Medical organ phantoms are increasingly used for training of personnel and validation of new procedures. This thesis is one in a series of works developing an MRI-compatible liver phantom that reproduces breathing motions. To enable feed-forward control, accurate system models predicting the phantom's position are required.
The objective of this thesis was the identification and modelling of the dynamics of this pneumatically actuated liver phantom. The system was divided into a subsystem containing the Proportional Solenoid Valve (PSV), which regulates pressure, and a pneumatic-mechanical subsystem, containing the actuator and platform holding the phantom. With system identification techniques using data from frequency-, step- and ramp-response tests, a transfer function was fitted to the PSV-subsystem which accurately reproduced its frequency response. The model achieved a coefficient of determination of 0.99 for its magnitude response, and 0.96 for its phase response.
Experimental testing showed that the pneumatic-mechanical subsystem displayed nonlinearity and history dependence. Due to these effects, a Long-Short-Term-Memory (LSTM) neural network model was chosen to model its behaviour. Data collected under various operating conditions was used to train the model, resulting in the ability to capture dominant dynamics for input signals with an amplitude above 0.4 bar. Below this threshold, the actuator could not reliably move, introducing unpredictable behaviour. For amplitudes above 0.4 bar, this model achieved a root-mean-square error of 0.39 cm, and a coefficient of determination of 0.97.
These results demonstrate the feasibility of developing predictive models for the liver-phantom system and provide groundwork for the modelling and feed-forward control of redesigned systems.
