This study investigates the sparse reconstruction of unsteady surface pressure fields in ventilated cavitating flows using an experimentally validated numerical framework and a data-driven strategy. A systematic numerical investigation is conducted to characterize the transient flow evolution and provide the ground truth for method development and evaluation. Based on this database, an integrated framework combining Gappy proper orthogonal decomposition (POD) with a clustering-guided unsupervised learning strategy is developed to identify dynamically similar regions and determine representative virtual sensor locations for sparse reconstruction. POD analysis reveals pronounced low-dimensional characteristics of the pressure dynamics, indicating that the dominant flow evolution can be represented by a limited number of coherent modes. During the re-entrant-jet-dominated stage, the first POD mode contains approximately 85% of the modal energy, indicating coherent low-order dominance. In contrast, during cavity break-off and shedding, the first-mode contribution decreases to approximately 65%, accompanied by a redistribution of modal energy toward higher-order modes and enhanced multi-scale pressure fluctuations. These flow-physics characteristics explain the variation in reconstruction difficulty across flow stages and provide a physical rationale for sparse pressure-field reconstruction. The reconstructed pressure fields agree well with the numerical ground truth, with eight virtual sampling points providing an effective compromise between reconstruction accuracy and measurement cost. The proposed framework also exhibits promising in-regime transferability under intermittent and transitional cavity conditions, together with satisfactory robustness to clustering initialization and measurement noise. Overall, the proposed framework offers physical guidance for future sensor placement under limited-measurement conditions.
Meng et al. (Mon,) studied this question.