Multivariable control of dynamic structural test systems
Multi-actuator structural testing has traditionally been regarded, from a control point of view, as a multi-loop single-input, single-output problem. This approach does not take into account the interaction between. different actuators, due to the dynamics of the structure under test, which can be considerable. The result of this is often poor laboratory reproduction of the actual service data. This project shows that the mass of the structure under test has a considerable impact upon the stability of the traditional multi-loop, single-input, single-output control system. Where stability is prejudiced, the loop gains have to be reduced to maintain stability and this can degrade the performance of the test. In these circumstances multivariable control offers the potential for a significant improvement in performance. Two experimental rigs are used in this project, both exhibit major interaction and pose a significant control problem. The first rig consists of a laboratory scale cantilever beam excited by two electro-dynamic vibrators with displacements measured by Linear Variable Differential Transformers (L VDTs). The second, industrial-scale, rig consists of a large steel frame excited by two hydraulic actuators with applied force measured by load cells. Multivariable controllers are designed and implemented on these rigs based on the frequency-domain Characteristic Locus method. The multivariable controllers are shown to demonstrate superior performance to traditional multi-loop controllers. Mathematical models of the rigs are not required for controller design, instead experimental frequency responses are all that are needed. This is a major attraction of the Characteristic Locus method since the task.of modelling the dynamics of a multichannel structural test system is not trivial. However, obtaining the frequency response of the second rig is made difficult by the imposition of closed-loop control during the identification experiment. A technique is presented to overcome this problem using an existing correlation method.