Delay Compensation for Real-Time Hybrid Testing

Abstract

Real-time hybrid testing has the potential of being a disruptive addition to the repertoire of experimentalists working in the field of marine technology, where rigorous experimental testing is both commonplace and necessary. The method combines conventional experimental testing with numerical methods in a constructive way, thereby countering weaknesses of either paradigm. However, the integration of experimental and numerical procedures must be done with great care. The fast dynamics and sensitivity to errors often present at the small model scales translate into a requirement of state of the art control design. Of fundamental importance to the majority of real-time hybrid tests is therefore compensation and mitigation of time delays.

It is in this thesis attempted to compensate for delays by means of model based prediction utilized in conjunction with observers instead of the much more common polynomial extrapolation methods. The work resulted in a simple but effective methodology for fusing delayed measurements from the optical motion measurement system, used to measure position and orientation of the experimental model, in a kinematic observer. An observer for linear position and velocity was used in conjunction with an attitude observer and asymptotic stability was shown for the interconnected system. The observers fuse old accelerometer and gyroscope measurements with the delayed position and attitude measurements from the optical motion measurement system. A prediction using only accelerometer and gyroscope data is subsequently performed.

Usage of nonlinear finite element codes with iterative solvers in the real-time hybrid test result in time-varying processing delay. A simple adaptive compensation strategy utilizing a parameter estimator was devised and tested on constant delay experimental data. The method yielded promising results, but the experimental data used did not contain enough nonlinear or otherwise complicated behavior to give a proper validation of the method. It was concluded that further work was needed.