A 501(c)(3) non-profit organization Applied AI research for public benefit
CheAI Research Inc.

Anhydrous ethanol separation takes 57.6 % of the heating value at 23.9 g/L, the rule-limited beer of a yeast fermenter model

We computed how the separation energy of an anhydrous ethanol plant depends on beer concentration and compared it with the range of a yeast-fermenter model.

CheAI Research Inc. | Research article | November 2025

Abstract

The ethanol concentration of the fermenter output, the beer, sets how much water a fuel-ethanol plant must boil away. We modelled a plant that makes 5.0 t/h of 99.5 wt % ethanol with a beer column, an extractive column with ethylene glycol (EG) and a vacuum solvent-recovery column, with the solvent recycle closed. DWSIM solved every column rigorously, and a kinetic yeast-fermenter model supplied the beer. The separation energy fell monotonically from 25.4 MJ/kg at 12.5 g/L, 94.7 % of the ethanol lower heating value (LHV), to 7.15 MJ/kg at 100 g/L. It came within 10 % of the high-concentration value at 62 to 79 g/L, depending on the reference point. Ethanol inhibition limited the fermenter model to 23.9 g/L under temperature, conversion and volume rules, where separation still cost 57.6 % of the LHV. Some design solves from 40 g/L upward failed, so the high-concentration energies are not proven minima.

1. Introduction

A beer column strips the ethanol from a dilute beer to near the azeotrope. Molecular sieves [1] or extractive distillation with a heavy solvent such as EG [2] then dehydrate it to 99.5 wt % or more. The beer column dominates the energy demand, because its feed is mostly water. Earlier studies linked a stronger feed to lower distillation energy and cost [1], [3], [4], and heat-integrated designs can narrow the gap for dilute beers [5].

No study cited here combines rigorous column solves, a closed solvent loop and a kinetic fermenter range in one energy curve. Such a curve shows how strong the beer must be before a further increase saves little, and whether the fermenter can reach it. Here the energy came within 10 % of its high-concentration value at 62 to 79 g/L, but the fermenter model reached only 23.9 g/L under its rules.

2. Methods

A glucose solution fed a continuous stirred-tank yeast fermenter. We used the kinetics of Nagy [6] as implemented in an earlier study of the same reactor [7], with exponential ethanol inhibition of growth [8]. At steady state, the biomass balance bounds the ethanol concentration by

CP ≤ (1/KP) ln(μX(T)/D),

with KP = 0.139 L/g, growth constant μX(T) and dilution rate D. We solved the steady state at 6,300 grid points. Three judgment rules (at most 35 °C, at least 90 % glucose conversion, at most twice the baseline volume) set the top of the fermenter band. The band starts at the 12.5 g/L baseline of the earlier study.

The beer was idealised to ethanol and water and preheated in one feed/stillage interchanger (10 K approach). The 50-stage beer column sent a 90 wt % distillate to the extractive column, and a recovery column at 20 kPa returned the EG. DWSIM 9.0.5 [9] solved all columns rigorously, with non-random two-liquid (NRTL) parameters fitted to measured binary equilibria [10]. We fed the converged recycle through the columns again as fresh solvent. Acceptance required no mole-fraction change above 10-5 and unchanged recycle flow and temperature.

The separation energy was the three reboiler duties plus the steam preheat left after the interchanger, per kg of product. At nine beer concentrations, the reflux ratio R was searched for at least 99.5 % ethanol recovery. Failed or timed-out solves counted as unknown, so each energy is the best feasible design found. The knee is the concentration where the energy comes within 10 % of the value at 75 or 100 g/L.

3. Results

3.1 Separation energy falls monotonically with beer concentration

The energy fell from 25.41 MJ/kg at 12.5 g/L to 15.44 MJ/kg at the band top of 23.9 g/L (Figure 1). It reached 7.99 MJ/kg at 75 g/L and 7.15 MJ/kg at 100 g/L. These values are 94.7, 57.6, 29.8 and 26.6 % of the LHV. On the designs found, a stronger beer always needed less energy, with diminishing returns. Without the interchanger, the energy fell from 46.96 to 9.44 MJ/kg, so the interchanger saved almost half at 12.5 g/L and a quarter at 100 g/L.

Separation energy against beer concentration.
Figure 1. Separation energy falls monotonically with beer concentration. The dashed line is the upper bound without the interchanger. Hollow circles mark points with failed solves, crosses mark the knees, and the grey band is the fermenter band.

The knee was at 62.2 g/L against the 75 g/L reference and at 78.8 g/L against 100 g/L. The fermenter band ends below both knees by a factor of at least 2.6. A 3 % change in the reference energy moved the knee by 2.4 to 3.5 g/L against 75 g/L and by 4.8 to 5.5 g/L against 100 g/L.

3.2 Ethanol inhibition caps the beer from the fermenter

At the baseline (60 g/L glucose fed, D = 0.051 h-1, 29.6 °C), the fermenter converted only 50 % of the glucose and gave 12.5 g/L ethanol. The three rules gave a band top of 23.9 g/L, at 35 °C, 63 g/L glucose fed and 90.3 % conversion (Figure 2). The band top follows the selected rules, not the limit of the model. At 35 °C and about 8.4 times the baseline volume, the model reached 38.1 g/L.

A higher sugar feed did not raise the ceiling, because the bound does not depend on the feed. The highest steady state anywhere on the map, 40.4 g/L, converted only 23.6 % of its glucose. The points above 23.9 g/L are therefore separation-only results, not states the fermenter model reaches under the band rules.

Highest ethanol concentration against fermenter volume.
Figure 2. Ethanol inhibition caps the beer concentration of the fermenter model. Highest steady-state ethanol concentration at 90 % conversion or more against the volume needed for 5.0 t/h of product, at three temperatures. Dotted and dashed lines mark once and twice the baseline volume.

3.3 The beer column sets the trend, and the EG loop adds a constant floor

From 12.5 to 100 g/L, the beer-column reboiler duty fell from 18.98 to 3.68 MJ/kg and the residual preheat from 3.20 to 0.23 MJ/kg (Figure 3). The EG loop added a constant 3.23 MJ/kg, because it received the same 90 wt % distillate at every point, and its share grew from 13 % to 45 %.

Contributions of the four terms to the separation energy.
Figure 3. The beer column sets the trend, and the EG loop adds a constant floor.

3.4 Baseline and equilibrium checks agree, and the closed loop converges

The fermenter reproduced the baseline ethanol concentration within 0.12 %, and the NRTL set matched 33 held-out ternary bubble temperatures with an average absolute deviation of 0.17 K. The closed EG loop converged to a verified fixed point with 99.588 wt % product. At 40, 55 and 75 g/L, 5 to 9 of the 27 design solves failed or timed out.

4. Discussion

The energy fell steeply below about 40 g/L and slowly above it, with no optimum at an intermediate concentration. At low concentration the beer column is stripping-limited, and its duty scales roughly with the water it boils. At high concentration its term becomes comparable to the fixed EG-loop term, so a stronger beer saves little.

Even the model maximum of 40.4 g/L lies well below the 16.7 % (v/v), about 130 g/L, of a continuous very-high-gravity cascade [11]. The trend agrees in direction with earlier studies [1], [3], [4]. Specialised designs let beers below 40 g/L approach the energy of those above 120 g/L [5], which is consistent with the large interchanger saving here. Heat recovery decides the absolute energy as much as beer concentration does.

At the band top, the beer-column reboiler and residual preheat make up 79 % of the energy. A different entrainer or an extractive dividing-wall column [12] acts on the remaining share. A stronger beer, or heat integration of the beer column, acts on the larger share.

5. Limitations

The fermenter is the weakest model. Its yield constants are independent, so its carbon balance closes to only 95.5 %, and its response to an inlet-flow step failed the check. Its energy balance is lumped, and it neglects ethanol lost with the vented CO2. The model uses hours, whereas the earlier study labels the same numerical time values as minutes. The plant uses only its steady-state ethanol/water ratio and flow, and the band rests on three judgment rules and a jacket-duty check at the 1,000 L model scale. On the feed side, biomass removal was not modelled, and residual glucose was estimated (at most 0.30 % of the energy) with a heat capacity extrapolated to 373 K, not simulated. Glucose effects on phase equilibrium beyond ideal water activity, fouling and non-volatiles such as glycerol and salts are not covered.

Each energy is the best feasible design found under a fixed rule, not a global or economic optimum. From 40 g/L upward, failed solves are unknowns, so cheaper designs can exist. The 100 g/L point has no complete design search. The numerical uncertainty of the knee is not bounded, and the 10 % threshold is a judgment. The EG loop was not re-optimised along the curve, and the 85 wt % distillate case and the uncorrected vapour pressure at 75 g/L were not computed. The fitted ethanol vapour pressure is extrapolated above 366.6 K and was tested at 23.9 g/L only.

The closed loop was not solved again at 100 g/L. It has at least two steady states and did not return from every perturbed start, and the product margin is only 0.088 wt %. The EG make-up of 3.2 g/h counts vapour losses only, with no degradation, entrainment or purge. The recycle tear tolerance leaves 83 to 92 g/h of EG unresolved, 26 to 29 times the make-up. More broadly, the checks confirm arithmetic and bookkeeping but do not validate the models against plant data. Costs and capital are outside the scope.

6. Conclusions

On the best feasible designs found for this plant, the separation energy fell monotonically from 25.4 MJ/kg at 12.5 g/L to 7.15 MJ/kg at 100 g/L. It came within 10 % of the high-concentration value at 62 to 79 g/L, depending on the reference point. Limited by ethanol inhibition, the yeast fermenter model reached 23.9 g/L under the operating rules, where separation still took 57.6 % of the product heating value. Beer-column heat integration was not modelled, so its comparison with a stronger beer needs further process designs.

References

  1. Vane LM. Separation technologies for the recovery and dehydration of alcohols from fermentation broths. Biofuels, Bioproducts and Biorefining 2, 553-588, 2008. DOI 10.1002/bbb.108
  2. Lei Z, Li C, Chen B. Extractive distillation, a review. Separation & Purification Reviews 32, 121-213, 2003. DOI 10.1081/SPM-120026627
  3. Zacchi G, Axelsson A. Economic evaluation of preconcentration in production of ethanol from dilute sugar solutions. Biotechnology and Bioengineering 34, 223-233, 1989. DOI 10.1002/bit.260340211
  4. Wingren A, Galbe M, Zacchi G. Techno-economic evaluation of producing ethanol from softwood. Comparison of SSF and SHF and identification of bottlenecks. Biotechnology Progress 19, 1109-1117, 2003. DOI 10.1021/bp0340180
  5. Madson PW, Lococo DB. Recovery of volatile products from dilute high-fouling process streams. Applied Biochemistry and Biotechnology 84-86, 1049-1062, 2000. DOI link
  6. Nagy ZK. Model based control of a yeast fermentation bioreactor using optimally designed artificial neural networks. Chemical Engineering Journal 127, 95-109, 2007. DOI 10.1016/j.cej.2006.10.015
  7. Ho T. Characterization and Control of a Yeast Fermentation Reactor. Unpublished technical report, CheAI Research Inc., 2021.
  8. Aiba S, Shoda M, Nagatani M. Kinetics of product inhibition in alcohol fermentation. Biotechnology and Bioengineering 10, 845-864, 1968. DOI 10.1002/bit.260100610
  9. Medeiros DWO. DWSIM, open-source chemical process simulator, version 9.0.5. 2025. https://dwsim.org
  10. Kamihama N, Matsuda H, Kurihara K, Tochigi K, Oba S. Isobaric vapor-liquid equilibria for ethanol + water + ethylene glycol and its constituent three binary systems. Journal of Chemical & Engineering Data 57, 339-344, 2012. DOI 10.1021/je2008704
  11. Bayrock DP, Ingledew WM. Application of multistage continuous fermentation for production of fuel alcohol by very-high-gravity fermentation technology. Journal of Industrial Microbiology and Biotechnology 27, 87-93, 2001. DOI 10.1038/sj.jim.7000167
  12. Kiss AA, Ignat RM. Innovative single step bioethanol dehydration in an extractive dividing-wall column. Separation and Purification Technology 98, 290-297, 2012. DOI 10.1016/j.seppur.2012.06.029