Computational corroboration of a force-driven knee simulation platform

Abstract

Force-driven (FD) simulation offers potential advantages for knee-implant testing; kinematics can adapt to implant geometry [1]. To be effective FD testing must produce well-defined, repeatable conditions – this requires a robust control scheme, leading to greater overall system complexity. It is important to understand the ‘holistic’ operation of the test system (mechanics and control), in order to correctly interpret results; for example, whether specific test outcomes are an artefact of implant design, input waveforms, controller operation, or rig mechanical dynamics. Therefore when using a commercial platform, there is considerable benefit in using computational modelling to augment the testing process, providing additional visualisation capabilities to gain insights into the test. This requires an in-silico model encompassing the full system (controller and mechanical rig) and capturing all aspects of the test dynamics.

Computational models of the AMTI 6-station knee wear simulator have been reported previously[2, 3]. The study by Lanovaz et al is noteworthy in that it identified unintentional system dynamics (flexion arm pliancy). However these models were displacement-driven. We have previously reported a similar displacement-driven model of the simulator, using rigid-body dynamics software[4]. Here, we describe the augmentation of this mechanical model with FD control-system simulation, and subsequent corroboration to validate this model.

The controller was developed in MATLAB/Simulink, using closed-loop PID control coupled with post-hoc dynamically iterating adaptive algorithms and progressive ramping on the different input channels (as with the in-vitro controller). Simulation times were found to increase, and because of the post-hoc nature of the control scheme, multiple trials were required per analysis. We circumvented this increase in computational overhead by initially using high-speed ‘surrogate’ models based on Hertzian contact, and subsequently switching to more detailed ‘discretised’ contact representation only once the tracking error was within a specified tolerance. This provides dynamically-adjustable control over the “performance-accuracy trade-off” through the simulation process.

The in-silico model was verified using bespoke experimental data for several simplified test-cases, and then for full FD ISO-specified gait[5]. Further mechanical pliancy was identified in-vitro (on the tibial side) which needed to be included to reproduce in-vitro results. Tracking was found to be greatly improved through the use of the adaptive controller.

This work paves the way for a better understanding of historical FD test data on this platform, better characterises the mechanics of the in-vitro rig, and provides additional visualisation capability, to analyse contact area, pressures, low-point and contact point motion, and cross-shear in-silico. Ultimately, it is hoped that this more holistic analysis approach can help to answer important questions on the best use of FD simulation in knee implant wear assessment.

More information

Identifiers

Catalogue record

Export record