Identification and control of peristaltic pumps in hemodialysis machines

Abstract

Abstract-In hemodialysis machines peristaltic pumps are responsible for the transfer of fluids. The main characteristic of these pumps is that they can transport the solutions with significant error. That depends on the tube segment deviation caused through the production. Due to the fact that the medical fluid represents sometimes drugs and in the same time it is required to control the fluid balance of the patient, it is important to transfer these fluids as accurate as possible. The goal of the paper is to design a controller that satisfies the mentioned requirements. First, the system identification is realized by classical moving average method (ARX), followed by the design of two controllers: a classical PID one and a fuzzy controller. After comparing their simulation results, the most preferable one was implemented in practice. Real simulation results of the implemented controller end the paper.

Adaptive soft computing methods for control of hemodialysis machines

“…H(s) = K pump / s (1) The skeleton of the model used for simulations is the same as in [9] (Fig. 1).…”

“…Results are presented mostly through the defined metrics from [9]. The four controllers (PID, adaptive fuzzy system, classical ANFIS, modified ANFIS) are compared through the given properties and only the one with the best performance is realized in practice in a real hemodialysis machine.…”

“…0% slope error is applied, which assumes an ideal tube segment [9]. The error was considered compensated, when the error signal first reached value 0 and did not leave its ±1 ml environment.…”

“…The main goal is to find a lightweight, fast controller, which is adaptive (in speak of the tube segment) and has no residual error. Therefore adaptive neuro-fuzzy inference systems (ANFIS) are designed and compared to the previous results [9,10]. The ANFIS controller is tested on a real hemodialysis system as well.…”

“…In previous works a hemodialysis machine model was identified and control possibilities were compared both with classical and basic soft computing methods [9,10]. Practical implementation was tested as well.…”