Presents a numerical study related to the application of a neural network or neural observer for state estimation purposes in the overall problem of active control of structural vibrations. The application of this type of estimation structure presents several advantages relative to classic approaches, such as any type of observer or Kalman filter. The results related to the training session and to the numerical implementation of the state estimation for a vibrating cantilevered aluminum beam are presented. The influence of different parameters, such as learning rate, number of hidden layers, number of processing elements for each hidden layer, and size of the input data pattern on the training section is shown.
展开▼
