THE DEVELOPMENT OF A NOVEL DIAGNOSTIC TECHNIQUE BY COMBINING MINIMUM-DISTANCE PATTERN-RECOGNITION AND DISCRETE WAVELET TRANSFORM

• Main Author: HILARIUS TUTUT SANDEWAN (NIM : 23108318); Pembimbing : Ir. Ign. Pulung Nurprasetio, MSME
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/15947
• Institution: Institut Teknologi Bandung

This thesis is focused on the development of a novel method for machinery diagnostics. The main platform is derived from pattern recognition technique called minimum distance method. In this respect, the method starts by forming the feature vector. Fault identification is performed based on the distance between the measured feature vector and the feature vectors in the data base. The main contribution of this thesis is the formation of a novel diagnostic technique which is formed by combining the minimum distance pattern recognition and decimated discrete wavelet transform (DWT). The feature vector is assembled from the wavelet filter coefficients. To show the effectiveness of the method, a numerical simulation is performed. In this case, two standard faults are simulated, namely the unbalance and mechanical looseness. The simulation shows promising results, which lead to further implementation using the available test rigs in the laboratory. As an interim result, the thesis also elaborates denoising technique using DWT. The main objective is to remove measurement noise from the output of vibration sensor. The old way of noise removal is through averaging, in the case of white noise, or filtering, in the case of colored noise. However, both the above procedure will distort the original signal. As a result, the signal becomes more difficult to analyze. Unlike filtering or averaging, DWT will not distort nor change the phase of the processed signal. DWT will simply decompose the signal into several allocated frequency bands within the measurement range, i.e., from 0 Hz up to the Nyquist frequency. The number of bands or intervals depends on the DWT levels. Since the signal has been decomposed, one may leisurely pick the component within the preferred frequency band. Using the above approach, DWT has been successfully implemented at the Mechanics and Mechanical Construction Laboratory, FMAE – ITB to remove grounding noise (50 Hz) from the measured signal.