulations’^14–16; (b) no validation of the dynamic behaviour (dynamic $H - Q$ curve) at all^11,12 or (c) validation using either pulsatile flow rates $Q$ ^17,18 or pressure curves,¹⁵ but not both together in a dynamic $H - Q$ curve.

The authors believe that further research is necessary in the field of validating flow simulations in VADs, particularly regarding the impact of simulation parameters on dynamic $H - Q$ curves. To facilitate proper validation, we will introduce a novel quantification metric, the Overlapping Ratio (OR) method. This method allows for the validation of simulated dynamic $H - Q$ curves against a given experimental or theoretical validation curve by calculating the ratio of the overlapping areas between the two curves.

The aim of this study is to investigate the impact of simulation parameters on the simulated pump performance in a VAD simulation with dynamic boundary conditions. This research builds upon the work of Hahne et al.¹⁰ In contrast to their general analysis of dynamic VAD flow interacting with the cardiovascular system, our study focusses on refining simulation methodology to accurately reproduce pump performance. Therefore, we systematically investigate the effects of varying simulation parameters – mesh size, time step and simulation method – to determine the optimal configuration for our bidirectionally coupled, dynamic VAD simulations for an accurate replication of the dynamic $H - Q$ curve.

Methods

Cardiovascular system model

To achieve patient-specific, dynamic (pulsatile) boundary conditions for the VAD CFD, we developed a bidirectional coupling between a lumped parameter model (LPM) simulating the cardiovascular system and the time-resolved 3D-CFD of the HeartMate 3. This coupling strategy accounts for the time-dependent interaction of the cardiovascular system with the VAD and is extensively detailed in a previous study.¹⁰ A brief overview: Over several heartbeats, the LPM and the VAD CFD exchange after every time step crucial flow variables such as pressure and flow rate in a bidirectional and time-dependent manner. As both the LPM and the CFD respond to changes in these variables (e.g. the pressure-volume loop of the heart changes under VAD support), this interaction can generate pulsatile, dynamic boundary conditions for the VAD CFD.

The HF patient was represented in this study by adjusting the pressure-volume loop (p–V loop) of the left ventricle to the measured data as presented in Warriner et al.¹⁹ Specifically, we utilized the mean p–V loop derived from 129 HF patients diagnosed with advanced heart failure according to ACC/AHA Stage D. This results in a modelled, damaged left ventricle with an end-diastolic left heart volume of 273 ml and an ejection fraction of 25%. It can therefore be reasonably asserted that the boundary conditions accurately reflect the characteristics of a ‘generic’, advanced heart failure patient.

Simulation method and turbulence modelling

Large-Eddy simulations (LES) and unsteady Reynolds-averaged Navier-Stokes (URANS) computations were employed to analyse the flow field in the HeartMate 3. LES offers a significant advantage by providing a detailed understanding of the flow field, including turbulence and turbulent stresses, as it directly resolves a portion of the turbulent flow. However, this advantage comes at the cost of longer computation times and more extensive flow verification compared to URANS. Both simulation methods utilized the commercial solver ANSYS Fluent 2023R2. Advection and diffusion fluxes were discretized using the Bounded Central Difference Scheme for LES and a Second Order Upwind Scheme for URANS.

For URANS turbulence modelling, the $k - ω$ -SST model was used, incorporating the curvature correction method developed by Menter²⁰ and Smirnov et al.²¹ Conversely, the WALE model was employed for sub grid scale (turbulence) modelling in LES.²²

For both methods, an implicit, second order differencing scheme was used for the discretization of the transient term. In addition, a double precision solver was used. The numerical model of the VAD contains both rotary and stationary domains. Both parts are connected by transient sliding interfaces.

Computational setup

Geometry and computational domain

The investigated VAD is the HeartMate 3 from Abbott Laboratories. We received the geometry digitally, which has been utilized in previous research studies.¹⁴ Inflow and outflow cannulas were included in the computational domain.

Meshing

Block-structured grids with hexahedral-elements were manually created using ICEM CFD 2022R1. All hydraulic parts of the pump were considered. Two different meshes were created for URANS to examine the influence of the mesh on the dynamic $H - Q$ curve.

The mesh with 1 million nodes (Figure 1(a)), and the mesh, with 11 million nodes (Figure 1(b)), both exhibit the similar block structure. In the 1 million-node mesh, grid angles are maintained above 33°, and aspect ratios are below 600. In the 11 million-node mesh, grid angles exceed 36°, and aspect ratios are less than 100.

Figure 1.

Overview of the mesh around the impeller: (a) 1 million mesh, (b) the same domain with 11 million mesh and (c) A zoomed view in the 11 million impeller mesh.

For studies utilizing the LES method, the 11 million-node mesh was also employed. The suitability of this mesh for LES was determined based on criteria beyond angles and aspect ratio considerations. Specifically, unless this mesh was not explicitly constructed for wall-resolving LES,²³ we will demonstrate in the results that the LES computation can be attributed as ‘LES-worthy’ in most parts of the domain. A larger, explicitly constructed LES mesh was not a viable option due to computational constraints in the current, bidirectional coupling strategy.

Boundary conditions and time steps

Blood was modelled as a Newtonian fluid, a valid assumption supported by our previous study demonstrating that blood’s non-Newtonian properties in low shear rate regions of the VADs have a negligible effect on the pump characteristics.²⁴ The fluid viscosity was $μ = 3.5 mPas$ and the density was $ρ = 1050 kg / m^{3}$ .²⁵

Pulsating, dynamic pressure boundary conditions for the VAD CFD were automatically provided by the LPM of the cardiovascular system as described in Hahne et al.¹⁰ The CFD returns the instantaneous VAD flow rate to the LPM to account for the dynamic response in the cardiovascular system under VAD support. This exchange of boundary conditions occurs at every computed time step in the LPM and CFD. The rotational speed of the impeller was set to $n = 5500 1 / \min .$ Two time step sizes of 2° and 20° rotation of the impeller were used in our parametric study to analyse the influence of temporal discretization, see in Table 1.

Table 1.

Simulations with parameter space.

Simulation	Method	Grid size	Time step
URANS-1M-2°	URANS	1 million nodes	2° rotation of impeller
URANS-1M-20°	URANS	1 million nodes	20° rotation of impeller
URANS-11M-20°	URANS	11 million nodes	20° rotation of impeller
URANS-11M-2°	URANS	11 million nodes	2° rotation of impeller
LES-11M-2°	LES	11 million nodes	2° rotation of impeller

Each simulation computed data from two heartbeats, with the dynamic $H - Q$ curve extracted from the final beat. Iterative convergence was monitored during the CFD simulation by ensuring that the RMS residuals were approximately $10^{- 5}$ .

Analysis of parameter space for result evaluation

The aim of this study is to assess how various simulation parameters impact the dynamic $H - Q$ curve of a VAD simulation in interaction with the cardiovascular system. Our objective is to identify the simulation results that most closely align with the pulsating characteristic curve observed from in vitro experiments and a mathematical model by Boes et al.⁹ To achieve this, we varied parameters such as simulation method, grid size and time step (see Table 1).

All URANS computations were performed with different grid sizes and time steps. Additionally, a LES simulation was conducted using the finest grid size (11M grid nodes) and the smallest time step (equivalent to 2° impeller rotation). At this specific time step, with a root-mean-square Courant number of 1.4, the LES simulations align with the recommended Courant numbers for LES ( $\leq 1$ ) as per the literature.²⁶

Metric of quantification of the H–Q curves

In order to quantify and evaluate the dynamic $H - Q$ curves, a new, intuitive method was developed to create a quantification metric, in which the Overlapping Ratio ( $O R$ ) of the analysed curve and the validation curve is formed (as shown in Figure 2). This method calculates the ratio of the overlapping area to the total area (overlapping area + residual areas), providing a quantitative measure of the overlap between curves.

Figure 2.

An Example for the Overlapping Ratio (OR): The ratio between the overlapping area (in green) ( $A_{O v e r l a p}$ ) and the sum of the overlapping area ( $A_{O v e r l a p}$ ) and the residual areas (in red) ( $A_{R e s i d u a l s}$ ) of both curves.

The OR is the ratio between the overlapping area $(A_{o v e r l a p}$ ) and the sum of the overlapping area ( $A_{o v e r l a p}$ ) and the residual areas ( $A_{R e s i d u a l s}$ ) of the simulated curve and the validation curve (from experiments), as descriped in equation (1).

OR = \frac{A_{Overlap}}{A_{Overlap} + A_{Residuals}}

(1)

$OR = 1$ represents an identical match with the validation curve, while the value 0 indicates the absence of overlapping areas with the validation curve. It is important to note that even minor discrepancies can result in a significantly reduced OR value.

Results and discussion

Quality assessment of LES

Every CFD study needs to be verified and validated to account for the quality of the simulation. The verification deals with estimating numerical errors, for example, due to the computational grid or time step. Contrarily, the validation examines if the simulation results reflect realistic results, which is realized by comparing the simulation results with experimental data. The URANS and LES methods are verified by investigating the convergence of time steps and mesh sizes. In a broader sense, the comparison of different dynamic $H - Q$ curves and the comparison of those with a validation curve in the next paragraph serve as both verification and validation. Nonetheless, further validation of the LES was required as the meshes were designed for URANS and not specifically according to LES criteria.

To validate the LES, we aim to delve deeper into the quality of our LES computation. Given that LES resolves a portion of the turbulent flow field directly, as discussed by Torner et al.,^23,27 we seek to analyse the extent to which our LES simulation achieves this level of flow resolution and if the LES can be considered as ‘LES-worthy’. Therefore, a LES quality assessment study was conducted at a static operating point, which is in the ‘middle’ of the analysed, dynamic $H - Q$ curve: $(H, Q, n) = (60 mmHg, 5.5 l / \min, 5500 1 / \min)$ .

For LES quality assessment, three verification criteria were applied, which are also explained in previous publications from our group.^27–30 First, the degree of turbulent flow resolution was estimated in the core flow region. One way to assess this is to estimate the Kolmogorov length $η_{k}$ by the relation of kinematic viscosity $ν$ and turbulent dissipation rate $ε_{t u r b}$ :

η_{k} = {(\frac{ν^{3}}{ε_{turb}})}^{\frac{1}{4}},

(2)

which follows from the theory of isotropic turbulence.³¹ The local ratio $Δ / η_{k}$ is the first criteria and gives information about the degree of resolution of dissipative scales through the grid.²⁶ The grid variable $Δ$ was calculated in every grid cell by $Δ = {(Δ_{x} Δ_{y} Δ_{z})}^{1 / 3}$ . The ratio was $\leq 9$ throughout the whole flow domain, which is smaller than the ratios of $Δ / η_{k} < (12 - 25)$ defined as desirable for LES from Fröhlich²⁶ or Celik et al.³²

Secondly, a turbulence kinetic energy (TKE) frequency analysis as second criteria was performed in one spatial point of the impeller to assess the physical resolution of turbulence.²⁶ The resulting spectrum is shown in Figure 3. Aim of LES is to resolve the TKE spectrum up to the inertial subrange.²⁶ In this range, small turbulence scales are increasingly isotropic, making LES sub-grid scale modelling reasonable. According to Kolmogorov’s theory, this portion of the spectrum is characterized by a slope of $f^{- 5 / 3}$ (red line in Figure 3). In the figure, the middle part of the spectrum shows a continuous slope of $f^{- 5 / 3}$ over approximately a decade in the frequency range. This suggests that a significant portion of the inertial subrange is resolved by our LES.

Figure 3.

Turbulent kinetic energy (TKE) spectrum of the 11M mesh. The initial subrange converges towards $ƒ^{- \frac{3}{3}}$ .

As a final and third verification step, we applied the Power Loss Analysis (PLA) outlined by Torner et al.^27,33 to assess if flow losses are accurately resolved. The total power loss $P_{t o t, 1}$ was calculated from the difference between the power at the coupling and the hydraulic power increase. This is consistent with the pump’s performance, for example in terms of flow rate $Q$ and pressure head $H$ . In contrast, the total power loss $P_{t o t, 2}$ is the sum of the losses from direct, resolved turbulent and modelled turbulent dissipation. It reflecting the fluidic stresses present within the pump’s flow. In theory, the losses from $P_{t o t, 1}$ and $P_{t o t, 2}$ should be equal.

Since these flow losses relate to the pump characteristics ( $P_{t o t, 1}$ ) as well as with fluidic shear stresses ( $P_{t o t, 2}$ ) in a VAD,²⁷ we can make a statement if these variables are properly computed. The PLA methodology is detailed in Torner et al.²⁷ For this PLA, we performed additional LES computations on three further mesh sizes (1M, 22M and 70M) at the static operation point to account for the impact of the LES grid.

The results depicted in Figure 4 illustrate the outcomes of the Power Loss Analysis (PLA). The calculated power loss $P_{t o t, 1}$ , derived from pump performance metrics including head, torque, speed and flow rate, exhibits a convergent pattern across mesh sizes of 11M, 22M and 70M. Notably, there is a discernible reduction in the relative deviation between $P_{t o t, 2}$ and $P_{t o t, 1}$ as the mesh size increases. Here, $P_{t o t, 2}$ represents the power loss calculated by the internal dissipative losses, which are linked to fluidic stresses.³³

Figure 4.

LES results of the Power Loss Analysis at different grid sizes.

At the 70M mesh, constructed according to wall-resolving LES guidelines, the relative deviation is approximately one third compared to the result of the 1M mesh ( $- 14 %$ vs $- 39 %$ ), indicating better reproduction of fluidic stresses at high mesh sizes. For our mesh of interest (11M mesh), the relative deviation between $P_{t o t, 2}$ and $P_{t o t, 1}$ is also approximately half compared to the 1M mesh ( $- 22 %$ vs $- 39 %$ ). However, the relative deviation is still notably higher as with the 70M mesh, suggesting that further mesh refinement in needed to properly account for the fluidic stresses in the LES with 11M grid nodes.

Therefore, our quality assessment of the LES indicates a satisfactory outcome from the first two verification methods, leading us to conclude that our LES simulation with the 11M mesh meets the criteria for being deemed ‘LES-worthy’. The PLA has revealed that the pump performance ( $P_{t o t, 1})$ is properly computed by the LES, although there is potential for further improvement in the calculation of the pump’s dissipative losses $P_{t o t, 2}$ . Consequently, additional refinement is required for the LES at the 11M mesh in order to account for dissipative losses and other related flow variables of interest, such as fluidic stresses, which will be crucial for blood damage assessment.

One might wonder why the finer LES meshes (e.g. 70M) were not used for the analysis of the dynamic $H - Q$ curves in this study. The reason for this is the current significant computational resource requirement of the bidirectional coupling of LPM and CFD. Currently, we require approximately 2 weeks of computation time on 30 local CPU cores to conduct the LES computation with 11M grid nodes and a 2° time step to simulate two heart beats resulting in a dynamic $H - Q$ curve for every heartbeat. However, we are currently working on accelerating the computations to also enable fully wall-resolved LES computation in interaction with the dynamic cardiovascular system.

Dynamic $H - Q$ curves

The dynamic $H - Q$ curves are presented in Figure 5, where they are compared for accuracy with the results of a mathematical model by Boes et al.⁹ The mathematical model has been derived with experimentally measured data of the HeartMate 3. It has been implemented in the Lumped Parameter Model to calculate the dynamic $H - Q$ curve (grey lines) used for validation. These $H - Q$ curves offer a comprehensive overview of our parameter study.

Figure 5.

Dynamic $H - Q$ curves with different simulation parameters. Comparison of URANS Simulations with: (a) and (b) different mesh sizes and (c) different step sizes and (d) different simulation methods.

In Figure 5(a), we illustrate the influence of a large time step (20°) on URANS computations with two different mesh sizes. Particularly during the dynamic transition from low to highest flow rates (diastole to end-systole) and vice versa (end-systole to diastole), both simulations exhibit significant deviations from the validation curve (grey line). The OR value increases of only 0.10 to 0.24 with the larger mesh size, which can be considered as still relatively low. This improper alignment is attributed to the large time step, which prevents the simulation from accurately capturing the dynamic changes in pump performance. Importantly, even with a finer mesh size, the computation of pump performance would not be improved in this case.

As shown in Figure 5(b), the chosen time step has again a significant effect on the result, increasing the OR value significantly from 0.24 to 0.63 by reducing the time step from 20° to 2° for the 11M mesh. The simulation with the smaller time step is more accurate in capturing the dynamic changes of the pump than the simulation with the larger one.

When comparing different meshes computed at the same, smaller time step (Figure 5(c)), an increase in grid size leads just to a slightly improved $H - Q$ curve with a rise of OR value from 0.44 to 0.63. Here, it is evident from the comparison of Figure 5(a–c) that in our study the time step has a more significant influence on the accurate reproduction of the dynamic $H - Q$ curve as changes in mesh size.

Lastly, the results between the URANS and LES methods for the finest mesh (11M) and smallest time step (2°) are compared in Figure 5(d). Both simulation methods produced comparable results for the pump performance, resembling the validation curve with just small deviations (OR values: 0.64 vs 0.63). However, the curve generated with URANS exhibited a smoother progression compared to the LES curve. In LES, some temporal fluctuation of the pressure head at a certain $Q$ are noticeable indicating the LES ability of capturing large-scale turbulent eddies.

It can be concluded that the selected time step has the most significant impact on the result, increasing the OR value from 0.24 to 0.63 by reducing the time step from 20° to 2°. An examination of the H–Q curves with the highest OR (as illustrated in Figure 5(d)) reveals a notable correlation with the validation curve. Consequently, we consider an OR value exceeding 0.60 as a first guess of a desirable target to quantify a properly computed dynamic $H - Q$ curve.

Summary and conclusion

In this study, we investigated several simulation parameters for their influence on computing the simulated, dynamic $H - Q$ curve of a VAD. Modifying the simulation parameters led to variations in the dynamic $H - Q$ curve. Specifically, we found that small grid sizes and large time steps are unfavourable for accurately capturing dynamic pump performance, with the time step emerging as the more crucial parameter. Consequently, we propose the utilisation of a small time step size, for example, $C F L \approx 1,$ in conjunction with an appropriate, large mesh size (quantifiable using the PLA). Additionally, employing the LES method led to some turbulent fluctuations in the pressure head at a certain $Q$ , although the overall dynamic curves remained similar compared to URANS. Both URANS and LES with the finest grid and the smallest time step were able to reproduce the dynamic $H - Q$ curve with an OR above 0.6.

This study offers a structured overview of the influence of simulation parameters on the dynamic pump performance of a VAD and implements a new intuitive method to quantify the accurate reproduction of the dynamic $H - Q$ curves. However, it is important to acknowledge some limitations, which should be addressed in the future. Firstly, the investigation focussed solely on the HeartMate 3 blood pump within the cardiovascular system of a ‘generic’, virtual patient. Consequently, the generalizability of the findings to different VADs and patient populations remains to be explored in future research. Additionally, the meshes employed in our simulations do not meet the requirements for a wall-resolved LES. Despite an accurate reproduction of the pump performance, our findings from the LES quality assessment indicate that further refinement of the LES grid is necessary to precisely assess the fluidic stresses within the blood flow. Therefore, future studies should aim to address these limitations to provide also comprehensive insights into the stress fields and blood damage potentials of VADs under dynamic flow conditions.

Footnotes

Acknowledgements

The authors gratefully thank Dr. Bente Thamsen for providing the digital geometry of the HeartMate 3.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Vincenz Crone

Benjamin Torner

References

Severino

Mather

Pucci

, et al. Advanced heart failure and end-stage heart failure: does a difference exist. Diagnostics 2019; 9: 170.

Jorde

Saeed

Koehl

, et al. The society of thoracic surgeons intermacs 2023 annual report: focus on magnetically levitated devices. Ann Thorac Surg 2024; 117: 33–44.

Mehra

Goldstein

Cleveland

, et al. Five-year outcomes in patients with fully magnetically levitated vs axial-flow left ventricular assist devices in the MOMENTUM 3 randomized trial. JAMA 2022; 328: 1233–1242.

Mehra

MR.

The burden of haemocompatibility with left ventricular assist systems: a complex weave. Eur Heart J 2019; 40: 673–677.

Köhne

Haemolysis induced by mechanical circulatory support devices: unsolved problems. Perfusion 2020; 35: 474–483.

Schüle

Thamsen

Blümel

, et al. Experimental and numerical investigation of an axial rotary blood pump. Artif Organs 2016; 40(11): E192–E202.

Tchoukina

Smallfield

Shah

KB.

Device management and flow optimization on left ventricular assist device support. Crit Care Clin 2018; 34: 453–463.

Noor

Parker

, et al. Investigation of the Characteristics of HeartWare HVAD and Thoratec HeartMate II Under Steady and Pulsatile Flow Conditions. Artif Organs 2016; 40: 549–560.

Boes

Thamsen

Haas

, et al. Hydraulic characterization of implantable rotary blood pumps. IEEE Trans Biomed Eng 2019; 66: 1618–1627.

10.

Hahne

Crone

Thomas

, et al. Interaction of a ventricular assist device with patient-specific cardiovascular systems: in-silico study with bidirectional coupling. ASAIO J. Epub ahead of print 27 March 2024. DOI: 10.1097/MAT.0000000000002181

11.

Boraschi

Bozzi

Thamsen

, et al. Thrombotic risk of rotor speed modulation regimes of contemporary centrifugal continuous-flow left ventricular assist devices. ASAIO J 2021; 67: 737–745.

12.

Fang

Boraschi

, et al. Insights into the low rate of in-pump thrombosis with the HeartMate 3: Does the artificial pulse improve washout? Front Cardiovasc Med 2022; 9: 775780.

13.

Wisniewski

Einfluss der pulsatilen Durchströmung einer axialen Blutpumpe auf den Radialspaltwirbel und Bewertung der resultierenden Blutschädigung. Rostock: Shaker Verlag, 2022.

14.

Wiegmann

Thamsen

Zélicourt

D de

, et al. Fluid dynamics in the HeartMate 3: influence of the artificial pulse feature and residual cardiac pulsation. Artif Organs 2019; 43: 363–376.

15.

Gou

Huang

, et al. Evaluation of the hemolysis and fluid dynamics of a ventricular assist device under the pulsatile flow condition. J Hydrodyn 2019; 31: 965–975.

16.

Song

Untaroiu

Wood

, et al. Design and transient computational fluid dynamics study of a continuous axial flow ventricular assist device. ASAIO Journal 2004; 50: 215–224.

17.

Huang

Lei

Ying

, et al. Numerical hemolysis performance evaluation of a rotary blood pump under different speed modulation profiles. Front Physiol 2023; 14: 1116266.

18.

Chen

Jena

Giridharan

, et al. Shear stress and blood trauma under constant and pulse-modulated speed CF-VAD operations: CFD analysis of the HVAD. Med Biol Eng Comput 2019; 57: 807–818.

19.

Warriner

Brown

Varma

, et al. Closing the loop: modelling of heart failure progression from health to end-stage using a meta-analysis of left ventricular pressure-volume loops. PLoS One 2014; 9: e114153.

20.

Menter

FR.

Two-equation eddy-viscosity turbulence models for engineering applications. AIAA Journal 1994; 32: 1598–1605.

21.

Smirnov

Kapetanovic

Braaten

, et al. Application of the SAS Turbulence Model to Buoyancy Driven Cavity Flows. In: Proceedings of the ASME Turbo Expo 2009: Presented at the 2009 ASME Turbo Expo, June 8–12, 2009, Orlando, FL, USA, pp.1207–1216. New York, NY: ASME.

22.

Nicoud

Ducros

Subgrid-scale stress modelling based on the square of the velocity gradient tensor. Flow Turbul Combust 1999; 62: 183–200.

23.

Torner

Konnigk

Hallier

, et al. Large eddy simulation in a rotary blood pump: viscous shear stress computation and comparison with unsteady Reynolds-averaged Navier-Stokes simulation. Int J Artif Organs 2018; 41: 752–763.

24.

Knüppel

Thomas

Wurm

F-H

, et al. Suitability of different blood-analogous fluids in determining the pump characteristics of a ventricular assist device. Fluids 2023; 8: 151.

25.

Fraser

Taskin

Griffith

, et al. The use of computational fluid dynamics in the development of ventricular assist devices. Med Eng Phys 2011; 33: 263–280.

26.

Fröhlich

Large Eddy Simulation turbulenter Strömungen: Mit 145 Abbildungen und 14 Tabellen. Wiesbaden: B.G. Teubner Verlag/GWV Fachverlage GmbH, 2006.

27.

Torner

Konnigk

Abroug

, et al. Turbulence and turbulent flow structures in a ventricular assist device-A numerical study using the large-eddy simulation. Int J Numer Method Biomed Eng 2021; 37: e3431.

28.

Torner

Hallier

Witte

, et al. Large-eddy and unsteady Reynolds-averaged Navier-stokes simulations of an axial flow pump for cardiac support. In: Proceedings of ASME Turbo Expo 2017, Charlotte, NC, USA, 26–30 June 2017. American Society of Mechanical Engineers.

29.

Torner

Flow in a ventricular assist device - pump performance & blood damage prediction: application area 7: biomedical flows. ERCOFTAC Knowledge Base Wiki, https://kbwiki.ercoftac.org/w/index.php/AC7-03 (2023, accessed 23 April 2024).

30.

Torner

Erforschung der Strömung in einem Herzunterstützungssystem unter Berücksichtigung des Turbulenzeinflusses auf die Blutschädigungsvorhersage. Dissertation, Düren, Shaker Verlag, 2019.

31.

Pope

SB.

Turbulent flows. Cambridge: Cambridge University Press, 2012.

32.

Celik

Klein

Freitag

, et al. Assessment measures for URANS/DES/LES: an overview with applications. J Turbul 2009; 7: N48.

33.

Torner

Konnigk

Wurm

F-H.

Influence of turbulent shear stresses on the numerical blood damage prediction in a ventricular assist device. Int J Artif Organs 2019; 42: 735–747.

Dynamic VAD simulations: Performing accurate simulations of ventricular assist devices in interaction with the cardiovascular system

Abstract

Keywords

Introduction

Methods

Cardiovascular system model

Simulation method and turbulence modelling

Computational setup

Geometry and computational domain

Meshing

Boundary conditions and time steps

Analysis of parameter space for result evaluation

Metric of quantification of the H–Q curves

Results and discussion

Quality assessment of LES

Dynamic H − Q curves

Summary and conclusion

Footnotes

Acknowledgements

Declaration of conflicting interests

Funding

ORCID iDs

References

Dynamic $H - Q$ curves