Digital twin technology has transformed asset performance management by creating dynamic virtual replicas of physical systems. However, accurately modeling turbomachinery, such as centrifugal pumps, blowers, and compressors, for fault prediction remains a challenge due to their complex operating behavior and sensitivity to various factors. Cavitation in centrifugal pumps, a phenomenon in which micron-sized vapor cavities rapidly implode and cause significant surface damage, is particularly difficult to detect. Traditional methods such as vibration analysis, which rely on high-frequency data, often fail to provide a comprehensive understanding due to limitations in scope and resolution. Additionally, developing a dynamic model of pump vibration involves complex processes such as computational fluid dynamics (CFD), finite element analysis (FEA), and multi-body simulation. These approaches frequently fall short in accuracy when updating with real-time sensor data, due to the high computational requirements and the lack of clear transfer functions between different model types. To address these challenges, the integration of CFD computations with reduced-order modeling (ROM) offers a promising solution. High-fidelity computations, which capture the multiphase complex interfacial flow phenomena using the Schnerr-Sauer cavitation model for centrifugal pumps are employed. These simulations generate extensive datasets that, while accurate, are computationally intensive and not feasible for real-time applications. To make these simulations practical for real-time monitoring and fault prediction, Non-Intrusive Proper Orthogonal Decomposition (POD), combined with Radial Basis Functions (RBF), is used to reduce model order. POD allows for the extraction of the most significant features from the high-dimensional simulation data, and RBFs provide nonlinear interpolation to capture the dynamics of the model. Results show that the POD-based ROM achieves up to 98% accuracy in reconstructing high-fidelity simulations. This significant reduction in computational time, which the ROM reduces to milliseconds compared to the extensive durations required for high-fidelity simulations, enables real-time monitoring and rapid fault detection. Additionally, the size of the ROM is very small, making it not only computationally efficient but also cost effective for cloud deployment and storage, further enhancing its practicality for large-scale applications.