A digital-twin testbench for centrifugal pump systems

This paper presents the Centrifugal Pump Digital Twin Testbench (CPDTT), an instrumented and fully automated experimental platform developed to study mechanical, electrical, and hydraulic failure modes in centrifugal pump systems under controlled conditions. The setup features an advanced architecture for control, data acquisition, and processing, together with a system-level Digital Twin (DT) that enables the implementation of predictive-maintenance algorithms and supports maintenance-oriented decision-making. The resulting linkage supports fault explainability by connecting observed experimental behaviors to specific failure modes. This platform is used to validate novel machine-learning algorithms based on power spectral density (PSD) analysis, aiming to extract features that provide information about the health condition of the different centrifugal-pump components. Cavitation, along with other mechanical and electrical failures, is among the most critical phenomena affecting pump efficiency and lifetime. The DT incorporates physics-based simulation models of the real system, enabling the generation of labeled energy datasets in healthy conditions and under induced faults, and supporting experimentation with supervised learning workflows. A detailed description of the system and the applied methodologies is provided. Overall, the testbench and its digital twin provide a practical environment for analyzing fault behavior in hydraulic systems, validating monitoring algorithms, and assessing computational performance. The methodology used to induce incipient cavitation and the corresponding detection strategies are also described.