Shape estimation from incomplete measurements: a neural-net approach

Accurate estimation of the shape of precision space structures is the key to the success of spaceborne optical systems. A combined simulated annealing and neural-network approach is proposed whereby one can infer the current deformed state of the structure from a limited number of on-board measurements. The approach is especially effective when most of the computations must be done on-ground or off-line, and only minimal calculations are allowed for near real-time on-board processing. It is shown that the performance of the network and its ability to estimate the shape accurately is highly dependent upon the off-line training and tuning of the model to a specific family of expected disturbances. Details of the methodology and results of numerical simulations are given for various on-board estimation scenarios.