点击次数：

发布时间：2019-03-10

论文类型：期刊论文

发表时间：2008-08-01

发表刊物：MECHANICAL SYSTEMS AND SIGNAL PROCESSING

收录刊物：Scopus、EI、SCIE

卷号：22

期号：6

页面范围：1427-1440

ISSN号：0888-3270

关键字：damage identification; information fusion; Dempster-Shafer evidence theory; Shannon entropy; artificial neural networks

摘要：This paper proposes a damage identification method based on the combination of artificial neural network (ANN), Dempster-Shafer (D-S) evidence theory-based information fusion and the Shannon entropy, to form a weighted and selective information fusion technique to reduce the impact of uncertainties on damage identification. The initial damage decision is first made by several individual ANNs with different inputs. Considering that damage identification accuracy is dependent on different ANNs, optimal weighting coefficients obtained by genetic algorithm (GA) are assigned to ANNs. The decision obtained from the ANN with the largest weighting coefficient is the most reliable. Damage identification-based D-S evidence theory is carried out by combining the decision of the ANN with the largest weighting coefficient with that of the ANN with the second largest weighting coefficient. Next, Shannon entropies of the decisions of the ANN with the largest weighting coefficient and that obtained by the information fusion are calculated to measure the uncertainty level of the decisions, respectively. The decision with smallest entropy remains in the next information fusion operation because this decision has less uncertainty. The decision with smaller entropy will be combined with the decision of the ANN with the third largest weighting coefficient. The operation is repeated until the last ANN with the minimum-weighting coefficient is fused. Numerical study on the Binzhou Yellow River Highway Bridge was carried out to validate the accuracy of the proposed damage identification method. The simulated results indicate that the damage identification method based on the weighted and selective information fusion technique can improve damage identification accuracy in comparison with ANN alone and direct information fusion techniques when the measurement noise level is the same. (C) 2007 Elsevier Ltd. All rights reserved.