ABSTRACT
As was already emphasized (cf. Section 2.4.4), one of the main difficulties associated with optimization of sensor locations is the dependence of optimal solutions on the true values θtrue of the parameters to be estimated. Since these values are unknown, an obvious and common approach is to use one of the locally optimal designs described in previous chapters for some prior estimate θ0 of θtrue in lieu of θtrue itself (it can be, e.g., a nominal value for θ or a result of a preliminary experiment). But θ0 may be far from θtrue and, simultaneously, properties of locally optimal designs can be very sensitive to changes in θ [92]. Such prior uncertainty on θ0 is not taken into account by any optimization procedure to determine local designs and an experimental setting thus obtained may consequently be far from optimal. This has even raised some doubts among experimenters about the practical use of nonlinear experimental design at all [348].
