[Truncated abstract] Centrifugal pumps are designed to operate at a capacity that maximizes the efficiency of the pump. Operation below this design capacity results in reduction of pump efficiency as the geometric configuration of the impeller and casing no longer provide an ideal flow pattern. Consequently there are changes in the flow field within the pump, including flow separation and regions of localized, non-uniform, unsteady flow. This flow interacts with rotating and stationary components inside the pump creating additional disturbance and hydraulic excitation. It is anticipated that the local hydrodynamic and global hydroacoustic excitation due to partial-flow operation will affect the structural vibration measured on the pump. In this study, the effects of partial-flow operation on the vibration signal at the pump bearing housing are measured on a number of industrial double-suction pump units. These are a particular class of centrifugal pump commonly used in high volume applications such as water distribution. The aims are to understand how the vibration signals change at the different operating conditions and to determine which parameters are best suited to monitoring the observed changes. Comparison is made between the response, under similar operating conditions, of pumps both within sets of identical size and design, across sets of similar design and different sizes, and between different designs. The pumps are all in-service industrial units. In double-suction pumps the impeller motion in the axial direction (along the line of the shaft) is not constrained. Due to symmetry in impeller design, axial stability is maintained by equal and opposite hydraulic forces on the two opposing sides of the impeller. The potential for loss of axial hydraulic balance during partial-flow operation is examined from a theoretical perspective, based on a literature review, and by measurement of the axial displacement of the shaft. Structural vibration is measured using accelerometers mounted at the non-drive end bearing housing in the horizontal and axial orientations. Changes in signal contribution and characteristics are examined using a variety of qualitative and quantitative techniques. Test signals are used to assess the limitations of the techniques and the effect of parameter selection on the interpretation of the signals.

Frequency response analysis (FRA) has emerged as the de-facto industry standard condition-monitoring tool to assess mechanical integrity of transformer windings during its service life. It the prerequisite detection sensitivity and customized portable commercial instruments are also available. Considering its importance and taking cognizance of its hidden potential, international bodies, viz., IEEE, IEC, and CIGRE have published standards/guides on its use and interpretation. In spite of all the progress witnessed over the past two decades, FRA has still not attained the status of a diagnostics tool. Probing the vast literature and research carried out in this points to the fact that lack of a rigorous mathematical basis to explain the underlying complex processes is, perhaps, one of the main reasons for the present predicament of FRA method. How-ever, it must be acknowledged that domain-knowledge is di cult to generalize in this. Having said that, the diagnostic part, which involves, the task of working back-wards starting from the FRA data to interpret a winding damage, locate it, and assess its severity, has so far remained teasingly elusive. As a consequence, FRA continues to remain as a sensitive condition-monitoring tool. Given its inherent potential, this situation seems to be a paradox, and so, calls for investigations. Once a mechanical damage has been detected by FRA, the next task is to locate its position and estimate its severity. An engineer expects FRA to provide these answers, so that corrective action, if needed, can be determined and initiated. In this context, even though FRA has attained global acceptance as a monitoring tool, it has failed as a diagnostic tool. Therefore, e orts that aim to address this issue are desirable. Driven by this motivation, the author's thesis proposes to explore a new school of thought in this direction, viz., to theoretically analyze the problem of localization of an incipient/minor mechanical damage (displacement in particular) and also assess its severity. Such an investigation seems to have not been undertaken previously. So, the goal is to establish a relationship to capture the complex interactions that exist between specific winding damages, winding parameters, and their overall in hence on the natural frequency deviations observable in the FRA data. Hence, exploring this possibility, subject to the constraint that the proposed method shall use inputs that are measurable at the terminals, becomes the primary objective of this research. In this thesis, a generalized analytical framework for handling winding displacements and FRA data has been successfully formulated. The formulation provides a general platform for localization and severity assessment of true radial and axial winding dis-placements occurring in an actual winding. An analytical solution becomes possible mainly due to manipulation of the system matrix, i.e., to consider the harmonic sum of squares of natural frequencies, instead of just the natural frequencies. This manipulation leads to an elegant closed-form expression that connects the displacement location and its severity, to changes in natural frequencies. For its implementation, short circuit natural frequencies and a few other terminal measurements are the only inputs that are necessary. This formulation is initially used in Chapter-3 to demonstrate localization of radial displacement in an isolated, actual, single, air-cored continuous-disk winding. Armed with this success, the supplicant proceeds further to show (in Chapter-4) how a minor manipulation of the formulation renders it suitable for localization of actual axial displacements as well. Extensive experimental verification was done and the results are encouraging. Accuracy of localization of radial/axial displacement is uniformly good for all positions, and so is the estimation of severity. Further details are presented in the thesis.