Cavitation modelling and ML integration in the DT prediction of extreme pressure heads

• Experiments–ML integration strengthens Digital Twin prediction. • Cavitation model in 1D MOC validated experimentally. • Neural network predictions improve risk analysis. • ML enables cavitation risk detection. • Air–valve misuse destabilises transients in low sloped pipelines. This research presents a hybrid approach exemplifying the convergence of deterministic modelling and data-driven prediction to enhance simulation fidelity in complex hydraulic systems. It integrates experimental testing, numerical modelling, and Machine Learning (ML) to analyse hydraulic transients and cavitation. Laboratory water hammer tests validated simulations using a developed Cavitation Model (CM) and an adapted Bentley’s Hammer, both based on the 1D Method of Characteristics (MOC) with discrete cavity representation when vapour pressure is reached. Comparisons between experimental data and CM outputs provided insights into Digital Twin (DT) applications. Air valve behaviour was also highlighted, as improper integration can destabilise transient flows in slightly sloped profiles. In parallel, an ML framework was developed to predict maximum (Hmax) and minimum (Hmin) pressure heads from flow rates (25–950 l/h) and pipe diameters (0.005–0.06 m). Using 5‑fold cross-validation with 35 % of data reserved for testing, a three-layer neural network achieved R 2 values of 0.86 (Hmax) and 0.94 (Hmin) in validation, and 0.46 (Hmax) and 0.91 (Hmin) in testing. The strong alignment between predicted and simulated/measured values, particularly demonstrates the integrated model’s value in identifying cavitation risks and enabling proactive system management and design optimisation without reliance on complex hydraulic simulations.