The need of energy production using renewable resources has been increasing nowadays. Thus, there has been a high investment in wind power machines to increase their quality and capability. The high reparation cost of these machines has shifted the focus of interest of companies and researchers to find effective methods to diagnose and to predict the status of wind power machines. Our research attempts to evaluate and to develop a new diagnostic and prediction system through the implementation and improvement of a dimension reduction technique joined with Mel-Frequency Cepstral Coefficients, a highly used technique in voice recognition. As a tool to diagnose the status, the hidden Markov models are implemented. As a result, the prediction and the diagnosis of the status of the system were successfully detected with a great level of accuracy.