<p>In this study, long-term vibration monitoring and analysis were conducted on a two-stage centrifugal pump operating at 2975&#xa0;rpm to investigate the progression of bearing degradation. Over three years, vibration data were systematically collected and analyzed using spectral, time domain, and envelope analysis, and power spectrum techniques. The power, velocity, and acceleration spectrum results revealed predominantly random high-frequency excitations related to the surface roughness of the rolling elements, with no distinct defect frequency components. These broadband excitations were mainly distributed across random frequency bands, reflecting the non-periodic impacts generated as the roughened rolling elements passed over the inner and outer raceways. The envelope analysis clearly showed fault-related amplitude modulations, with interactions between the roughened ball surfaces and the cage producing characteristic signals indicative of pitting-induced cage damage. For the first time, it was demonstrated that artificially introduced pitting on the rolling elements could excite the fundamental train frequency (FTF) and its harmonics through ball–cage interactions, thereby providing a novel diagnostic indicator for the early detection of cage-related bearing faults. Following fault identification, the thrust bearing was replaced, after which the vibration levels returned to normal. Post-inspection confirmed the presence of pitting damage on the rolling elements. This long-term investigation underscores the importance of multi-domain vibration analysis and highlights the superior diagnostic performance of power spectrum and envelope spectrum analysis—particularly in the velocity and acceleration domains—compared with conventional spectral and time-domain approaches, especially for the early identification of bearing pitting defects.</p>