Name: MARCUS VINICIUS BATISTA MOUTINHO
Publication date: 09/08/2016
Advisor:
Name |
Role |
|---|---|
| CELSO JOSE MUNARO | Advisor * |
Examining board:
| Name |
Role |
|---|---|
| CELSO JOSE MUNARO | Advisor * |
| THOMAS WALTER RAUBER | External Examiner * |
Summary: Process monitoring methods using statistical approaches assume that the data have a normal distribution. Moreover, many of these techniques require that the plant operation remains in the same region, resulting in the generation of a large number of false alarms if not fulfilled. In this work, a robust data clustering technique is used for treating plants operating in multiple operating points. The methodology is applied to fault detection in plugs of control valves, which belongs to this class of problem. Furthermore, the fault is considered difficult to detect due to the difficulty of installing sensors. For evaluation of
the methodology, the benchmark DAMADICS was used. The clustering technique presented has the ability to handle a certain percentage of outliers in data that may arise, including in transient state. This feature optimizes the step of pre-processing of data. A comparison with the traditional method (no clustering) is performed highlighting its main features and superiority.
