Iterative learning control of medical applications

  • Iterativ lernende Regelung medizintechnischer Systeme

Stemmler, Maike; Abel, Dirk (Thesis advisor); Leonhardt, Steffen (Thesis advisor)

Aachen : RWTH Aachen University (2023)
Dissertation / PhD Thesis

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2023

Abstract

Natural control mechanisms compose the basis for e.g. the maintenance of homeostasis. However, the aging process, diseases or accidents can disturb their functionality. To recover them, control system theory can be applied. But natural control mechanisms usually have several different functions and there exist multiple cross-couplings. Furthermore, there often is no direct interaction between the technical and biological system, although in most cases the direct interaction is the intended use. However, humans are able to quickly understand complex correlations and learn from previous trials to improve the performance of subsequent ones. This ability is extremely useful for processes that operate in a repetitive manner. Since iterative learning control (ILC) algorithms adopt this characteristic, they are well suited to control repetitive medical processes. According to this, the aim of this thesis is to design ILC approaches for medical applications that adapt the control action on the abilities and individual needs of each patient. Therefore, the patient has to be a part of the control loop. Consequently, the controller model may include non-linearities, time-varying parameters, and inter- and intrapatient parameter variations. Furthermore, the measurability of relevant signals and the quantifiability of the therapeutic aim are not always given. This thesis investigates the main challenges that arise from the patient’s integration in the control loop and the resulting conclusions for the controller design. To verify the results and demonstrate the improvements and advantages of ILC strategies applied to medical applications, two exemplary applications are considered in this thesis, i.e. the control of a left-ventricular assist device and the robot-assisted neuromuscular training. In both cases, the system’s dynamics are identified and an applicable ILC strategy is developed, implemented, and tested in simulation. For this purpose, process-specific adoptions are required, which are analyzed, presented, and discussed in detail. In norm-optimal iterative learning control (NOILC) schemes various different constraints and control objectives can be considered and weighted regarding their importance. For this reason, NOILC schemes instead of conventional, simpler ILC schemes are used throughout this thesis. Apart from this, the concept of state and parameter estimation is introduced and the designed control scheme for the neuromuscular training is also tested in an experimental study with human subjects. Both, the simulation as well as the experimental results demonstrate the superiority of the NOILC scheme compared to conventional therapy approaches. In both applications, the manual adjustment, as well as the monitoring effort, can be reduced to a minimum, as the control schemes automatically adapt to e.g. changing hemodynamic conditions. In conclusion, the results show that NOILC schemes enable patient adaptive control of repetitive medical processes. By addressing physiological variables, natural control mechanisms can be recovered and deformities can be corrected.

Institutions

  • Chair and Institute of Automatic Control [416610]

Identifier

Downloads