Data-driven regime identification in subsurface gas–solid flows using volumetric tomography

Underground gas releases through granular media display complex and transitional flow behavior that remains difficult to characterize using surface observations or isolated spatiotemporal snapshots. Few studies have associated underground gas leak scenarios with flow regimes, and those attempts are typically qualitative in nature. This study proposes a data-driven methodology for the systematic analysis and interpretation of subsurface gas–solid dynamics using volumetric tomography (i.e. Electrical Volume Capacitance Tomography - ECVT) measurements. Experimental data were acquired from bench-scale experiments varying inlet flow conditions, orifice diameters, and granular materials. Spectral analysis of transient solid volume fraction signals revealed distinct frequency-domain features, which were subsequently classified using partitional clustering of periodograms. A parametric assessment examined the sensitivity of clustering outcomes to key methodological choices, including feature extraction settings, clustering algorithms, distance metrics, number of clusters, and dataset size, using multiple quality evaluation metrics and post-hoc comparisons against visually interpreted regimes. The most robust configuration was obtained using Welch-based power spectral densities with a resolution of 129 frequency bins, clustered via a condition-based TimeSeriesKMeans algorithm employing the Dynamic Time Warping (DTW) metric. Across 944 pressure runs, the proposed framework spanned a broad spectrum of gas–solid flow behaviors and consistently identified transitions between regimes. Physical interpretation was derived by layering qualitative regime attribution onto the unsupervised clustering results, explicitly acknowledging the inherent ambiguity and overlap between transitional states. The analysis captured regimes such as gas migration and multiple forms of fluidization, with four clusters sufficient to represent the most distinguishing regime characteristics. Overall, the methodology provides a generalizable framework for organizing and interpreting complex subsurface gas dynamics and may be extended to numerical simulations or related physical systems, such as subsea gas releases.