Online Learning and Control of An Internal Combustion Engine for UAS Using Simplex Tessellation and Recursive Least Squares

As Unmanned Aircraft Systems demand more fuel flexibility, the control of the engines for these systems will need to adapt to unknown fuels. To do so, a computationally efficient method for the online learning and adaptive control of an engine based on real-time input and output engine measurements is developed. The method, based on recursive least-squares estimation and multi-dimensional piecewise-linear splines, has been developed for systems with one input (injection timing), two inputs (injection timing, glow-plug power/fuel mass), as well as for general systems with arbitrary dimensions. The online learning model in turn generates an adaptive feedforward signal which is combined with an integral feedback with decoupling control to achieve a desired combustion phasing (CA50) and other outputs such as mean effective pressure (MEP) or power.