Self-constructing wavelet neural network algorithm for nonlinear control of large structures

Abstract

An adaptive control algorithm is presented for nonlinear vibration control of large structures subjected to dynamic loading. It is based on integration of a self-constructing wavelet neural network (SCWNN) developed specifically for structural system identification with an adaptive fuzzy sliding mode control approach. The algorithm is particularly suitable when the physical properties such as the stiffnesses and damping ratios of the structural system are unknown or partially known which is the case when a structure is subjected to an extreme dynamic event such as an earthquake as the structural properties change during the event. SCWNN is developed for functional approximation of the nonlinear behavior of large structures using neural networks and wavelets. In contrast to earlier work, the identification and control are processed simultaneously which makes the resulting adaptive control more applicable to real life situations. A two-part growing and pruning criterion is developed to construct the hidden layer in the neural network automatically. A fuzzy compensation controller is developed to reduce the chattering phenomenon. The robustness of the proposed algorithm is achieved by deriving a set of adaptive laws for determining the unknown parameters of wavelet neural networks using two Lyapunov functions. No offline training of neural network is necessary for the system identification process. In addition, the earthquake signals are considered as unidentified. This is particularly important for on-line vibration control of large civil structures since the external dynamic loading due to earthquake is not available in advance. The model is applied to vibration control of a continuous cast-in-place prestressed concrete box-girder bridge benchmark problem seismically excited highway.