DEVELOPMENT OF TIME SYNCHRONOUS AVERAGING METHOD INTEGRATED WITH MACHINE LEARNING METHOD TO MINIMIZE THE NUMBER OF AVERAGING

• Main Author: Nasirudin, M.
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/72045
• Institution: Institut Teknologi Bandung

The vibration signals obtained from measurement usually contain noise that can reduce the accuracy of balancing rotating machinery. One of the methods to reduce the effect of asynchronous noise signal with the reference is the Time Synchronous Averaging (TSA) Method. Unfortunately, the ability of the TSA Method to reduce noise is proportional to the number of averaging. The more the number of averaging, the better the ability of the TSA Method to reduce noise. This means that a large number of averaging are required to effectively reduce noise. Consequently, a large amount of measurement data is needed so that the vibration measurement process requires a long time and a large data storage memory. Meanwhile, the Machine Learning (ML) Method is a method that has been widely applied in various fields, including in the field of engineering maintenance, because this method could learn from a statistical data based on specific algorithm. Therefore, this research aims to develop the TSA Method integrated with the ML Method to minimize the number of averaging. This method is then applied to a case of vibration due to unbalanced mass in a two-stage rotor system. This research began with transfer measurement data consisting of vibration signals and keyphasor signals from the research of Sumartoyo. The measurement data was recorded from vibration due to unbalanced mass in a two-stage rotor system. The measurement data was then processed using the TSA Method. Then, the vibration signal resulting from the TSA Method was extracted into two sets of data, namely the training data set and the testing data set. The training data set was used in the ML Method training stage to obtain the ML Method Model. The ML method used was the Regression Method and the Regression Method algorithms used were Decision Tree (DT), Support Vector Machine (SVM), and Random Forest (RF). Furthermore, the obtained ML Method Model was tested with the testing data set to determine whether this method has successfully minimized the number of averaging. The result of the ML Method training showed that all three algorithms provided Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) values less than 1. This indicated that the predicted response values are close to the actual response values. However, the results of the ML Method testing showed that the RMSE and MAE values are higher compared to the training results. This is due to testing process which involves extrapolation in the ML Method. Furthermore, the predicted results using the ML Method are used to construct a 400-time averaged vibration signal. The resulting 400-time averaged vibration signal closely approximates the vibration signal from the TSA Method with a 400-time average.