Automatic time stepping algorithms for implicit numerical simulations of non-linear dynamics

Variable step strategies are specially well suited to deal with problems characterized by high non-linearity and contact/impact. Both phenomena are typical of dynamic simulations of the interactions between a turbine blade and its casing, the most dramatic example being blade loss. Constant step size strategies do not give satisfactory answer for this kind of problems, since it is very difficult, if not impossible, for the user to find an appropriate time step that does not lead to divergence nor generate extremely costly computations. An automatic time stepping algorithm is proposed, which takes into account the recent history of accelerations in the bodies under consideration. More precisely, the adaptation algorithm is based on estimators of the integration error of the differential dynamic balance equations. This allows for adaptation of the time step to capture correctly the transient phenomena, with characteristic times which can range from relatively long (in regime) to very short (blade loss), thus ensuring precision while keeping the computation cost to a minimum. Additionally, the proposed algorithm
automatically takes decisions regarding the necessity of updating the tangent matrix or stopping the iterations, further reducing the computational cost. As an illustration of the
capabilities of this algorithm, several numerical simulations of both academic and industrial (the contact/impact between a turbine blade and the casing) problems will be presented.