In the era of modern industries, use of advanced machines has increased the productivity of plants. Defect of such complex machines can increase the downtime and hence reduce productivity and ultimately the profit. Proper maintenance of systems is very crucial to ensure trouble free operation of machines. Predictive Maintenance techniques are those techniques which are used for identification of fault in its early stage. This avoids the sudden unavailability of machines and reduces the downtime. Predictive Maintenance techniques using vibration analysis is one of the very popular methods of Condition Monitoring. These techniques are evolving rapidly due to their efficient detection of fault in its incipient stage. However, non-stationary of the vibration signal produced particularly in belts in belt drive systems using conventional techniques of signal processing poses challenges. In this project work, the application of advanced signal processing techniques is investigated for the automatic fault detection in belts. This project work is focused on developing a novel concept of vibration signal processing for the automatic fault detection in belts using wavelet analysis and the belt condition classification and prediction using Machine Learning algorithms.