Diffusion-controlled decaffeination in packed bed extraction columns is governed by the serial mass-transfer resistance across the external solvent film, the stagnant pore liquid, and the swollen cell-wall matrix of green coffee beans. The rate-limiting transport step is not the convective delivery of solvent through the interstitial void volume but the migration of caffeine through the water-swollen bean matrix to the particle surface. This condition is quantified by the mass-transfer Biot number, Bi_m = k_f R / D_eff. When Bi_m exceeds 10, the internal diffusion resistance dominates the external film resistance. Industrial columns typically operate in this regime because the solvent film coefficient k_f is large relative to the effective intraparticle diffusivity D_eff. Caffeine is reduced from an initial concentration of 0.9–1.4 % dry matter to a final concentration not exceeding 0.1 % dry matter, as defined for decaffeinated coffee under Directive 1999/4/EC and quantified by ISO 20481:2008 high-performance liquid chromatography. The characteristic diffusion time for a spherical bean is τ_diff = R² / D_eff. For a representative bean radius of 3.5 mm and an effective diffusivity of 1 × 10⁻¹⁰ m²/s, the diffusion time scale is approximately 34 h; this value is consistent with industrial batch cycles of 8–12 h when solvent concentration driving force, prewetting, and temperature are incorporated into the time constant. The diffusion equation for a sphere with a uniform initial caffeine concentration and a surface concentration set by the solvent partition condition provides the concentration decay profile; the long-time solution is dominated by the first eigenvalue, giving a time constant proportional to R² / (π² D_eff). Packed-bed design therefore concentrates on reducing bean radius, increasing D_eff through controlled water plasticization, and maintaining a high solvent-side caffeine capacity by removing caffeine from the solvent loop before recycle.

Why Does Particle Radius Control Extraction Time More Than Solvent Velocity?

The dependence arises from the squared radius term in the diffusion time constant. Doubling the particle radius increases the diffusion time by a factor of four, whereas doubling the solvent velocity only reduces the external film resistance when the Biot number is below 10. In packed beds with superficial velocities constrained between 0.5 mm/s and 2.0 mm/s, the Sherwood number for caffeine in supercritical CO₂ is commonly in the range 2–10; the external film coefficient k_f calculated from Sh = k_f d_p / D_m is therefore large enough that increasing velocity has a diminishing effect on the total extraction rate. Direct single-bean extraction studies show that reducing coffee bean radius from 3.5 mm to 2.0 mm can shorten total extraction time by more than 50 %, a result consistent with the R² scaling, whereas increasing solvent flow by a factor of two can leave the extraction time nearly unchanged. Pre-wetting alters D_eff by swelling the cell-wall matrix and dissolving caffeine into an internal aqueous phase. At moisture contents below 20 wt%, caffeine remains largely crystalline or hydrogen-bonded within the cytoplasm, and effective diffusivity can be two orders of magnitude lower than at 40 wt% moisture. At moisture contents between 35 wt% and 45 wt%, the bean matrix is plasticized, cell-wall polymer chains are more mobile, and caffeine has continuous aqueous pathways to the particle surface. The partition coefficient at the bean–solvent interface, K = C_solvent / C_water, then determines the surface concentration driving the internal diffusion. For methylene chloride, the partition coefficient strongly favours the solvent phase; for supercritical CO₂, the water–CO₂ partition coefficient is pressure- and density-dependent and can be modified by polar entrainers or by saturating the CO₂ with water.

Across a full-scale extraction column, the pressure drop must be low enough to avoid bean compaction, yet high enough to prevent channelling and maintain liquid distribution. The Ergun equation for pressure drop in a packed bed is ΔP/L = 150 μ (1−ε)² u / (ε³ d_p² φ²) + 1.75 ρ (1−ε) u² / (ε³ d_p φ), where u is the superficial velocity, μ is the solvent viscosity, ρ is the solvent density, ε is the bed void fraction, d_p is the mean particle diameter, and φ is the particle sphericity. For green coffee beans with d_p of 7.0 mm, ε of 0.36, and φ near 0.75, the viscous and inertial contributions are both relevant at industrial superficial velocities. At 1 mm/s with a solvent density of 0.78 g/cm³ and viscosity of 9 × 10⁻⁵ Pa·s, the pressure drop is estimated in the range 0.2–0.4 bar/m; at 2 mm/s, the pressure drop can exceed 1 bar/m, and the mechanical stress on the bean bed becomes significant. Industrial extraction columns are therefore built with height-to-diameter ratios of 4:1 to 8:1, with bed heights commonly between 6 m and 12 m, and total pressure drops held below 5 bar. The vessel design is governed by ASME BPVC Section VIII Division 1 for pressure service, with internal distributor plates, hold-down screens, and liquid collection cones. Radial maldistribution is a documented production-scale failure mode: if the distributor plate becomes fouled with bean fines or swollen bean fragments, solvent preferentially flows through the centre of the bed, leaving the annular region undermixed. This condition has been observed to reduce apparent extraction efficiency by 10–20 % between screen-cleaning cycles. The packed bed is therefore not a fixed-bed reactor in the catalytic sense; it is a compressible biological bed whose void fraction, particle diameter, and sphericity change during prewetting and extraction.

When Bed Compaction from Water Swelling Collapses Interstitial Void Volume

Water prewetting is necessary to plasticize the bean matrix, but the same water uptake increases bean volume and can close the interstitial void spaces that the solvent must traverse. If the prewetting moisture exceeds 50–55 wt%, the green coffee beans swell by approximately 15–25 % in diameter, and the bed void fraction can fall from a dry value near 0.40 to below 0.30. Since the Ergun pressure drop is proportional to (1−ε)² / ε³, this reduction in void fraction can increase pressure drop by a factor of 2–4 at the same superficial velocity. The column then enters a regime in which the solvent flow is non-uniform, the beans at the bottom of the bed are mechanically compressed, and caffeine extraction fronts become highly asymmetric. This is a critical processing threshold: below 20 wt% moisture, diffusion is too slow for an economic cycle; above 50 wt% moisture, hydraulic transport becomes unstable. The optimal prewetting window is therefore 35–45 wt% for methylene chloride and ethyl acetate processes, and 40–55 wt% for supercritical CO₂ processes where the water also acts as an entrainer. The prewetting step is carried out in a separate rotating drum or screw conveyor with water spray nozzles and a residence time of 2–6 h to allow uniform moisture penetration. Near-infrared probes mounted in the drum discharge chute measure moisture every 10–30 s, and the set point is adjusted for the initial bean moisture determined by ISO 6673:2003 oven drying at 105 °C. Operational experience on production-scale lines shows that a batch-to-batch moisture variation of ±5 wt% can shift the extraction cycle time by 10–15 % and increase the pressure drop by 20 %. At the upper end of the moisture range, the bed must be loaded gently and screened to remove fines; otherwise the bottom discharge screen can become blinded by swollen bean fragments and the extractor must be emptied for cleaning.

In methylene chloride and ethyl acetate decaffeination trains, the solvent entering the extraction column is not fresh solvent in a single pass. The solvent circulates continuously through the packed bed, a caffeine-removal step, and a solvent-recovery evaporator. The caffeine concentration in the solvent leaving the bed is set by the equilibrium capacity, the contact time, and the internal diffusion rate from the beans. The recirculating solvent therefore has a finite inlet caffeine concentration that reduces the logarithmic mean driving force along the bed. The packed bed is normally operated as a battery of two to four columns in series, with solvent flowing countercurrent to the progression of bean batches. This arrangement increases the effective driving force and reduces solvent-to-feed ratios. The solvent-to-feed ratio for methylene chloride processes is typically between 5 L/kg and 20 L/kg of green coffee per cycle; for supercritical CO₂ processes, the solvent mass flow rate is specified in terms of interstitial velocity rather than a fixed liquid ratio because the density and solvent capacity change with pressure and temperature. In all cases, the solvent must be subjected to caffeine removal before returning to the bed. For methylene chloride, caffeine is removed by liquid-liquid extraction with water, and the solvent is then evaporated and condensed. The residual solvent limits in the final decaffeinated coffee are specified under 21 CFR 173.255; compliance is verified by gas chromatography with a method limit of quantitation of 1 ppm or better. For supercritical CO₂, caffeine is separated from the expanded gas or liquid stream by pressure reduction, water scrubbing, or adsorption onto activated carbon. The inlet caffeine concentration to the extractor after these steps must be maintained below 0.05 g/L to prevent a significant loss of driving force.

Mass Transfer Zone Length, Axial Dispersion, and Caffeine Breakthrough in Recirculating Solvent Loops

The concentration front in a packed bed is not an ideal step; axial dispersion, intraparticle diffusion, and finite external film transfer broaden the breakthrough curve. The axial dispersion coefficient D_ax is commonly expressed through the axial Péclet number, Pe_ax = u_s L / D_ax. For green coffee beds with u_s between 1 mm/s and 3 mm/s, tracer residence time distributions give Pe_ax values between 20 and 100. This is intermediate dispersion: the mass transfer zone can occupy a substantial fraction of the bed. The height of the mass transfer zone is approximated by H_MTZ = (D_ax / u) ln(c_feed / c_breakthrough) plus a term proportional to the intraparticle diffusion time. In industrial decaffeination columns, the mass transfer zone is often 30–60 % of the bed height early in the cycle and contracts as the caffeine level in the bean decreases. The practical consequence is that a single pass of solvent through the bed removes only a fraction of the caffeine; multiple passes are required. The number of passes is determined by the caffeine mass balance, V_s dC/dt = Q (C_out − C_in), where V_s is the solvent volume in the loop and Q is the volumetric flow rate. When C_in is not zero, the driving force is reduced. Caffeine removal efficiency in the solvent loop must therefore be high; otherwise the cycle time expands. In a well-designed methylene chloride recirculation loop, the caffeine removal step reduces the recycled solvent concentration to below 0.05 g/L, while the concentration exiting the extraction column may be 0.2–0.5 g/L. The practical dynamic loading of the bed is therefore a function of the solvent capacity, the recirculation rate, and the removal step efficiency, not simply the equilibrium solubility. The packed bed is not saturated column-by-column in a simple frontal chromatographic sense; it operates as part of a coupled separation loop.

At 25 MPa, the extraction temperature window for supercritical CO₂ decaffeination is narrow because two opposing effects operate. Increasing temperature from 60 °C to 65 °C lowers the CO₂ density by roughly 3 %, which can reduce the caffeine solubility by 5–10 %; however, the same temperature increase lowers solvent viscosity and increases the caffeine diffusivity by a comparable amount. Decreasing temperature from 60 °C to 55 °C increases density and solubility but retards intraparticle diffusion and may reduce extraction rate. The net effect is that a temperature deviation of ±5 °C can shift the cycle time by 10–15 % and change the caffeine capacity of the solvent by a measurable amount. This is a critical processing threshold: the system is tuned around a density window of 0.68–0.90 g/cm³ and a caffeine diffusivity window of 1 × 10⁻¹⁰–5 × 10⁻¹⁰ m²/s. Pressure must also be controlled within ±0.5 MPa because a drop in pressure reduces density and can shrink the solvent capacity. The pump and back-pressure regulation system is designed to hold pressure within ±0.2 MPa at the column inlet. On production-scale lines, the extraction column is jacketed and heated in zones, with temperature measured by resistance temperature detectors mounted at 3–5 axial positions. The prewetting water temperature is maintained between 50 °C and 60 °C to avoid thermal degradation of chlorogenic acids and to prevent steam flashing into the bean. The extraction vessel is depressurized at the end of each cycle through a series of pressure letdown valves; rapid depressurization below 6 MPa can cause water in the beans to flash and crack the seed coat, which reduces product quality and increases fines generation. The spent beans are then vacuum or steam stripped to remove residual solvent and dried to a moisture below 12 wt% before storage. Storage at relative humidity above 60 % can increase water activity above 0.65, which favours mould growth and shortens the shelf life of the decaffeinated green coffee.