A Comparison of Neural Networks and Model Updating Methods for Damage Localization

摘要: In this paper a comparison between classical Model Updating methods and Neuronal Network for damage localization is presented. Both are based in modal parameters have been applied to the same experimental problem. The main objective of research establish comparatives qualities each method. consolidated method detection, but it long process that requires considerable computational cost. Artificial Networks (ANN) faster consequently they require less cost if neural network properly chosen trained. A laboratory test beam supported four points has developed order apply Finite Element Update detection. first step made by means Iterative Methods. second small introduced with both Neural methods. full numerical model update developed. Guyan reduction Independence Effective used select best transducers position. Correlation, validation adaptation data techniques used. are: reduction, Dynamic Improved Reduction System (IRS) SEREP reduction. correlation Modal Assurance Criterion (MAC), Coordinate (COMAC) Modified (MACMO). Also parameterization, sensibility analyses done before final updating. Different penalty functions depending number update. ANNs trained results obtained simulations Method. These introducing different beam. Ones ANN trained, evaluated as tool predict position natural frequencies Paper 49