A Dimensionless Morphological Framework for Permeability Prediction in Porous Rocks: Extending Kozeny–Carman with Coordination Number and Throat-Scale Corrections
Аннотация
The Kozeny–Carman (KC) equation remains the standard closed-form predictor of permeability in porous rocks, yet it systematically under-performs in heterogeneous carbonates where connectivity, rather than porosity alone, governs flow. We extend KC by a Buckingham-$$\Pi $$ reduction of $$\{k,\tau ,\phi _{\textrm{eff}},Z_{\textrm{avg}},S,\bar{T}\}$$ to a single dimensionless complexity number $$\Pi = \tau ^{\alpha }/[\phi _{\textrm{eff}}^{\beta }\,Z_{\textrm{avg}}^{\gamma }\,(S\bar{T})^{2\delta }]$$, closed by $$k = 1/(c_{\textrm{lit}}\,S^{2}\,\Pi )$$. The closure is dimensionally consistent for any exponent values, sign-consistent with the physics of every descriptor, and reduces exactly to KC when $$\alpha =2$$, $$\beta =3$$, $$\gamma =\delta =0$$. The model hierarchy M0–M4 was fitted on $$1\,963$$ sub-volumes extracted from twenty parent micro-CT scans (ten clean clastic sandstone and ten Brazilian pre-salt carbonate volumes), yielding 923 sandstone and $$1\,040$$ carbonate sub-volumes, with single-relaxation-time lattice Boltzmann permeability as the reference. The models were selected by the corrected Akaike Information Criterion. The preferred model was M3 (KC pinned at $$\alpha =2$$, $$\beta =3$$ with $$\gamma $$ and $$\delta $$ free), with $$\hat{\gamma } = 0.67$$ and $$\hat{\delta } = 0.50$$, achieving log-space $$R^{2} = 0.986$$, log-MAE $$= 0.24$$ (typical multiplicative error $$1.27\times $$), and $$100\%$$ of samples within $$\pm 1$$ decade of the LBM ground truth. The unconstrained M4 fit returns $$\hat{\alpha } = 2.01$$ and $$\hat{\beta } = 3.00$$, empirically confirming the classical squared-tortuosity and cubic porosity scalings to within $$1\%$$. The throat-scale exponent $$\hat{\delta } = 0.50$$ lies midway between the KC ($$\delta =0$$) and Katz–Thompson ($$\delta =1$$) endpoints. Cross-lithology validation under an intercept-refitted protocol was symmetric ($$R^{2}_{\log } \gtrsim 0.95$$ in either direction), indicating that the four exponents capture lithology-invariant physics rather than within-lithology over-fitting.