AbstractThe centrifugal pump is a primary agent in process manufacturing. The condition of any one of its component parts can affect its operations, and effective condition monitoring (CM) is necessary to avoid unpredicted failure, minimize disruption costs and improve machine accessibility.
The primary purpose of this research is to develop effective CM tools for centrifugal pumps, which can be easily deployed in practice. To this end two measurements: surface vibration and airborne acoustics are selected as data sources due to their being information rich in the dynamics of pump operations, especially the acoustic signal which can be obtained remotely with minimal disturbance of production lines. These information-rich datasets are adversely affected by noise contamination, so effective data processing tools need to be used. A gap found in present applications is the use of the modulation signal bispectrum (MSB) to improve fault detection due to its capability of suppressing random noise while retaining phase information, enhancing its ability to detect nonlinear components. A second gap was the lack of a structured comparison of the detection of faults in a centrifugal pump using its acoustic and vibration signals.
An analytic study has revealed that both vibration and acoustic data contain non-linear interactions between deterministic excitations such as impeller rotation, bearing defects and random hydraulic flow and system resonances. As a result, the data can be modelled as the modulations between these mechanisms, and demodulation tools such as MSB can usefully be used to diagnose characteristic features, based on experimental studies. Using MSB and comparing it with more traditional methods such as envelope analysis, and the power spectrum showed both airborne acoustic signals, and pump surface vibrations could be used to correctly detect and diagnose bearing faults, mechanical seal defects, and impeller blade wear seeded into the test pump.
It is demonstrated that both acoustic and vibration signals are capable of providing sufficient information to detect seeded bearing defects when using MSB analysis to remove random noise and improve modulation component detection. It was confirmed that remotely measured airborne acoustic signals are effective tools for monitoring pump health and detecting and distinguishing between inner and outer race faults and that the acoustic signals could provide better differentiation between healthy and defective cases than vibration signals when investigating inner race faults. However, in the case of the outer race fault, the vibration signal provided greater separation of baseline and fault states. Both the airborne sound and vibration signals generated by a seeded mechanical seal defect contained sufficient information to detect the presence of the fault. For fault detection, MSB and power spectrum analysis-based methods were used separately. The results show that harmonics of the shaft drive frequency (48.3 Hz) in the acoustic signal allowed for easier fault detection than the spectrum. The MSB of the airborne sound provided good separation between baseline and fault harmonics over a wide range of flow rates.
The MSB and power spectrum were also used to detect the seeded impeller faults at different flow rates. The experimental results show good separation of blade pass frequency between healthy and faulty cases for both the MSB analysis and the power spectrum of the acoustic signal. However, the harmonics of the shaft frequency (48.3 Hz) in the acoustic signals obtained using MSB provided a greater distinction between healthy and faulty cases for all harmonics than the power spectrum, suggesting that averaging MSB peaks in the low-frequency range which contains the shaft drive frequency should allow strong differentiation of impeller wear defects. Moreover, the vane passing frequency (338.3 Hz) for the baseline and seeded impeller faults showed clear separation of the acoustic peak magnitudes between baseline and both wear faults for MSB plots and the power spectrum. The experimental results show that the acoustic signal, whether analysed using MSB or power spectrum, outperformed vibration analysis.
|Date of Award||2023|
|Supervisor||Andrew Ball (Main Supervisor), Fengshou Gu (Co-Supervisor) & Ann Smith (Co-Supervisor)|