Technological developments had represented a challenge for the industrial in order to improve the machinery and the productive systems. Therefore, the importance of maintenance area has increased because it allows holding the systems health and its availability. Nonetheless, a diagnostic task requires expertise in the specific area of the diagnostic application, and highly trained and skilled personnel are needed. Thereby, automatic pattern recognition is highly desirable, and it can be achieved by classification of signals based on the information or features extracted from mechanical vibration signals. In this doctoral thesis is proposed a methodology to implement an automatic diagnostic system including several factors such as time-varying operating conditions, the unavailability of data with damaged machine and the identification of the faulty mechanism.