Fault analysis and detection from vibration signals in rotating machines with signal processing and machine learning methods

In this thesis, studies were carried out to determine the faults occurring at the drill bit end of the CNC machine, which is frequently used in the industry. Failure is inevitable for each machine used in production. Prediction of failures before they occur prevents both loss of manpower and high cost of repair. For this reason, analyzing faults in machines is a popular subject that has been applied for many years. When the previous studies are examined, the quantities such as voltage, current, temperature and vibration taken under the working conditions of the machines are used. In this study, a dataset containing the vibration signals measured during the operation of the CNC machine is used. In order to diagnose faults, the features have been extracted of the raw vibration signals. The feature extraction process allows obtaining a small amount of data which is valuable from a large amount of data. In this context, features were obtained in 4 areas which are: time domain, frequency domain, time-frequency domain and the combination of these. After feature extraction, normalization was performed. In this way, the features with different values can be processed efficiently at the same time. After the normalization process, the Long-Short Term Memory structure and various deep learning layers were realized. As a result of the transactions 99,53% accuracy rate was obtained. In order to compare the obtained accuracy rate Support Vector Machines were used because of Support Vector Machines are frequently used in this area. In this study, it is shown that fault analysis can be performed on machines with Long-Short Term Memory structure which is not so used in this field. In addition, the effects of adding different layers to the generated artificial neural network are shown.