Christopher A. Allison, Ruotong Huang, Anindityo Patmonoaji, Lydia Knuefing, Anna L. Herring
7 min
Abstract
Understanding interphase mass transfer is essential for a variety of applications in porous media, ranging from groundwater remediation to geologic energy storage. While X-ray micro-Computed Tomography (microCT) provides critical in situ observations, analyzing mass transfer requires models and workflows compatible with the limited spatial and temporal resolution. Current literature presents three analytical frameworks for evaluating interphase mass transfer using microCT data: the Slice-Averaged Concentration (SAC) approach, the Non-Classified per-Cluster (NPC) approach, and the Classified per-Cluster (CPC) approach. This study evaluates the results of all three approaches across four sets of time-lapse tomography sequences that observe hydrogen dissolution at varying solvent injection rates. To mitigate biases arising from dissolution-driven cluster remobilization, we introduce a volume-ratio filtering technique to all workflows to ensure that estimates more accurately reflect true mass transfer events. Our analysis finds that all three analytical approaches estimate average mass transfer coefficients within one order of magnitude of one another at the same solvent injection rate. However, the similarity between the estimates of each approach diverges when approximating more complex phenomena, such as aqueous solute concentration profiles. Ultimately, the utility of one approach over another is determined by the desired level of system detail, at the cost of the computational resources required to achieve it. Higher phenomenological resolution requires greater computational processing and refinement due to increased sensitivity to measurement and processing noise, as well as outlier events. We anticipate that the findings will provide a framework for researchers to match analytical approaches to their available computational resources and desired level of physical detail.
Alex: Okay, so SAC treats the rock column like stacked layers, while the cluster ones zoom in on each bubble separately?
Sam: Exactly. The non-classified per-cluster method measures volume and surface changes for every cluster, assuming fresh water around each one drives maximum dissolving. That's NPC. The classified version goes further: it sorts clusters by what happened—fully gone, partly shrunk, split off, or grown—then only uses the fully dissolved ones, figuring those hit the peak dissolving zone where water is freshest. They pair up images from scans taken one after another, matching gas pockets by comparing their centers and sizes—like spotting the same balloon in two photos taken minutes apart, allowing it to shrink or shift a bit.
Alex: So CPC is pickier, only trusting the cleanest shrinks. But with clusters moving around, how do they avoid junk data messing up the rates?
Sam: They add that volume-ratio filter—it skips intervals where new volume shows up almost as much as what's lost, dodging remobilization fakes. All three, after filtering, give mass transfer coefficients that match within about a factor of 10, even if concentrations vary more. This suggests the workflows trade detail for speed reliably, letting researchers pick based on their setup for things like storage predictions.
Alex: That's a clear step for trusting these scans without overcomplicating.
Alex: Rates line up, but you mentioned concentrations spread wider across methods—up to full solubility differences. Why do they disagree there, even if the core rates match?
Sam: The slice-averaged method spreads the dissolving effect across entire horizontal layers, assuming the water mixes evenly in each one—like blaming the whole classroom for a spill instead of just the desk nearby. This dilutes the numbers, keeping them safely between zero—no dissolved gas—and one—fully saturated. Per-cluster methods work backward from average rates to guess water freshness around each pocket, sometimes landing outside those bounds because real dissolving varies. The non-classified one includes pockets that grew, pushing estimates over saturation since growth means less dissolving than assumed. Classified avoids growth by picking only shrinking ones, but still gets undersaturated guesses from small pockets where measurements fuzz out.
Alex: So SAC smooths everything out like a big average, while cluster ones try pinpointing but hit snags from uneven real conditions?
Sam: Yes. These issues cluster near the dissolving front, especially at faster flows where tiny bits drag down the overall rate guess. All three show rates rising with injection speed, and per-cluster ones run higher than slice-averaged, but still within a factor of ten—like estimates from 1 to 10 units apart.
Alex: A factor of ten is rough agreement for rates. What about stacking up against other studies?
Sam: They fit within the same range as past work on gases like CO2 or solvents in bead packs or sandstones, though exact orders flip sometimes—like one finding slice-averaged higher. Differences stem from rock types or conditions, so more matching experiments would sharpen predictions. The paper notes simpler methods suit broad checks with less computing, while detailed ones reveal fine patterns like uneven fronts, at higher effort.
Alex: A balanced view, then—rates trustworthy across the board, concentrations as a method fingerprint. Solid for real storage planning.
Sam: Pulling it all together, the paper shows these three methods deliver mass transfer coefficients that align within a factor of ten, even as concentrations diverge. This validates picking workflows by what's available—slice-averaged for quick, low-compute overviews of big-picture rates, or per-cluster ones for spotting finer details like uneven solvent fronts, though they demand more processing power and stats to handle distributions. The study highlights real limits: it's tied to this one rock type and hydrogen gas, so results may shift in other setups like different sands or gases, as past work suggests.
Alex: Right, so SAC keeps it simple and macroscopic, while NPC and CPC dig into cluster-level events but trade off with variance or smaller samples. Overall, it offers a practical guide: standardize these μCT pipelines to balance computer time against detail, aiding real-time monitoring for underground storage or cleanup where predicting dissolution matters.
Sam: In essence, this work strengthens confidence in μCT for porous media studies, showing methodological choices yield consistent core insights despite trade-offs.
Alex: A meaningful step for reliable predictions in storage tech. Thanks for breaking it down, Sam. Thanks for listening to ResearchPod.