Optimal tuning for an abrupt change detection algorithm: Application to an underground gallery structure health monitoring
Abstract
The aim of this paper is to detect faults such as cracks in an underground structure to ensure its health monitoring. The proposed approach is based on an ARMAX modeling of a deformation sensor (a vibrating wire gauge), to generate a residual. This residual is then used by a statistical test, the dynamic cumulative sum, to generate a detection function which is compared to a threshold for decision. The main contribution of this paper is the use of the receiver operating characteristic curve to optimally tune the parameters of the statistical test and to have the best fault detection/false alarm rate. The originality of this work is that the receiver operating characteristic curve is used for two aims: First, to optimally tune the sliding windows width of the statistical test and second, to precisely tune the decision threshold. The effectiveness of the proposed method is highlighted on the application of an underground gallery structure health monitoring.