Fatigue tests for aircraft structural components are indispensable for the series approval of aircraft, but sometimes take years. The reasons for this are, in particular, challenges concerning the control of the servo axes for imprinting the load profiles. For this purpose, a hardware in the loop test bench was set up, consisting of a force-controlled servo cylinder, a load unit, and a real-time capable control system. The challenge is to realize the displacement control of the servo-hydraulic load unit as dynamically as possible to imprint the behavior of the simulated structural component as accurately as possible. To facilitate the design and development of displacement controllers, for imprinting the model response, a hydraulic simulation model was implemented in AMESim and integrated with MATLAB/Simulink to establish a control loop. By comparing the motion trajectories of the simulation model and the physical test bench under PID control, it is determined that the simulation model accurately represents the dynamic behavior of the test bench, thus serving as a suitable platform for evaluating control theories. Through data acquisition and neural network training, a NARMA-L2 neural network controller is obtained. A comparative analysis between the neural network controller and the PID controller revealed that the NARMA-L2 neural network controller achieved superior control performance on the simulation model. In frequency sweep tracking, NARMA-L2 control exhibits smaller phase differences and improves the cutoff frequency compared to PID. NARMA control exhibits excellent robustness, with little impact on the control performance from variations in supply pressure. The controlled system demonstrates good control effectiveness across different operating points.
|