Case-based diagnosis of batch processes based on latent structures

Resumen

The aim of this thesis is to present a methodological approach for the automatic monitoring of batch processes based on a combination of statistical models and machine learning methods. The former is used to model the process based on the relationships among the different monitored variables throughout time, while the latter is used to improve the diagnosis capabilities of the system. Statistical methods do not relate faulty observations with its root cause (they only list the subset of variables whose behaviour has been altered) and they lack of learning capabilities. By using case-based reasoning (CBR) for the diagnosis, faulty observations can be associated with more significant information (like causes). Statistical models also provide a new representation of the observations, on an orthogonal basis, that improves the use of the distance-based approaches of the CBR, giving a better performance