Abstract
This thesis discusses automatic fault diagnosis in industrial systems with artificial intelligence
techniques, in particular fuzzy and possibilistic reasoning. Initially, the problems
to be solved are presented and strategies to deal with them are suggested in the
scope of artificial intelligence, with special emphasis in fuzzy relational models which
will be the base of the main contribution. Expert systems which combine fuzzy logic
and probability have been also studied, as well as Bayesian networks. Tests with the
mentioned techniques have been carried out. After its evaluation, given the drawbacks
that some of them had, a decision was made to implement a new methodology in order
to improve the existing solutions. This methodology views possibilistic fuzzy diagnosis
as an optimization problem. The methodology converts the linguistic assertions in the
expert system rulebases in a set of linear restrictions from relational techniques. These
equations are used with linear programming code. Some modifications require quadratic
programming. The obtained results in a practical oil analysis application are promising,
presenting easily interpretable outputs and taking into account uncertainty in rules and
measurements.