Safe and reliable operation of the systems relies on the use of online condition monitoring and diagnostic systems that aim to take immediate actions upon the occurrence of a fault. Machine learning techniques are widely used for designing data-driven diagnostic models. The training procedure of a data-driven model usually requires a large amount of labeled data, which may not be always practical. This problem can be untangled by resorting to semi-supervised learning approaches, which enables the decision making procedure using only a few numbers of labeled samples coupled with a large number of unlabeled samples. Thus, it is crucial to conduct a critical study on the use of semi-supervised learning for the purpose of fault diagnosis.

Another issue of concern is fault diagnosis in non-stationary environments, where data streams evolve over time, and as a result, model-based and most of the data-driven models are impractical. In this work, this has been addressed by means of an adaptive data-driven diagnostic model.