Application of optimal iterative learning control to the dynamic testing of mechanical structures

Abstract

Dynamic testing of mechanical components is carried out across a variety of industries to assess fatigue life or to facilitate optimal design. Such machines are predominantly hydraulically actuated and systems range from single-channel testing of small components to multichannel testing of motor vehicles and airframes. A typical requirement is that in-service loads or displacements are replicated as closely as possible on the tested structure. Such a control problem cannot normally be solved with standard feedback control and it is common to employ iterative methods whereby the actuator command signals are modified in a series of repetitive trials. The industry standard approach, the so-called inverse algorithm, does not always converge to a suitable solution. In this paper an alternative iterative approach, derived using optimization methods, is presented and shown to have superior robustness properties. This is demonstrated through the use of simulation studies and application to both laboratoryscale and full industrial-scale test rigs. Some key theoretical results are also presented that provide important support for the experimental results. In practice, the robustness properties of the new algorithm can enable the required accuracy to be obtained in a greatly reduced number of iterations in relation to the conventional approach, thereby significantly accelerating the test commissioning process.