Initial guess generation strategies for spaceplane trajectory optimisation

Trajectory optimisation for spaceplanes is a highly complex problem due to the nonlinearity of the dynamics, long integration times, high energy environment and a broad spectrum of different flight conditions from sea level to space. In this paper, strategies are analysed for the fast and autonomous generation of initial guesses for a gradient-based solver for the ascent trajectory of a multi-stage reusable spaceplane launch vehicle. Different multi-start strategies are used to generate an archive of solutions with the performances analysed for computational run time, convergence rate and violation level of the constraints. A focus is also put on methods that reduce the dependency on the expertise of the user to produce a problem-specific first guess. Different approaches are analysed that introduce a weighting of the constraints relative to the objective function, add low levels of white noise, and conduct an initial sorting using larger integration time steps. A promising compromise between convergence rate, run time and automation is achieved with the introduction of low level white noise to unconverged solutions from a population of first guess solutions created using Latin Hypercube Sampling.