Erfan Ahadi - Master of Science Student, Iran University of science and technology
Mohammad Riahi - Professor of mechanical engineering, Iran University of science and technology

In recent years, intelligent health monitoring of industrial machines has gained attention of many researchers. Early detection and isolation of mechanical faults in rotating machines can lead to higher performance, less manufacturing costs and increase insafety and reliability. Since 2006 Deep Learning (DE) has been implemented in many intelligent fault diagnosis systems using vibration signals due to its capabilities in handling huge amounts of data and end to end structure without any need to manual featureextraction. In this paper the most popular DL methods are introduced, their applications in fault diagnosis of rotating machines are reviewed and finally experimental tests are conducted to evaluate DL methods performance in comparison with traditionalfault diagnosis techniques. Results prove higher effectiveness of deep leaning methods when facing huge scale of data without any need for condition monitoring expertise. On the other hand, deep learning methods demand high computational power and highknowledge on computer programming. Moreover, if a high quality hand crafted feature set is available, shallow learning methods may perform as good as deep learning methods with lower computational costs and training time

Fault Diagnosis, Artificial Intelligence, Deep Learning, Rotating Machines, Vibration Signals