Probabilistic Assessment of the Efficiency of Algorithms for Crack Detection on Instrumented RC Beams: a non-Model Based Method

International audienceDamage detection of instrumented structures has known large developments during the last decades. Some of the approaches are called model-based as they lie on a mechanical model of the structure and aim to model and identify the crack characteristics from monitoring. The main disadvantage of this approach for real structures is to model other effects that may change the response of the sensor: temperature, traffic loading, etc. Other approaches are called non-model based methods: they lie on the analysis of changes in the response only. Their main disadvantage is that the identification of the crack characteristics is unavailable. In parallel with the diffusion of monitoring systems, the development of decision support systems (DSS) has been started, that utilize the results of the monitoring data interpretation process for supporting decisions concerning maintenance/rehabilitation planning. The overall process, from data interpretation and damage identification to the definition of the lifetime or fragility curves and finally to decision models is however affected by several uncertainties. The talk will focus on the uncertainties of the damage identification process suggesting a probabilistic measurement of the change in the response through an approach largely used for Non Destructive Testing calibration. The reliability estimate of the damage identification algorithms will be evaluated defining the probability of detection (PoD), the probability of false alarms (PFA) and a receiver operating characteristic (ROC) curve. For a given detection threshold, the couple (PoD, PFA) defines the efficiency of a detection algorithm. The objective is to maximize the probability of detection and minimize the probability of false alarm. The performance of the Proper Orthogonal Decomposition in detecting and localizing damage will be presented, regarding the application to long-term static monitoring data resulting from an experiment conducted in real environmental conditions on two pre-stressed concrete beam specimens. These ROC curves can be obtained by considering two or more algorithms and the same defect range, one technique and two defect ranges, one technique with two settings and the same defect range, or one technique applied in various conditions (even if the testing procedure is rigorously followed during inspection). In the following we will consider two settings (rough data and pre-processed data) and the same defect. The paper will also present a combination of several sensors measurements to illustrate the gain obtained from a redundant monitoring