Please use this identifier to cite or link to this item:
http://hdl.handle.net/10603/221422
Title: | Gear Box Vibration Analysis Using Machine Learning Methods for Fault Diagnosis |
Researcher: | Saravanan . N |
Guide(s): | Ramachandran K.I |
Keywords: | Engineering and Technology Vibration analysis Gear box Machine learning methods Gear fault diagnosis Mechanical Engineering |
University: | Amrita Vishwa Vidyapeetham (University) |
Completed Date: | 5/02/2010 |
Abstract: | This thesis is about the use of machine learning methods for gear box vibration analysis for condition monitoring. Condition monitoring is already much practiced in many of todayand#8223;s engine rooms and plants, either by skilled engineers or diagnostic expert systems. However, techniques that rely on automatic pattern recognition have only recently been introduced into this field. Pattern recognition is a research area with a long-standing history, traditionally focused on finding optimal decision functions for static well-sampled classes of data. Besides issues encountered in any pattern recognition problem (feature extraction, small sample sizes, generalization), we face some special issues in condition monitoring of rotating equipment. This requires the use of (relatively novel) methods for blind source separation, novelty detection and dynamic pattern recognition. The knowledge of the condition of a machine may be obtained by selecting a suitable index and monitoring its value at regular intervals. With measured data (signal) one can do trend monitoring, condition checking and fault diagnosis. Fault diagnosis of bevel gear box using vibration signals was taken up for detailed study and forms the main theme of this research work. Fault diagnosis includes the methods such as shock pulse method; wear debris analysis, sound and acoustics emission and vibration analysis. Again each method has its own area of applications with pros and cons. Amongst these conventional techniques, fast Fourier transform (FFT) stands out and is widely used in industries. Characteristic frequencies are used as basis for FFT based techniques; characteristic frequency is a function of the speed of the shaft. Further, there are several other factors that interfere with the vibration signals. FFT based techniques are sensitive to such noises. Clearly, there is a need for an automated fault diagnosis technique with fault tolerance capability. Machine learning techniques seemed to be a candidate fulfilling these requirements... |
Pagination: | XVIII, 243 |
URI: | http://hdl.handle.net/10603/221422 |
Appears in Departments: | Department of Mechanical Engineering (Amrita School of Engineering) |
Files in This Item:
Items in Shodhganga are licensed under Creative Commons Licence Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).
Altmetric Badge: