Toward a monitoring of storage facilities for radioactive waste

Document available in extended abstract form only. ANDRA's CIGEO project aims at proposing safe long-term solution for radioactive intermediate-level long-lived waste and high-level waste at a depth of 500 meters. It will be designed to accommodate the wastes permanently. In these circumstances, the structure health monitoring must be ensured. This is the aim of the instrumentation and processing technique. This study strives to solve two major issues in order to implement a Fault Detection and Isolation system for structures health monitoring: the sensors positioning and the data processing. Proposed Approach Two types of diagnosis approaches can be distinguished: with or without model. Based model approaches need a good knowledge of the model to have a reliable diagnosis. In case of complex or large system, the second type of methods is preferably used. In this work, the complexity of physical models available prompted us to compute behavioral models used in signal processing. The objective is to extract fault indicators (residuals for example) for reliable diagnosis. Subsequently, ARMAX (Auto Regressive Moving Average with external inputs) model is used to find the local parameters and their evolution. The objective is to predict measured variables analytically. In order to test the method, we selected the measurements performed by a vibrating wire named DFO2018-2. This vibrating wire is a displacement sensor installed in the concrete of the main shaft of Bure's URL at -455 m (more precisely on the SMGR1). As an external input, we selected the temperature measurement provided by the weather station; the temperature is a good external input affecting the sensor measurements and not perturbed itself by the system observed. The ARMAX model parameters are estimated over 1 year. The prediction is done for the following years. If measurements deviate from the model, we suspect a fault in the system behavior. Improvements of designing the model will be developed for this case in particular to minimize the noise and optimize automatically A(q) and B(q). The ARMAX will be then applied to another case of studies: the ORS (observation of liners and supporting structure) measurements