Fault diagnosis of the vibrating beam system via multimodel neural ensembles

This paper develops an effective approach to model design and fault diagnosis for the deflections of the actuated cantilever beam under the regime of repetitive non-linear loads. In particular, the problems of reconstruction of non-linear actuation and linear dynamics of the beam are covered in terms of multi-model neural ensembles with dedicated data fusion rule. Taking advantage of the repetitive character of the control task and available measurement data, the uncertainty related to a potential faults of actuators can be properly addressed using a data-driven iterative learning scheme for modeling and controller synthesis. Then, applying the bank of available models for normal and faulty system states embedded in extension of dedicated observer scheme it is possible to effectively detect and isolate the potential actuator faults. The resulting approach is flexible enough to accurately model the faults occurring on nonlinear actuators and, additionally, take into account the disturbances and noise acting upon the system. The potential of the new design is illustrated via physically relevant real-world application of dynamic displacement control and actuator fault diagnosis for an actuated vibrating beam system.