Gestion patrimoniale des réseaux d’assainissement : Impact de la qualité des données et du paramétrage du modèle utilisé sur l’évaluation de l’état des tronçons et des patrimoines

摘要

Asset management ensures that the best decisions are made for elements of the asset in order to reduce risks, optimize performance and minimize costs. A proactive asset management includes development of prioritization schemes for selection of inspection and rehabilitation needs. In this regard, we have identified the following bottlenecks which are addressed in this manuscript. First, there is a need of elaborating inspection programs based on deterioration models in order to be more cost-effective. The influence of availability and quality of data (imprecision, uncertainty, incompleteness) on models’ predictive power has not been studied in depth. We propose two methods to establish the list of the most informative factors from a representative sample of an asset stock. Among other results, this study would suggest that having imprecision in the database is preferable to having a database with no information on one specific factor. Second, once segments are inspected, they should be evaluated by a condition grading protocol. Though various condition grading protocols have been developed, they all fail to undertake the sensitivity of managers and stakeholders to the over or under-estimation of assets’ condition grade as many sources of uncertainty could be found within the condition assessment process. We propose a procedure in order to deal with this uncertainty. We also carry out some sensitivity analyses of parameters employed in this procedure. The results of these sensitivity tests are then applied to a part of the Greater Lyon asset stock. The results show that the assessment of segments into a condition grade depends heavily on the hypotheses that a manager could make about the under or over-estimation of its assets’ condition. Third, at the moment small number of utilities has completely inspected and evaluated their asset stocks. Therefore, the use of a representative sample from an asset stock in order to calibrate decision-support models as deterioration models seems mandatory. Nevertheless, in this regard we should tackle with following problematic issues: 1) How to draw a representative sample of an asset stock which reflects the characteristics of this asset stock? and 2) What is the impact of used sample on the calibration outcomes of these multivariate models? Hence, by drawing several samples with different sizes according to different sampling methods and applying Monte Carlo method, we have proposed a procedure in order to study the influence of available sample on the outcomes of a multivariate model. By proposing some statistical analyses, we showed that the calibration process depends extremely on available sample which could results, if this latter is not drawn properly, into erroneous conclusions.