Tesis doctoral de Magda Liliana Ruiz Ordóñez
This thesis focuses on the monitoring, fault detection and diagnosis of wastewater treatment plants (wwtp), which are important fields of research for a wide range of engineering disciplines. The main objective is to evaluate and apply a novel artificial intelligent methodology based on situation assessment for monitoring and diagnosis of sequencing batch reactor (sbr) operation. To this end, multivariate statistical process control (mspc) in combination with case-based reasoning (cbr) methodology was developed, which was evaluated on three different sbr (pilot and lab-scales) plants and validated on bsm1 plant layout. Results showed that, mpca is a robust technique for monitoring and fault detection of sbr operation. The mpca was successfully tested for on-line (real-time) monitoring of pilot scale-sbr performing nitrogen removal – the first time this is achieved (to our best knowledge). The mpca methodology is now ready to be used as part of daily operation of the sbrs. For the diagnosis part, a comprehensive evaluation of the cbr methodology for automatic diagnosis of sbr process operation (biomath and lequia) was performed – the first time an artificial intelligent method applied within wwtps. The methodology was then tested on the bsm1 plant layout which were used to construct abnormal events, e.G. Faults, sensor failures, etc. The cbr method used input from the mpca (rather than raw process data) and the best descriptors for the assessment of the situation (cases) were found to be principal components and errors (q) of the statistical model. The main results showed that the cbr successfully diagnosed a wide range of operational problems such as sludge bulking, influent inhibition/toxicity, high influent flow and sensor faults. The diagnosis performance of cbr method using several statistical extensions such as mpca, dynamic pca and pca were also studied. This comparison showed that the mpca + cbr combination has a good diagnosis performance. However, a more theoretical and in-depth study of which inputs and descriptors to use for the situation assessment step in the cbr are needed to further improve the diagnosis. In addition, the ability of cbr to maintain and update the knowledge was also studied and tested successfully using drop and ib family of algorithms. This showed that repeating the cycle of learning helps maintaining and updating the case-base of the cbr. Overall, this adaptive and intelligent aspects of the method makes it a good candidate for helping the management in the daily plant operation as an automatic diagnosis and real-time warning tool. Such artificial intelligent methods are promising tools which has the potential to contribute to good management and operation of plants. Further research is, however, needed to improve and consolidate the application of cbr to wwtp operations, including input descriptors, retrieve and update algorithms and decision making rules. All in all, this is expected to save operational costs as well as improve plant performance to comply with the goals of urban water management.
Datos académicos de la tesis doctoral «Multivariate statiscal process control and case-base reasoning for situation assessment of sequencing batch reactors«
- Título de la tesis: Multivariate statiscal process control and case-base reasoning for situation assessment of sequencing batch reactors
- Autor: Magda Liliana Ruiz Ordóñez
- Universidad: Girona
- Fecha de lectura de la tesis: 16/06/2008
Dirección y tribunal
- Director de la tesis
- Jordi Colomer Feliu
- Tribunal
- Presidente del tribunal: Alberto josé Ferrer riquelme
- ulf Jeppsson (vocal)
- gurkan Sin (vocal)
- jean-philippe Steyer (vocal)