Influence of measurement errors on structural damage identification using artificial neural networks

The effect of measurement errors on structural damage identification using artificial neural networks (ANN) was investigated in this study. By using back-propagation (BP) networks with proper input vectors, numerical simulation tests for damage detection on a six-storey frame were conducted with measurement errors in deterministic as well as probabilistic senses. The identifiability using ANN for damage location and extent was studied for the cases of measurement errors with different degrees. The results showed that there exists a critical level of measurement error beyond which the probability of correct identification is sharply decreased. The identifiability using the neural networks in the presence of modeling and measurement errors is finally verified using experimental data on a two-storey steel frame.