Outils automatiques d'évaluation de la qualité des données pour le suivi en continu de la qualité des eaux usées
|Advisor:||Vanrolleghem, Peter A.|
|Abstract:||Nowadays, in the wastewater field (sewers, water resource recovery facilities -WRRFs, rivers), the monitoring and control of wastewater quality is performed with several on-line sensors. However, a good monitoring strategy should be reliable and provide good data quality. The current fault detection methods have shown that problems such as fouling lead to 10-60 % of the data being discarded. However, helping users in understanding, analysing and processing detected faults (sensors clogging, faulty calibration, suboptimal installation and maintenance) will allow reducing the percentage of data loss and reaching good data on wastewater quality. In this Master thesis, we propose two full workflows allowing the collection of raw data and their transformation into actionable information (i.e. for sensor fault detection, control or process monitoring).The two full modular frameworks were applied to time series data coming from thepilEAUte, bordEAUx and kamEAU projects collected in sewers and WRRFs. These methods have been made more easily applicable by writing Standard Operation Procedures (SOPs) on the use of these methods. In addition, the Matlab scripts are written in a modular way by building different function blocks that are compiled in a toolbox. The first method is a univariate tool composed of two main steps: Data filtering (outlier detection and smoothing) and fault detection. The second method is a multivariate tool using Principal Component Analysis, also composed of two steps: (i) the development of the PCA model and (ii) the fault detection by the PCA. Finally, for the three aforementioned projects, data treatment has led to only 0.1-12% of the data being discarded.|
|Document Type:||Mémoire de maîtrise|
|Open Access Date:||31 October 2019|
|Collection:||Thèses et mémoires|
All documents in CorpusUL are protected by Copyright Act of Canada.