ICAISC '08 Proceedings of the 9th international conference on Artificial Intelligence and Soft Computing
Application of semi-Bayesian neural networks in the identification of load causing beam yielding
ICANN'10 Proceedings of the 20th international conference on Artificial neural networks: Part I
| Hi-index | 0.00 |
A hybrid computational system, composed of the finite element method (FEM) and cascade neural network system (CNNs), is applied to the identification of three geometrical parameters of elastic arches, i.e. span l, height f and cross-sectional thickness h. FEM is used in the direct (forward) analysis, which corresponds to the mapping @a={l,f,h}-{@w"j}, where: @a - vector of control parameters, @w"j - arch eigenfrequencies. The reverse analysis is related to the identification procedure in which the reverse mapping is performed {@w"j}-{@a"i}. For the identification purposes a recurrent, three level CNNs of structure (D^k-H^k-1)"s was formulated, where: k - recurrence step, s=I, II, III-levels of cascade system. The Semi-Bayesian approach is introduced for the design of CNNs applying the MML Maximum Marginal Likelihood) criterion. The computation of hyperparameters is performed by means of the Bayesian procedure evidence. The numerical analysis proves a great numerical efficiency of the proposed hybrid approach for both the perfect (noiseless) values of eigenfrequencies and noisy ones simulated by an added artificial noise.