Modeling and Compensation of Hysteretic Nonlinearities in Robotic Catheters: Machine Learning Methods versus Mechanical Solutions
Project together with KU Leuven.
Introduction: To treat coronary artery disease (CAD) effectively, catheter-based techniques are often employed in Percutaneous Coronary Intervention (PCI) procedures. The application of catheters in PCI leads to high success rates for recanalization. However, the presence of hysteresis, backlash and deadzones (see Figure.1 ) in catheters may hinder clinicians from reaching accurate intraluminal positioning and steering. Machine learning (ML) methods are appealing for their ability to accurately represent complex nonlinear behavior, albeit highly dependent on the specific application. This makes them suitable for the modeling of the nonlinear systems. Bowden-cables, which are axially incompressible, is a potential mechanical solution for reducing hysteresis phenomenon in robotic catheters [1] [2]. However, the Bowden-cable cannot completely eliminate the hysteresis [3]. How to further improve the performance of the Bowden-cable on hysteresis compensation is still worth studying.
This thesis aims to investigate the modeling and compensation of the hysteresis phenomenon based on two different solutions: Machine learning methods (software) and Mechanical Solutions (hardware). An example of machine learning (ML) methods could be Probability Graphical Models (PGMs). Specifically, the PGMs include e.g. Hidden Markov Models (HMMs), Bayesian Networks, and Conditional Random Field (CRF) techniques. PGMs can be extended to be applied on time series data by considering probabilistic dependencies between the entire time series. Another promising type of ML method is the Support Vector Machine (SVM). The ability of SVM to solve nonlinear regression estimation problems makes SVM successful in time series forecasting. SVM has become a hot topic of intensive study due to its successful application in classification and regression tasks. As for mechanical solutions, Bowden-cable is a potential solution for minimizing hysteresis. Some methods e.g. adding a cable tension control mechanism will be investigated to further improve the performance of Bowden-cables.
There are already two experimental setups available in our lab for data collection and algorithm validation. One is driven by a Pneumatic Artificial Muscle (PAM) (Fig. 2), while the other one is driven by stepper motors via cables (Fig. 3). First, the student(s) have to collect data from one/both of experimental setup. Next, one has to program in Python machine learning libraries and other supplementary libraries such as Pytorch, Numpy, Pandas, Matplotlib to implement the above-mentioned algorithms. Finally, the best modeling performance with machine learning approaches will be compared to the mechanical solutions as well as a classical hysteresis model e.g. generalized Prandt-Ishlinskii model. In a second step these models are adopted for improving the control of the said muscle/cable such that more precise positioning can be achieved and improved trajectories can be followed both in free space and in contact situations. Objectives: Collecting data from an experimental setup that is available in the MISIT lab; Modeling and compensating for the hysteresis behavior in a robotic catheter based on different probabilistic graphical models in machine learning such as Hidden Markov models, Bayesian Networks and Conditional Random Field techniques or other supervised learning models e,g. Support Vector Machine; Comparing the modeling performance of different machine learning methods to a mechanical solution e.g. Bodwen-cables. Implementing Controllers based on the developed models and demonstrate the overall performance. The students are expected to fulfill at least two of the following requirements: Proficiency in Python programming; Knowledge and strong interests in machine learning methods, e.g. probabilistic graphical models; Proficiency in mechanical design and mechatronics prototyping Consolidated knowledge in probability and statistics.
Reference [1] Ali A., Sakes A., Arkenbout E.A., Henselmans P., Starkenburg R. van, Szili-Torok T., Breedveld P. (2019). Catheter steering in interventional cardiology: mechanical analysis and novel solution. Proc. Inst. Mech. Eng. Part H: Journal of Engineering in Medicine, 12 p. [2] https://www.bitegroup.nl/maneuverable-devices/sigma-catheter-steering-inside-the-heart/ [3] Kesner, S.B. and Howe, R.D., 2011. Position control of motion compensation cardiac catheters. IEEE Transactions on Robotics, 27(6), pp.1045-1055.
Contact:
J.Dankelman, j.dankelman@tudelft.nl or
Di Wu di.wu@kuleuven.be |