Seminaris d'alumnes de doctorat de les Universitats Valencianes.
EXPLORING NON-LINEAR RELATIONSHIPS IN DATA: AN OVERVIEW OF KERNEL-BASED METHODS FOR INDUSTRIAL BATCH PROCESS ANALYSIS
Nowadays, most of the manufacturing industries in the world perform batch processes in their plants. To guarantee and preserve high quality of the final products and to minimize the number of off-specification process runs, batch monitoring schemes are designed starting from historical data so that faults and failures might be quickly, easily and efficiently recognized and their possible root causes might be correctly identified. These two phases of process monitoring are also known as fault detection and fault diagnosis, respectively. The most widely used techniques to build the aforementioned process monitoring schemes are Principal Component Analysis (PCA) and Partial Least Squares regression (PLS). However, when resorting to PCA or PLS, if the data under study are affected by complex non-linear relationships, their analysis and interpretation may be seriously jeopardized, since both assume their structure is linear. In this circumstance, a good alternative is represented by the so-called kernel-based techniques, which also comprehend support vector machines and have already been broadly used in chemistry, biology, informatics and continuous process monitoring. The main goal of this work is to explore the potential of these methods for monitoring industrial batch processes. To this end, simulated and real industrial process datasets are analysed to check the effectiveness of the described methodology in this field.
Departament d'Estadística i Investigació Operativa Aplicades i Qualitat
Universitat Politècnica de València
Dijous, 12 de Març
12:30h, Saló de Graus.
o dirigirse a email@example.com
Data d'actualització: 4 de maig de 2012 12:28.
Llista de notícies