Multistage artificial neural network in structural damage detection

Abstract

This study addressed two main current issues in the area of vibration-based damage detection. The first issue was the development of a pragmatic method for damage detection through the use of a limited number of measurements. A full set of measurements was required to establish the reliable result, especially when mode shape and frequency were used as indicators for damage detection. However, this condition is usually difficult to achieve in real-life applications. Hence, in this study, a multistage artificial neural network (ANN) was employed to predict the unmeasured data at all the unmeasured point locations to obtain full measurement before proceeding to damage detection. The accuracy and efficiency of the proposed method for damage detection was investigated. Furthermore, the sensitivity of the number of measurement points in the proposed method was also investigated through a parametric study. The second issue was the integration of the uncertainties into the proposed multistage ANN. The existence of uncertainties is inevitable in practical applications because of modelling and measurement errors. These uncertainties were incorporated into the multistage ANN through a probabilistic approach. The results were in the means of the probability of damage existence, which were computed using the Rosenblueth’s point-estimate method. The results of this study evidenced that the multistage ANN was capable of predicting the unmeasured data at the unmeasured point locations, and subsequently, was successful in predicting the damage locations and severities. The incorporation of uncertainties into the multistage ANN further improved the proposed method. The results were supported through the demonstration of numerical examples and an experimental example of a prestressed concrete panel. It is concluded that the proposed method has great potential to overcome the issue of using a limited number of sensors in the vibrationbased damaged detection field.