Application of information fusion and Shannon entropy in structural damage detection

Vibration-based damage identification is a useful tool for structural health monitoring. But, the damage detection results always have uncertainty because of the measurement noise, modeling error and environment changes. In this paper, information fusion based on D-S (Dempster-Shafer) evidence theory and Shannon entropy are employed for decreasing the uncertainty and improving accuracy of damage identification. Regarding that the multiple evidence from different information sources are different importance and not all the evidences are effective for the final decision. The different importance of the evidences is considered by assigning weighting coefficient. Shannon entropy is a measurement of uncertainty. In this paper it is employed to measure the uncertainty of damage identification results. The first step of the procedure is training several artificial neural networks with different input parameters to obtain the damage decisions respectively. Second, weighing coefficients are assigned to neural networks according to the reliability of the neural networks. The Genetic Algorithm is employed to optimize the weighing coefficients. Third, the weighted decisions are assigned to information fusion center. And in fusion center, a selective fusion method is proposed. Numerical studies on the Binzhou Yellow River Highway Bridge are carried out. The results indicate that the method proposed can improve the damage identification accuracy and increase the reliability of damage identification to compare with the method by neural networks alone.