Development of sensor-based machinery vibration assessment system for effective faults diagnostic in condition based maintenance using machine learning / Syed Mohd Syafiq Syed Mahmud

Syed Mohd Syafiq, Syed Mahmud (2022) Development of sensor-based machinery vibration assessment system for effective faults diagnostic in condition based maintenance using machine learning / Syed Mohd Syafiq Syed Mahmud. Masters thesis, Universiti Malaya.

	PDF (The Candidate’s Agreement) Restricted to Repository staff only Download (711Kb)
	PDF (Thesis M.A) Download (1737Kb)

Abstract

Rotating machines such as turbines, motors, pumps, and fans generally generate the vibration when operate under normal condition. However, the presence of excessive force within the component of rotating machine can produce the high level of vibration. Without a systematic monitoring system, a high vibration can accelerate the machine wear, consume excess power, and damage the equipment. Consequently, the machine requires to shut down for maintenance and resulting unplanned downtime and increase the cost of maintenance. Furthermore, high level of vibration also can affect the machine performance and even worse affect the safety problems in engineered systems. High vibration caused by imbalance and misalignment become a major issue and leading to component damage such as shaft, seals, coupling and bearings in various rotating engineering application. Misalignments occurred when rotational centerlines are not collinear while imbalance is occurred when shaft geometric and mass centerline are not coincided. In a traditional way, the vibration data are present in plot such as orbit, time waveform, spectrum etc. However, to analysis and diagnose the vibration fault required a skilled worker who has vast experience, high technical knowledge, and expertise in vibration analysis. Furthermore, to diagnose the vibration fault would consume more time before final maintenance decision can be made. Therefore, the advance monitoring system that can diagnose the machinery fault is developed to supervise machine condition effectively. The system is build based on sophisticated deep learning program language by using modern python program to provide an assist and support in supervised the trend, performance of the machine and provide recommend decision. By using Convolutional Neural Network (CNN) method in deep learning offer a higher accuracy in prediction and detecting the vibration faults. The system is develop using 1D Convolutional Neural Network that can provide very simple in structure, easy to understand and flexible in design. Based on the validation result, the system successfully to achieve 100% accuracy in predict and detection each vibration faults correctly. In addition to this, the system is assessed with difference type of input data and the system achieve low accuracy when reducing the training data and using unprocessed data. However, this method required a huge number of training and validation dataset in order to get a better accuracy in predict and detecting the vibration faults. The fewer dataset could provide a result in a poor approximation and subsequently produce an incorrect interpretation of vibration faults.

Actions (For repository staff only : Login required)