Abstarct: Abstract In this thesis, first, the methods of modeling valve stiction are examined, and after that, an applied algebraic model obtained from data-driven models is presented. After choosing an optimal model, a brief discussion about the wide range of works performed regarding stiction detection and diagnosis is discussed. The first proposed method for detecting stiction in a non-invasive and model-free has been presented by the extension of a support vector machine to the generalized statistical variables to detect the occurrence of stiction independent of process dynamics. Finally, by developing the obtained knowledge and using the machine learning method on statistical variables, a multi-step algorithm is presented that is capable of automatically detecting and classifying oscillating factors in the control loop as the second method. In this method, after training the classifiers of the support vector machine, Bayes, and k-nearest neighbors, the best classifier is evaluated and selected, and the decision tree specifies the probability of occurrence of each oscillation factor in the loop. In this method, two statistical variables T^2 and SPE have been used to increase the features. The effectiveness of this method has been demonstrated by applying it to benchmark industrial data.