#TITLE_ALTERNATIVE#

• Main Author: WIJAYA NIM : 23110028; (Dosen Pembimbing : Prof. Dr. Ir. IGN Wiratmaja Puja);, FREDDY
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/15726
• Institution: Institut Teknologi Bandung

Maintenance is an important issue in the industry. There are many different approaches in maintenance. One of them is Condition Based Maintenance (CBM). CBM consists of several stages, i.e., detection, isolation, and identification of faults. The focus of this thesis is in the identification stage. Identification may be performed by analyzing the vibration of machineries. <br /> <br /> <br /> <br /> <br /> In this research work, a novel time-series based pattern-recognition technique is employed [Dianviviyanthi and Nurprasetio, 2000]. The method combines minimum-distance <br /> <br /> <br /> <br /> <br /> and time-series modeling to identify faults in rotating machineries. AutoRegressive Moving Average (ARMA) forms are utilized to model the vibration time series data. Based on the above, feature vector consisting of ARMA model coefficients is formed. The ARMA coefficients are calculated using MATLAB®. Reference feature vectors are obtained from the learning stage, in which data are acquired from machinery possessing specific fault. In this <br /> <br /> <br /> <br /> <br /> thesis, 30 data sets are used in the learning stage to form the reference feature vector. To identify faults in a particular machine, a new feature vector is formed based on the data from the unknown case. The parametric distance between the new feature vector and the reference feature vectors in the data base will determine the existing fault. <br /> <br /> <br /> <br /> <br /> For the laboratory experiment, test rig developed by Bamanto [2010] and Sakti [2011] is utilized. The test rig may simulate rotating-unbalance and unbalance with mechanical looseness. The first attempt of implementing the time-series based method was quite successful for the unbalance case. However, the method showed possibilities of <br /> <br /> <br /> <br /> <br /> misidentification for the combined unbalance and mechanical looseness case. In order to improve the method, wavelet decomposition is implemented. The result shows that upon <br /> <br /> <br /> <br /> <br /> combining time-series modeling and wavelet decomposition, the problem of misidentification can be circumvented. Therefore, the new diagnostic method based on time-series modeling and wavelet decomposition may be recommended to complement the older frequency domain (FFT) based method.