Neural Network Limitations on the Model Updating Application

The primary benefits from finite element model updating with neural networks are to avoid the common problem of coordinate incompleteness and robustness in the presence of noise. If, however, this is the only benefit, then it would be questionable whether or not the use of neural network for model updating would be worth the effort. For model updating application, the actual cost of the neural network is time. This requires the time for creating the training pairs and the training of the neural network. In this study, the ability of neural network for model updating is determined by two experimental modal testing. As a result of present paper, neural network is only suitable for models with small number of updated parameters such as classical model or joint identification problems.