The accurate prediction of pressure drop across randomly packed beds is critical for the design of chemical reactors, adsorption columns, and thermal energy storage systems, yet existing semiempirical correlations often lack consistent accuracy across diverse geometric and flow regimes. This study presents a residual-driven sensitivity analysis framework that systematically evaluates the influence of key operational and geometric parameters, namely, superficial velocity $U_s$, normalized bed length $L/D$, and normalized column diameter $D/d_p$, on the prediction error of classical pressure drop models. Using an extensive experimental dataset collected in our laboratory, comprising over 3000 measurements across four column diameters (50–200 mm) and six particle sizes (2.5–42.6 mm), we integrate robust outlier detection, iterative data sanitation, and multimethod sensitivity assessment combining Pearson correlation, tree-based feature importance, and partial dependence analysis. Results show that for the raw dataset, errors are collectively influenced by all three parameters, with no single factor dominating. Through successive cleaning iterations that progressively remove data with strong wall effects ($D/d_p<4$), low length-to-diameter ratios, and extreme geometries, the analysis reveals a clear transition: Superficial velocity emerges as the overwhelmingly dominant driver of residual error, accounting for 74.6% of the normalized Gini importance in the final filtered dataset. This evolution demonstrates that after mitigating geometric confounders, the core limitation of widely used models such as the Ergun equation lies in their inability to accurately capture flow-regime-dependent behavior. The work provides a validated methodology for sensitivity analysis in packed-bed flows and offers practical guidance for model selection and experimental data-quality assurance, highlighting that future model improvements should prioritize velocity-dependent corrections rather than further refinement of geometric terms.

Close

PrevNext

View figure in article

• Flows in porous media
• Fluidized beds
• Pressure techniques