Modern Control Engineering is still greatly challenged by complex highly nonlinear dynamical systems with many inputs and outputs. Anyone who deeply studies the problem of automating kite energy soon finds that while kites can can be controlled for long periods, such control is not guaranteed, and any kite eventually finds a way to crash. We work on faith that daunting AWE automation challenges will yield to persistent study, for there is no equivalent success story yet anywhere in the history of control.
Kite control needs to be robust, optimal, tractable, transparent, and scalable. Robust means reliable and that ordinary changes in the wind and kite set-up do not crash the system. Optimal means that the control system need not work too hard for the desired output, for example, servos do not constantly churn in normal operation. Tractable means that the system can adequately compute its solutions in realtime. Transparent means you can easily look into the system and figure out what is going on. Scalable means that modular kite units can be aggregated in large numbers to work in coordination, such as dense non-interfering arrays and flocking behaviors.
Model Predictive Control (MPC) is a style of control engineering that looks at the state of a system while consulting a model to predict an optimal control strategy. Its been likened to the function of headlights at night, where a brighter light reaches father for more certain direction. Models take many forms, but a state-space or state-machine in clean matrix form is desirable. The ultimate model is a complete look-up table where every state has a correct control response specified, but this is only fully practical for situations like sailing by a tide-table or toy worlds like tic-tac-toe. Look-up (search) is very fast compared to systems that must compute an unknown solution in a vast state-space.
Non-Linear Model Predictive Control (NMPC) extends MPC to address effects like chaos. This is a fashionable paradigm in kite control, from Leuven to Torino, but don't be fooled by the reams of preliminary equations bandied about; we are still far from any adequate kite model and characterization of the chaos. Progress will consist of large high-dimensional real-world wind data sets applied to realistic kite models rich with failure-mode states to actively avoid.
At present: effective complex kite control is only practical by human piloting and supervision assisted by ad-hoc classical control loops. Embodied situated logic methods (inherent stabilities; "training wheels") are key expedients. Likely AWE control winners will be early fielded systems able to generate revenue while evolving toward full automation.
FairIP/CoopIP ~Dave Santos June 14,2010 M1634
Comment and development of this topic will be occurring here. All, send notes!
Terms and aspects:
Commentary is welcome: