• Kenisuomo Luwei

Student thesis: Phd


The study uses an experimental approach in contributing to advance vibration-based fault identification (VFI) techniques for rotor and ball bearing in rotating machines using data fusion of time and frequency domain features. Consequently, it proposes novel VFI approaches which can effectively detect a wide range of rotor and bearing faults in a single analysis. Vibration signals are collected from three test rigs; the flanged-based flexible test rig (FFTR) designed in earlier research, and the spring-based flexible test rig 1 and 2 (SFTR-1 & 2) are improved versions of FFTR. Dynamic characterisation provided the first few natural frequencies of the rigs, which helped in the selection of machines running speed. The FFTR ran below its first critical speed, while the STFR ran below and above its first critical speed. Simulated rotor-related faults included unbalance, misalignment, crack shaft, looseness, and shaft rub, while the simulated bearing fault were cage defects. The preliminary investigation considered only rotor-related faults using acceleration features from vibration signals from FFTR, a build-up from an earlier unified multispeed approach (UMA). Data trending was carried out using selected acceleration features from the time domain; root mean square (RMS), crest factor (CF) and kurtosis (Ku), and frequency domain; 1x – 5x and spectrum energy (SE), and the classification result reaffirms the UMA. Furthermore, acceleration was converted to velocity and similar features were obtained for improved classification. Afterwards, acceleration-based time and velocity-based frequency domain features were fused for improved classification. A proposed novel data fusion of acceleration and velocity features (dFAVF) model helped strengthen the improved approach using signals from STFR-1 with rotor and bearing faults classified in a single analysis. Although spectrum analysis amplitude is helpful in the proposed method, a drawback is the loss of phase during computation due to its complex conjugate. However, the poly-Coherent Composite Bispectrum (pCCB) for rotor fault diagnosis proposed in an earlier study combined different frequency components, which retained their phase during analysis. It also combined multiple sensor data. Therefore, rotor conditions from the SFTR-1 were analysed, and the faults were classified using the first few pCCB components extracted. Additional pCCB components and complex number representations of the components were classified separately and compared. Another proposed novel approach is the fused acceleration-based time domain features with poly-coherent composite bispectrum components (AT-pCCB) model. The valuable result from the pCCB classification of rotor faults led to incorporating bearing features. Thus, developing the AT-pCCB showed improved classification for rotor and bearing faults in a single analysis. Observation of more rotor and bearing conditions and comparing similar machines with varying foundation flexibility, i.e., the SFTR-1 and SFTR-2, using the proposed dFAVF and AT-pCCB models, proved the insensitivity of the proposed model in fault identification in such a scenario. Comparing faulty to baseline conditions for all scenarios showed the quantified classification for effective diagnosis. The distinct separation between the rotor and bearing faults could be from their low and high-frequency ranges. Consequently, these fault identification models have viable prospects for industrial application.
Date of Award31 Dec 2022
Original languageEnglish
Awarding Institution
  • The University of Manchester
SupervisorJyoti Sinha (Supervisor) & Akilu Yunusa-Kaltungo (Supervisor)


  • Vibration-based Fault Identification (VFI)
  • Rotating Machines
  • Rotor and Bearing
  • Data Fusion
  • fault classification

Cite this