Dynamic Modelling of a Fluidic Muscle with a Comparison of Hysteresis Approaches

Abstract: n recent years, there has been a surge in interest and research into the utilisation of soft actuators within the field of robotics, driven by the novel capabilities of their inherently compliant material. One such actuator is the Pneumatic Artificial Muscle (PAM) which offers a high power-to-mass ratio, compliance, safety, and biological mimicry when compared to their traditional counterparts. However, because of their flexible and complex physical structure and the compressibility of air inside the PAM, they exhibit nonlinear dynamic behaviour, largely due to the influence of the hysteresis phenomenon. In order to implement strategies to counteract this effect, it first needs to be modelled. As such, this thesis investigates two approaches, namely the Maxwell-Slip (MS) and generalised Bouc-Wen (BW) models. Firstly, the test muscle's initial braid angle, maximum displacement, and maximum force are determined to establish the static force using a modified model. Data is then collected on the PAM's force-displacement hysteresis for 2-6 bar of pressure. Using the results from these experiments, the MS and BW model parameters are identified through optimisation. With the static and hysteresis force components characterised, two complete dynamic models are created. The findings show that, when compared to the collected force-displacement data, the BW model has greater accuracy for all pressures except at 4 bar, although both approaches demonstrate results within a satisfactory margin. Lastly, a model validation is conducted to compare the models using a new dataset, separate from the one on which they were trained. Data for this test is recorded at a pressure of 4 bar with a more complex reference that covers four different regions of the muscle's displacement range. Thereafter, both dynamic models are applied to assess their performance. It is evident from the results that the BW model produces a better outcome than the MS, achieving a normalised error of 5.3746% as compared to the latter's 12.835%. The higher accuracy of the generalised BoucWen method is likely due to it having a more complex structure, specialised parameters, and the ability to model asymmetric hysteresis. The Maxwell-Slip model may however still be preferable in some applications due to its relative simplicity and faster optimisation.