The use of Mann iteration in the stable inversion of NARX models that have
been converted to state space form is investigated to either recover the convergence
or improve the accuracy of the best approximate solution under conditions
when Picard iteration fails to converge. Attention is given to the use of filtering
and time-varying iteration gains. The results are potentially of use in response
reconstruction for fatigue testing purposes where the inverse of a NARX model,
obtained by system identification, may be used to achieve the reconstruction.