Sensorless Estimation of Pressure Head and Flow of a Continuous Flow Artificial Heart Based on Input Power and Rotational Speed

Abstract

The present study has proposed a new method for estimating the pressure head (P(t)[mm Hg]) and flow (Q(t)[L/min]) of a centrifugal pump on the basis of voltage (V(t)[V]), current (I(t)[A]), and rotational speed (N(t)[k(rpm)]) of the DC motor for a pump without any additional sensors. In the proposed estimation method, two auto-regressive exogenous (ARX) models are employed. One ARX model has an output, P(t) or Q(t), and three inputs, VI(t) = V(t)I(t) and N(t) and the steady state gain (K) of the system from VI(t) to N(t). It can be assumed that K may include the information on viscosity of blood. The coefficient parameters of this ARX model are identified in an off-line fashion before implantation of the pump. After implantation, P(t) or Q(t) is estimated by the same ARX model with the already identified parameters. The other ARX model is used to identify Kon the basis of VI(t) and N(t) in an on-line fashion every time the viscosity of blood may change. In the experiment, a mock circulatory system consisting of a centrifugal pump and a reservoir with 37% glycerin or water was employed. The root mean square error between measured Q(t) and its estimate obtained from the proposed method was 1.66L/min. On the other hand, a different method based on a single ARX model with inputs of VI(t) and N(t), but without the additional input of K, yielded the corresponding estimation error of 2.22L/min. This means that the proposed method can reduce its estimation error by about 25% in comparison with a method that cannot cope with the change in blood viscosity.

Noninvasive Average Flow Estimation for an Implantable Rotary Blood Pump: A New Algorithm Incorporating the Role of Blood Viscosity

“…It is anticipated that further clinical data, as well as acute experiments in animals with models of cardiac failure, will confirm the early positive outcomes presented herein. The flow estimator described here may also represent an important starting point for the development of an algorithm for noninvasive instantaneous flow estimation, through, for example, an ARX model (4,12). Such an estimate of instantaneous flow is an essential tool for clinicians to gain further insight into the interactions between the body and the pump, and is a necessary development on the way to designing a totally implantable, long-term artificial heart.…”

“…Due to the similarity in experimental procedures, the algorithm by Wakisaka et al (8) was tested on the flow maps collected using the VentrAssist; however, results were unsatisfactory, as the constant offset term accounting for HCT failed to describe the difference in flow rates between flow maps. Yoshizawa et al (4), analyzing data collected in a mock loop using water and a 37% aqueous glycerol solution, proposed a new algorithm for a Kyocera C1E3 pump (Kyocera Corporation, Kyoto, Japan). Their model was summarized by the equation …”

“…It is interesting to compare the form of the final equation with the findings of other empiric studies: unlike previous estimators, Eq. 5 does not feature HCT as an external gain factor (as in Yoshizawa et al [4]) or as an additional linear term (as in Wakisaka et al [8]). Blood viscosity was instead selectively incorporated in some of the parametric coefficients, specifically, those related to input electrical power.…”

“…Many research groups claim to have developed satisfactory algorithms for sensorless flow estimation (for a review, see [2]). However, it is evident upon closer inspection of the reported equations that the design characteristics of different pumps cause the proposed algorithms, which have been obtained through an empiric process of pump flow mapping, to be substantially different from each other, and possibly unique to every individual pump design (3)(4)(5). A clear example of the extreme variability of algorithms is also outlined by Ayre et al (6), where it is evident that, even when using the same pump, a small change in impeller shape can introduce significant modifications in the relationship between power, speed, and flow.…”

A Dynamical Model for Pulsatile Flow Estimation in a Left Ventricular Assist Device

“…This is the major limitation of the presented model. Using an autoregressive with exogenous input (ARX) model, Yoshizawa et al (2002) developed a pulsatile flow estimator (Yoshizawa et al, 2002). A Mean absolute error of 1.66 L/min, and a correlation coefficient of 0.85 were obtained when another ARX model has used to compensate for HCT .…”