Estimation of dialysis treatment efficiency by means of system identification

When treating patients suffering from renal failure with hemodialysis, an obvious point of interest is the actual blood cleaning efficiency of the dialyzer (artificial kidney). This efficiency is called clearance or dialysance. The method currently used for estimating clearance is based on doing a step-change on the process. Due to the nature of the process, this method is slow and has a relatively large output spread. This master thesis investigates a new method of finding clearance by means of system identification. The dialyzer is modelled as a discrete-time system, and perturbed by use of a pseudo-binary random sequence. The input/output-data is then fed into an optimal Kalman filter for parameter estimation. The gain and offset of the identified system is directly related to the dialyzer clearance of the treatment. The method shows promising results, usually converging to good parameters within 15 minutes, and then tracking changes continuously for the rest of the treatment. It also provides better accuracy, with a considerable reduction in spread compared to the old method. Main obstacles stem from variable time-delays in the system and measurement offsets.