Abstract

In this paper, we present extensions to the Anaerobic Digestion Model No. 1 (ADM1) to simulate hydrogen sulphide in biogas and solids retention efficiency. The extended model was calibrated and validated against data from a large-scale covered in-ground anaerobic reactor (CIGAR), processing sugarcane vinasse. Comparative scenarios and set-ups of a CIGAR with and without a settling tank unit (settler) were simulated to investigate the reactor's performance. Biogas flow, methane content, and yield with settler were 15,983 Nm3/d, 57%, and 0.198 Nm3CH4/kgCOD, respectively, which were 9.4%, 1.8%, and 11.64%, higher than without the settler. Improvements are combination of influent flow rate 116% higher and increased solids retention time by using a settler. The optimised modelled reactor, the volume of which was reduced by 50%, was able to produce 83% more methane per volume of reactor with half the retention time. After model calibration and validation, we assessed the quality of predictions and its utility. The overall quality of predictions was assessed as high accuracy quantitative for CH4 and medium for H2S and biogas flow. A practical demonstration of ADM1 to industrial application is presented here to identify the potential optimisation and behaviour of a large-scale anaerobic reactor, reducing, consequently, expenditure, risk, and time.

INTRODUCTION

Vinasse is a wastewater with high chemical oxygen demand (COD) and high sulfate, produced in large volumes during sugarcane ethanol distillation. Anaerobic digestion (AD) is a sustainable bioprocess to unlock the value of sugarcane vinasse (SV) as an energy feedstock through biogas recovery, from which additional bioenergy can be produced, generating electricity in gas engines, steam in boilers or even replacing diesel in sugarcane agricultural operations. There is lack of retrofitted technology from pilot to large-scale (Fuess & Zaiat 2018) and current stage of research on AD of SV is still scarce in Brazil (Moraes et al. 2015).

The breakthroughs on AD technologies to new applications are largely guided by mathematical models and to date The Anaerobic Digestion Model No. 1 (ADM1) (Batstone et al. 2002a) is the state of the art model. However, since its publication substantial extensions have been proposed to ADM1 (Batstone et al. 2006). Two examples of extensions are sulfate reduction (Fedorovich et al. 2003; Batstone 2006; Barrera et al. 2013, 2015) and decoupling solids retention time (SRT) from hydraulic retention time (HRT) to simulate high rate AD systems (Fedorovich et al. 2003; Zaher et al. 2003; Feldman et al. 2018).

The sulfate extension suggested in Batstone et al. (2006) under-predicted H2S and over-predicted volatile fatty acids, while in Fedorovich et al. (2003) it was not calibrated to predict hydrogen sulfide in biogas. These limitations were overcome in Barrera et al. (2015) where the extension was validated for sulfate-rich vinasse. It included biochemical routes for depleted hydrogen once sulfate reducing bacteria (SRB) use volatile fatty acids as source of electrons. This extension may be applied to model AD of Brazilian SV, which average SO4−2:COD ratio of 0.055, shows evidence of a sulfate-rich liquid substrate (Elia Neto et al. 2009). Ascertaining sulfate reduction dynamics in AD by modelling extension to ADM1 is a useful tool to predict undesirable H2S in biogas and costs with its removal. Unlike model validation in Barrera et al. (2015) for sulfate-rich vinasse using data from laboratory-scale experiments, we show the applicability of their study on sulfate extension to validate a model against data from industry.

Since 1950, the importance of SRT in AD is recognised as a tool to reduce anaerobic reactors size (McCarty 2001). High rate anaerobic reactors are based on decoupling SRT from HRT by promoting biomass retention within reactors (Van Lier et al. 2008). Biomass retention can be achieved by settling, attachment or granulation (Dereli et al. 2012), although the formation of granular anaerobic biomass, the core of most efficient anaerobic reactors, is occasionally impossible or unstable. In this case, drawing anaerobic biomass out of the effluent in a settling tank and recycling it back to the reactor, increases SRT irrespective of HRT (McCarty 2001; Tauseef et al. 2013). This hydraulic configuration is called anaerobic contact process (ACP) and is suitable to disperse or flocculent anaerobic biomass. The retention of biomass allows higher microorganism concentration and lower food to microorganism ratios (F/M), resulting in lower biomass and higher methane production (Low & Chase 1999; Turovskiy & Mathai 2006; Appels et al. 2008; Ruiz et al. 2011).

In the original ADM1, which is based on a continuous stirred tank reactor (CSTR), the SRT is equivalent to HRT (Batstone et al. 2002b) and is limited to biomass growth rate (Abbasi et al. 2011). To model a high rate process in ADM1, Batstone et al. (2002b) suggest the introduction of a variable to the mass outflow term of the mass balance equation of ADM1 to represent the SRT above HRT in a CSTR (Batstone et al. 2002b). Feldman et al. (2018) considered an ideal solids separation unit to model the operation of an anaerobic granular Internal Circulation reactor (Feldman et al. 2018). Zaher et al. (2003) suggest in their model a factor (fxout = SRT/HRT) multiplying the outflow term of mass balance equation which adds proportionality to the model (at constant fxout, decreasing HRT will affect SRT proportionally), efficiency adjustments (different values for different solids retention efficiencies) and shock events simulations. (Zaher et al. 2003). Kleerebezem & Loosdrecht (2006) consider that a constant SRT value for AD modelling, as suggested by Batstone et al. (2002a), may lead to unrealistic high solids concentrations and, instead, a maximum solids concentration in the reactors should be considered. Reichert (1994) presents an activated sludge model including a settler and sludge recycle, which is suitable to model solids retention in an ACP. In Reichert (1994) model, since the settler is nonideal (i.e. not all solids are retained in the settler), excessively high solids concentration on the reactor, as described by Kleerebezem & Loosdrecht (2006), is avoided.

Two ADM1 key-objectives are optimisation of plant design and operational analysis of AD systems to fulfil industry needs. (Batstone 2006; Ozkan-Yucel & Gökçay 2010; Kazadi Mbamba et al. 2016). Nevertheless, the literature on the application of ADM1 to industrial case studies is scarce (Dereli 2019) with a few extended benefit-cost analysis.

Modelling sulfate reduction and hydraulic variation in an AD system would address operational analysis and critical issues on design, showing what benefits and limitations are available from industrial application of ADM1.

A recent case study on ADM1 to cover the gaps of its industrial application was published in Elaiuy et al. (2018). In this work, the authors calibrated and validated the model of a large-scale covered in-ground anaerobic reactor (CIGAR) in Brazil, which processes SV to produce biogas and generate bioelectricity for supply to the local grid. However, neither model extension nor optimisation alternatives were explored in their model.

Thus, the novelty of this paper is to present a further step to the work of Elaiuy et al. (2018) towards the neglected sulfate extension and investigate possible design optimisation by evaluating: (i) sulfate reduction processes in AD of SV and predictions of H2S in biogas; (ii) extension to ADM1 to incorporate a settler and sludge recycle for the improvement of biomass retention efficiency in the CIGAR reflecting on biogas production and methane yields; (iii) potential CAPEX and OPEX savings.

METHODS

CIGAR configuration and operation

The modelled CIGAR is one of the components of a biogas plant. It is an anaerobic reactor built in 2010 on a sugarcane mill in Brazil. Soil excavation was carried out to fit in-ground the 15,000 m3 CIGAR, which was lined and covered with high density polyethylene (HDPE) to store biogas underneath a headspace of 4,800 m3. Inside the CIGAR, vertical HDPE baffles divided the reactor into three communicating chambers. Chamber 1 (C1 – 60% of total reactor volume) was fed with a mixture of raw vinasse and effluent from C1 (recycle stream) mixed in an external mix tank. This mixture was pumped upwards through pipes at the bottom of the reactor, as in a typical upflow anaerobic sludge blanket (UASB) reactor. The remaining volume of the reactor was split into two chambers, chamber 2 (C2) and chamber 3 (C3). C2 was fed only with effluent from C1, and C3 was designed to settle the anaerobic sludge, which may be recycled back to C1.

The CIGAR was designed to operate under mesophilic conditions (37 °C), 39.5 m3/h flow rate of SV, 1.99 kgCOD/m3/d organic loading rate (OLR), 15 days HRT, 0.227 Nm3CH4/kgCOD methane yield to produce 491 Nm3/h of biogas with 55% CH4.

A schematic layout of the reactor is presented in Figure 1.

Figure 1

CIGAR – original and proposed configurations.

Figure 1

CIGAR – original and proposed configurations.

Analytical methods

Data were collected from CIGAR operation over the harvest season of 2012 (Season_12) from May to December for 220 days. Eventually, data were not available due to operational problems. Physico-chemical analyses were carried out by plant operators according to the protocols described by Standard Methods for the Examination of Water and Wastewater (APHA 2005), in an on-site laboratory and included analysis of influent, recycle, and effluent streams (Table 1). Biochemical characterisation of the substrate as carbohydrates, lipids, and proteins was performed following the analytical methods described in Elaiuy et al. (2018).

Table 1

Physico-chemical analyses, method, and frequency

ParameterMethodFrequency
Temperature APHA 2550 Continuously (online) 
pH APHA 4500 Continuously (online) for mixed influent, daily for other streams 
COD concentration APHA 5,220 Daily 
Total solids (TS) APHA 2,540B Weekly 
Total volatile solids (TVS) APHA 2,540E Weekly 
Total suspended solids (TSS) APHA 2,540D Weekly 
Volatile suspended solids (VSS) APHA 2,540E Daily 
Volatile fatty acids (VFA) APHA 5,560 Daily 
Partial alkalinity APHA 2,320 Daily 
Total Kjeldahl nitrogen (TKN) APHA 4,500 Biweekly 
Total ammonia nitrogen (TAN) APHA 4,500 Biweekly 
Sulfate (SO4−2Turbidimetric (APHA 9,038) Randomly 
ParameterMethodFrequency
Temperature APHA 2550 Continuously (online) 
pH APHA 4500 Continuously (online) for mixed influent, daily for other streams 
COD concentration APHA 5,220 Daily 
Total solids (TS) APHA 2,540B Weekly 
Total volatile solids (TVS) APHA 2,540E Weekly 
Total suspended solids (TSS) APHA 2,540D Weekly 
Volatile suspended solids (VSS) APHA 2,540E Daily 
Volatile fatty acids (VFA) APHA 5,560 Daily 
Partial alkalinity APHA 2,320 Daily 
Total Kjeldahl nitrogen (TKN) APHA 4,500 Biweekly 
Total ammonia nitrogen (TAN) APHA 4,500 Biweekly 
Sulfate (SO4−2Turbidimetric (APHA 9,038) Randomly 

Influent flow was measured continuously using a magnetic flow meter OPTIFLUX KC1000F/6 (Krohne) with IFC100 signal converter. Biogas flow was continuously measured using a thermal dispersion mass flow meter FT-2 (Clontech). H2S was measured daily with precision detection tubes (Kitagawa manual pump model AP20, tubes model 120SH – range of 0.1–4% H2S). CH4 was measured 6 times a day with a handheld GEM2000 biogas analyser (Landtec).

Soluble COD removal efficiency

Theoretical oxygen demand (ThOD) of 1 kg of microorganisms as volatile suspended solids (VSS) with an estimated composition of C5H7O2N can be calculated as 1.42 kgCOD/kgVSS (Van Lier et al. 2008). Subtracting the ThOD of VSS from the total COD results in a proxy soluble COD. Thus, the proxy of soluble COD can be calculated by the following equation:
formula
(1)
COD removal efficiency for the proxy soluble COD can be calculated as follows:
formula
(2)
Similarly, total COD removal can be calculated for effluent as follows:
formula
(3)
where ‘i’ refers to recycle or effluent stream.

During Season_12, high values of VSS in effluent and recycle, from chambers 3 and 1, respectively, denoted anaerobic biomass being washed out from both chambers, which was confirmed by Imhoff cone settleable solids tests. The comparison of proxy soluble COD removal efficiency between effluent and recycle, based on operational data from Season_12, will be presented and discussed further on to justify possible optimisation of plant design.

Settler – preliminary test

In 2012, a small settler was connected to the CIGAR to evaluate effluent solids reduction. Its overall performance was an average retention of 53% volatile solids (VS) loaded and 54% COD loaded in a concentrated stream of thickened sludge, corresponding to 20% of settler inflow. Average VS of thickened sludge was 13,792 mg/L, 2.6 times higher than the VS of settler inflow (i.e., CIGAR effluent). Similarly, average COD of thickened sludge was equal to 39,430 mg/L, 2.7 times higher than the settler inflow COD.

Based on this preliminary test, using a settler to clarify the effluent from the CIGAR and return the thickened sludge back to the reactor, would increase the SRT and biomass concentration in the CIGAR. By doing so, it is expected higher methane yield, lower biomass production, and better effluent quality. Besides, we assume that a settler would be more efficient than C3, which was designed for the same purpose, simplifying reactor design, construction, and operation. This assumption is investigated in this paper by the inclusion of a settler and sludge recycle to ADM1 and will be discussed later. Figure 1 presents a schematic flow diagram for the original condition and proposed configuration, along with a simplified mass balance.

Modelling framework

The model under study, implemented in Aquasim 2.1G, combines two updates to the previous model described in Elaiuy et al. (2018): the sulfur chain reactions and the sludge retention and recycle. The first update evaluates the H2S in biogas to better assess its concentration for suitable biogas cleaning process, where Season_12 data were used to calibrate and validate the model. The second update simulates the overall performance of the CIGAR with improved solids retention and possible design modifications, aiming at cost reduction and improved performance, where typical vinasse COD values were used for simulations with the previously validated model.

Sulfur chain extension in ADM1

Original ADM1 does not include the sulfur chain processes (Batstone et al. 2002b). To account to this chain of reactions, three species of sulfur reducing bacteria (SRB) were added to the model, propionate sulfate reducing bacteria (pSRB), acetate sulfate reducing bacteria (aSRB), and hydrogenotrophic sulfate reducing bacteria (hSRB), including their respective substrates uptake and decay processes.

A simplified extension including only hSRB would not consider volatile fatty acids (VFA) consumption by SRB, limiting the assessment of their impact on methane production (Batstone et al. 2006). On the other hand, included in the model extension, butyrate and valerate seemed unnecessary due to their low concentration in the AD of SV under study, adding unnecessary complexity to the model (Barrera et al. 2015).

Also, three acid/base dissociation process (H2SO4, HSO4 and H2S) and H2S gas transfer from liquid to gas phase process were added to the original ADM1.

SRBs considered consume, each one, all together with sulfate, hydrogen, propionate and acetate.

Reaction rates were considered as dual Monod kinetics, where rates are simultaneously dependent on two substrates concentration: hydrogen, propionate, and acetate for each one of the species considered and sulfate concentration for all SRB.

SRB kinetic and stoichiometric parameters were adopted from Barrera et al. (2015) and Solon (2015), respectively. The following reactions were added, as shown below.
formula
(4)
formula
(5)
formula
(6)

Sulfide produced by sulfate reduction is found in the gas and liquid stream. In the liquid stream, it may be undissociated (H2S) or dissociated (HS) in proportions defined by dissociation constant, Henry's Law constant and environmental conditions, including reactor pressure, temperature and pH. Acid-base equilibrium for both H2S and H2SO4 were implemented as algebraic equations, with the two dissociations of H2SO4 included, as suggested by Knobel & Lewis (2002).

Sulfur was introduced to the model as an input variable, read according to available data for sulfate. The statistical mode was used when data for sulfate were rather sparse. By doing so, misled SO4−2 values interpolated by Aquasim were avoided. H2S transfer to gas phase was implemented similarly to other gases (CH4, CO2 and H2), with gas constants from Sander (1999).

Inhibition by free H2S was considered for hydrogen, acetate, butyrate, propionate, and valerate degrading organisms, as well as for all SRB. The formulation used was presented in Fedorovich et al. (2003) as a first order inhibition kinetics. For ease of calculation, only one constant for all bacterial groups affected by H2S inhibition was used.

The model including sulfur consists of 25 biochemical process, 11 of them subject to one or more inhibition process, 9 the acid-base equilibrium process, and 4 the liquid-gas transfer physical process, implemented as differential and algebraical equations system in a completely stirred tank reactor composed of a liquid phase and a headspace.

Parameter estimation and model validation procedure

The model incorporating the sulfur chain processes includes the maximum substrate uptake rate (MSUR) and half saturation constant (HSC) of the three SRB groups considered.

Before parameter estimation, the model was tested with the same set of parameters that describes the SRB groups in Barrera et al. (2015) against dataset from Season_12. After visual inspection of results, as suggested by Donoso-Bravo et al. (2011), simulations for biogas and methane agreed well with measured data. However, deviations were noted between simulated and measured values for H2S during the first 70 days and good correlation for the remaining days (data not shown). For this reason, from 220 days of operational data of Season_12, the first 70 days were discarded.

The remaining 150 days of Season_12 were split into two subsets, one for parameter estimation and the other one for cross validation.

The first subset of data, used for parameter estimation, ranged from day 70 to day 150. The initial 10 days (70–80) were used as a ramp and did not account to model error calculations. The second subset of data, used for cross validation, ranged from day 140 to day 220 and again, the first 10 days (140–150) used as a ramp were discarded for error calculations.

SRB constants (MSUR and HSC for acetate hydrogen, propionate, and sulfate) were estimated by Aquasim built-in function for parameter estimation. The function target was to minimise the difference between simulated and measured H2S concentration in biogas.

With respect to model accuracy, the relative absolute error (rAE) was used to evaluate deviations between measured and simulated values.

rAE is calculated following the equation below:
formula
(7)
where is the ith measured value, is the model prediction at the time corresponding to data point , which by its turn is a function of the set of parameters to be estimated, and is the number of observations.

For rAE results, we adopted the following classification from Batstone et al. (2002a), thereby qualifying the model quantitative predictions as high accuracy (rAE ± 10%) and medium accuracy (rAE 10–30%).

Sludge retention

Model inputs and description

Influent COD was set constant to 31.5 kgCOD/m3, adopted from Elia Neto et al. (2009) who have reported an average characterisation of vinasse from 20 ethanol mills across Brazil. The main idea behind setting this value from another author was to assume an average COD for Brazilian SV, in order to avoid specific mill-type influent and project a potential national biogas production.

Disintegration of particulate matter (originated from influent and decayed bacterial biomass) was considered as first order kinetics reaction, resulting in carbohydrates, proteins, lipids and inerts (soluble and particulate). Substrate was biochemically fractionated as 44% carbohydrates (fch), 30% proteins (fpr), and 26% lipids (fli), according to Elaiuy et al. (2018). Although particulate composites vary in time, fractions of carbohydrates, proteins, and lipids were assumed constant in the simulations. Besides, their hydrolysis constant was kept the same, 0.66 d−1 for all three components.

The degradation extent (fd) describing the degradable ThOD fraction of substrate converted to methane was set to 50% as estimated in Elaiuy et al. (2018). The kinetic parameters to model sulfate reduction process in the AD of SV were initially kept the same as in Barrera et al. (2015). Apart from these, stoichiometric and kinetic parameters were based on the work of Rosen & Jeppsson (2006). The charge balance of the influent was determined from measured values of influent pH, VFA concentration, inorganic nitrogen (calculated as TAN), and inorganic carbon (calculated through partial alkalinity measurements) (Elaiuy et al. 2018).

Settler and sludge recycle modelling

To simulate the inflow of sludge recycle from the settler back to the CIGAR an advective link with a bifurcation was implemented in Aquasim, as suggested in Reichert (1994). To model the settler, a variable (recircX) defining the solids retention efficiency was implemented in Aquasim and set constant to 43% (10% lower than the real capacity of the tested settler) to allow some uncertainty of model predictions. This variable multiplies each solids concentration in the CIGAR's effluent, calculating the solids mass flow rate driven to recycle stream. To simplify the implementation the advective link was set only to transport solids. With this modelling structure, there is no need to calculate actual SRT, which is difficult in in-ground anaerobic reactors, where sampling the sludge bed is not practical or accurate.

Proposed CIGAR set-ups

The 40% volume of the CIGAR occupied by C2 and C3 were responsible for only 21% of total biogas over Season_12. These percentages were calculated comparing COD removal between recycle and effluent streams.

This low performance may be due to inefficient biomass retention and to address this problem the following operational set-ups (Setup1, Setup2, Setup3) for the CIGAR were modelled considering the applicability in full-scale system. The optimisation study is based on Setup3.

Setup1. Chambers are lumped together as a single reactor;

Setup2. Chambers are lumped together as a single reactor and connected to an external settler;

Setup3. Volume of the single reactor is reduced by 50% and connected to an external settler.

Modelled scenarios

Setup1 and Setup2 were subject to six different scenarios (SC1 to SC6), each one at higher flow rate (Table 2). By doing so, we could verify the collapse of the reactor due to biomass washout by increasing the flow rate and assess the performance of Setup1 and Setup2. In each scenario, an initial flow rate ramp of 10 days was undertaken to gradually increase the biomass concentration as the flow increases in a dynamic equilibrium. Therefore, as soon as the flow rate becomes constant, the system will have reached a steady state condition. If the ramp is too steep, chances are the reactor will collapse because biomass growth rate does not follow the increased flow rate.

Table 2

Characteristics of modelled scenarios

ScenariosInfluent flow rate (after 10 days ramp) [m3/d]HRT [d]Increase in influent flow rate compared to nominal value
SC1 948 15.2 0% 
SC2 1,090 13.2 15% 
SC3 1,232 11.7 30% 
SC4 1,417 10.2 50% 
SC5 1,514 9.5 60% 
SC6 2,050 7.0 116% 
ScenariosInfluent flow rate (after 10 days ramp) [m3/d]HRT [d]Increase in influent flow rate compared to nominal value
SC1 948 15.2 0% 
SC2 1,090 13.2 15% 
SC3 1,232 11.7 30% 
SC4 1,417 10.2 50% 
SC5 1,514 9.5 60% 
SC6 2,050 7.0 116% 

After the 10-day-ramp, each scenario had a constant flow over 210 days, resulting in a total of 220 days of CIGAR operation for each scenario (average operation period of sugarcane mills in Brazil).

Steady state simulations for methane yield, biogas flow and CH4 were performed in all set-ups and scenarios. Therefore, variations in biogas production and its composition could be attributed to biomass retention effects and not to other variables.

RESULTS

CIGAR monitoring

Characteristics of influent, effluent, recycle streams, and biogas composition covering 220 days are shown in Table 3. The average of each dataset was calculated ± one standard deviation. The varying characteristics can be noted by the dispersion of datasets relative to their average, indicating dynamic behaviour of loading conditions. It is noteworthy that SV characteristics vary over Season_12, depending on production schedules of ethanol and sugar by the sugarcane mill.

Table 3

Characteristics of influent, effluent, recycle streams, and biogas composition during entire season_12

MinimumAverageStandard deviationMaximum
Influent total COD (kg/m34.30 30.05 10.64 71.61 
Effluent total COD (kg/m32.51 16.17 9.87 52.46 
Effluent VSS (kg/m30.40 4.58 3.52 16.70 
Effluent soluble COD (kg/m30.84 9.36 7.54 33.61 
Recycle total COD (kg/m32.39 20.55 10.04 52.88 
Recycle VSS (kg/m30.40 6.44 4.53 20.90 
Recycle soluble COD (kg/m30.56 10.55 7.00 28.02 
Influent (g SO4−2 /m3250 1,357 579 2,750 
Influent SO4−2: COD 0.01 0.05 0.03 0.18 
H2S (ppm) 7,500 14,856 3,889 30,000 
CH4 (%) 31.5 57.2 6.5 77.5 
MinimumAverageStandard deviationMaximum
Influent total COD (kg/m34.30 30.05 10.64 71.61 
Effluent total COD (kg/m32.51 16.17 9.87 52.46 
Effluent VSS (kg/m30.40 4.58 3.52 16.70 
Effluent soluble COD (kg/m30.84 9.36 7.54 33.61 
Recycle total COD (kg/m32.39 20.55 10.04 52.88 
Recycle VSS (kg/m30.40 6.44 4.53 20.90 
Recycle soluble COD (kg/m30.56 10.55 7.00 28.02 
Influent (g SO4−2 /m3250 1,357 579 2,750 
Influent SO4−2: COD 0.01 0.05 0.03 0.18 
H2S (ppm) 7,500 14,856 3,889 30,000 
CH4 (%) 31.5 57.2 6.5 77.5 

Soluble and total COD values for recycle (C2) and effluent (C3), in Table 3, indicate that effluent total COD is 21% lower than recycle total COD, and effluent soluble COD is 11% lower than recycle soluble COD.

Total COD removal efficiency for recycle and effluent streams were 32% and 46%, respectively. Soluble COD removal efficiency for recycle and effluent streams were 65% and 69%, respectively, indicating high soluble COD removal efficiency for both streams.

Assuming that the difference between total and soluble COD is due to anaerobic biomass washout, we presume that increased solid retention using a settler could improve the CIGAR overall performance. A settler could replace C2 and C3 in a more efficient way, reducing by 40% the volume of the CIGAR, whilst improving its overall performance. Furthermore, both chambers add constructive complexity that affects CAPEX and OPEX due to extra pumps, valves, pipes and HDPE baffles.

Sulfur chain modelling

Direct and cross validation

The kinetic parameters to model sulfate reduction process in the AD of SV (Table 4) were estimated against the first aforementioned dataset. The results of estimated parameters are mostly similar to Barrera et al. (2015), except for Km_aSRB, Ks_aSRB, Km_pSRB, and Ks_pSRB. A primary driver for these differences is the SO4−2:COD ratio. Whilst this ratio averaged 0.0502 in this work, a higher ratio was found in Barrera et al. (2015). There was competition for acetate between SRB and methanogens as long as inhibition related to COD:SO4 ratio was not severe (COD:SO4 < 0.1) (Barrera et al. 2013). The uptake of acetate and neglected propionate by methanogens reflected in lower Km_aSRB and higher Ks_aSRB. Nonetheless, Ks_SO4_aSRB, Ks_SO4_hSRB, and Ks_SO4_pSRB, which are needed to predict SSO4, are similar to parameters from Barrera et al. (2015).

Table 4

Comparison of kinetic parameters reported in Barrera et al. (2015) and estimated in this work

ParameterDescriptionBarrera et al. (2015) Estimated in this work
Km_aSRB MSURa for acetate SRB 18.5 4.47 
Km_hSRB MSURa for hydrogen SRB 63 65.0 
Km_pSRB MSURa for propionate SRB 23 30.0 
Ks_aSRB Acetate HSCb for acetate SRB 0.120 0.0201 
Ks_hSRB Hydrogen HSCb for hydrogen SRB 6E-06 5.41E-06 
Ks_pSRB Propionate HSCb for propionate SRB 0.110 0.0160 
Ks_SO4_aSRB Sulfate HSCb for acetate SRB 0.001 0.00100 
Ks_SO4_hSRB Sulfate HSCb for hydrogen SRB 0.00105 0.00105 
Ks_SO4_pSRB Sulfate HSCb for propionate SRB 0.002 0.00200 
ParameterDescriptionBarrera et al. (2015) Estimated in this work
Km_aSRB MSURa for acetate SRB 18.5 4.47 
Km_hSRB MSURa for hydrogen SRB 63 65.0 
Km_pSRB MSURa for propionate SRB 23 30.0 
Ks_aSRB Acetate HSCb for acetate SRB 0.120 0.0201 
Ks_hSRB Hydrogen HSCb for hydrogen SRB 6E-06 5.41E-06 
Ks_pSRB Propionate HSCb for propionate SRB 0.110 0.0160 
Ks_SO4_aSRB Sulfate HSCb for acetate SRB 0.001 0.00100 
Ks_SO4_hSRB Sulfate HSCb for hydrogen SRB 0.00105 0.00105 
Ks_SO4_pSRB Sulfate HSCb for propionate SRB 0.002 0.00200 

aMaximum substrate uptake rate.

bHalf saturation constant.

Since other parameters used in the present model were calibrated in previous work for AD of SV, only MSUR and HSC of SRB were selected for calibration.

The simulations shown in Figure 2(a) are split into direct and cross validation of the model for H2S and there is a good fit between simulated and measured values on both sides. The quality of prediction was sensitive to a few kinetic parameters: Km_aSRB, Ks_aSRB, Km_pSRB, and Ks_pSRB. Notwithstanding, direct validation shows higher quality of prediction (rAE 13.3%) than cross validation (rAE 17.4%), both are in the range of medium (10–30%) accurate quantitative prediction. The same quality of prediction for H2S was found in Barrera et al. (2015). Furthermore, given the scale and the uncontrolled environment of this work, rAE for both direct and cross validations can be considered satisfactory when compared to other studies, which reported similar rAE under laboratory conditions.

Figure 2

rAE of measured and simulated (a) H2S, (b) biogas flow and (c) CH4 – direct and cross validation.

Figure 2

rAE of measured and simulated (a) H2S, (b) biogas flow and (c) CH4 – direct and cross validation.

Figure 3

(a) Methane yield, (b) methane content, (c) biogas flow for modelled scenarios and setups.

Figure 3

(a) Methane yield, (b) methane content, (c) biogas flow for modelled scenarios and setups.

Simulated H2S concentration in biogas ranged from 0.94% to 2.01% and is close to ranges found in the literature using SV. For example, 1.5–3.0%, was reported in Cortes Pires et al. (2015); Leme & Seabra (2017); Nandy et al. (2002); Yasar et al. (2015) and 1.25–1.75% in Barrera et al. (2015).

Despite the variability in the influent composition over Season_12, the simulation is consistent after parameter estimation (Figure 2(a)). However, the quality of prediction was less accurate in the cross validation but yet classified as medium accuracy.

Most of H2S was overestimated by the model. 55% and 62% of simulated values were higher than measured values for direct and cross validation, respectively.

Also, 57% and 46% of simulated values had errors within +/−10% for direct and cross validation, respectively. However, if we consider the season average H2S content in biogas the difference between measured and simulated values was 3.08% for the first subset and 8.91% for the second subset.

The rAE for biogas flow in the direct validation of this work was 12.7% (Figure 2(b)), smaller than in Elaiuy et al. (2018), possibly because of free H2S inhibition to microorganisms, which was the highest among all inhibitions (data not shown).

Similar quality of predictions for biogas flow were presented in Dereli (2019), simulating a full-scale anaerobic reactor.

The same rAE of 6.1% for CH4 was found in the direct and cross validation, which can be visually confirmed by small deviation between simulated and measured outputs of CH4 in both sides of Figure 2(c). Nevertheless, this rAE is higher than the rAE found in Elaiuy et al. (2018), possibly because of the use of methane precursors in the H2S production instead of CH4 production.

Additionally, compared calculated averages for days 80–220 of simulated and measured biogas flow and CH4 show differences of −4.9% and −2.3%, respectively.

Sludge recycle – modelled scenarios

Simulation results of Setup1 and Setup2 for biogas flow, methane content, and methane yield are illustrated in Table 5. Comparing results, it was found that Setup2 had better performance than Setup1 in all simulated scenarios.

Table 5

Results of biogas flow, methane content, and methane yield from different scenarios and set-ups under study

ScenarioSetup1 (Without settler)
Setup2 (With settler)
Comparison Setup2 × Setup1
Biogas flow [Nm3/d]Methane content [%]Methane yield [Nm3CH4/kgCOD]Biogas flow [Nm3/d]Methane content [%]Methane yield [Nm3CH4/kgCOD]Biogas flowMethane contentMethane yield
SC1 9,858 56.8 0.193 10,593 57.2 0.208 7.5% 0.7% 7.5% 
SC2 11,125 56.7 0.190 11,951 57.2 0.205 7.4% 0.8% 8.0% 
SC3 12,411 56.6 0.186 13,471 57.1 0.203 8.5% 1.0% 8.8% 
SC4 14,047 56.3 0.181 15,287 57.0 0.200 8.8% 1.4% 10.4% 
SC5 14,611 56.0 0.178 15,983 57.0 0.198 9.4% 1.8% 11.6% 
SC6 Collapse 20,958 56.6 0.189 N/A 
ScenarioSetup1 (Without settler)
Setup2 (With settler)
Comparison Setup2 × Setup1
Biogas flow [Nm3/d]Methane content [%]Methane yield [Nm3CH4/kgCOD]Biogas flow [Nm3/d]Methane content [%]Methane yield [Nm3CH4/kgCOD]Biogas flowMethane contentMethane yield
SC1 9,858 56.8 0.193 10,593 57.2 0.208 7.5% 0.7% 7.5% 
SC2 11,125 56.7 0.190 11,951 57.2 0.205 7.4% 0.8% 8.0% 
SC3 12,411 56.6 0.186 13,471 57.1 0.203 8.5% 1.0% 8.8% 
SC4 14,047 56.3 0.181 15,287 57.0 0.200 8.8% 1.4% 10.4% 
SC5 14,611 56.0 0.178 15,983 57.0 0.198 9.4% 1.8% 11.6% 
SC6 Collapse 20,958 56.6 0.189 N/A 

Maximum influent flow rate for SC5 and SC6 were respectively 60% and 116% higher than the nominal reactor design flow rate, meaning above that, imminent biomass washout and reactor collapse. The maximum influent flow rate for Setup2 was higher than for Setup1 before reactor collapse. This is attributed to better biomass retention in Setup2, since it is 52% higher than in Setup1 regarding SC1 (data not shown). We assume that the influent flow rate could be even higher as long as the settler has an improved performance.

It is important to highlight that for Setup1 and Setup2 the stepwise increasing flow rate in each SC, results in a reduction in HRT followed by lower methane yield (Figure 3(a)) and methane content (Figure 3(b)). Nevertheless, these reductions were smaller for Setup2 than for Setup1.

Simulated methane yield of 0.193 Nm3CH4/kgCOD for Setup1 was similar to 0.196 Nm3CH4/kgCOD measured in 2012. These figures are smaller than CIGAR projected methane yield of 0.227 Nm3CH4/kgCOD. However, the amount of methane recovered per kilogram of COD removed is more realistic in the simulations since we adopted 50% of COD removal compared to nominal CIGAR design of 65%.

Average methane yield of 0.225 Nm3CH4/kgCOD for large-scale projects has been reported in Silva Neto et al. (2019). Another possible explanation to this discrepancy may be attributed to either higher degradation of substrates or superior SRT. To verify the latter, we simulated a settler with 80% solids retention efficiency and CIGAR influent flow rate of 948 m3/d, resulting in a methane yield of 0.221 Nm3CH4/kgCOD. This result is 6% higher than simulations where the settler was set to 43% of solids retention efficiency and only 4% lower than average methane yield reported in the literature.

The biogas flow was higher for Setup2 than for Setup1 (Figure 3(c)) as a result of increased biomass concentration, thus lower food to microorganism ratio (F/M).

The OLR was 4.5 kgCOD/m3/d for SC6 and 3.3 kgCOD/m3/d for SC5, which are close to values found in Wilkie et al. (2000), ranging between 4.6 and 5 kgCOD/m3/d for ACP digesting sugarcane vinasse. Also, OLRs for SC5 and SC6 were higher than the nominal CIGAR design and Season_12 measured values of 1.99 kgCOD/m3/d and 2.06 kgO2/m3/d, respectively.

CIGAR economics – Setup3 analysis

A comparative analysis between Setup1 and Setup3 after simulations is presented in Table 6. The reactor for Setup3 is 50% smaller than Setup1, and yet it was able to produce 83% more methane per cubic meter of reactor at half of HRT of Setup1. However, the methane yield and total biogas flow rate dropped by 7.8% and 8.3%, respectively. Notwithstanding, Setup3 has higher biogas production per volume of reactor, it is not possible to achieve the same biogas production as in Setup1. Even though the OLR is higher for Setup 3, the methane yield is smaller.

Table 6

Summarised comparison between original (Setup1) and optimised reactor (Setup3)

ParameterUnitOriginal reactor (Setup1)Optimised reactor (Setup3)Comparison optimised x original
HRT [d] 15.2 7.6 −50% 
Biogas flow rate [Nm3/d] 10,593 9,781 −8.30% 
Methane content [%] 57.2 56.8 −0.70% 
Methane yield [Nm3CH4/kgCOD] 0.208 0.191 −7.80% 
Methane production [Nm3CH4/m3reactor] 0.421 0.771 83% 
Total volume earthworks [m320,304 m3 10,867 m3 −46% 
HDPE liner/cover area [m²] 10,204 m² 5,109 m² −50% 
Total pipework [m] 757 m 345 m −54% 
Pumping system  4 pumps, 2 flow meters, 4 VSDa 2 pumps, 1 flow meter, 2 VSDs −50% 
ParameterUnitOriginal reactor (Setup1)Optimised reactor (Setup3)Comparison optimised x original
HRT [d] 15.2 7.6 −50% 
Biogas flow rate [Nm3/d] 10,593 9,781 −8.30% 
Methane content [%] 57.2 56.8 −0.70% 
Methane yield [Nm3CH4/kgCOD] 0.208 0.191 −7.80% 
Methane production [Nm3CH4/m3reactor] 0.421 0.771 83% 
Total volume earthworks [m320,304 m3 10,867 m3 −46% 
HDPE liner/cover area [m²] 10,204 m² 5,109 m² −50% 
Total pipework [m] 757 m 345 m −54% 
Pumping system  4 pumps, 2 flow meters, 4 VSDa 2 pumps, 1 flow meter, 2 VSDs −50% 

aVSD = Variable Speed Driver.

By reducing the size of the CIGAR and simplifying its design, as shown in Setup 3, it will have an impact on CAPEX planned. Since the majority of capital expenditures occur upfront in the construction stage, costs allocated to land, material, and equipped facilities, for example, will be significantly reduced. OPEX will follow the same trend, including, for example, electricity consumption, staffing, and general overhead. In summary, the optimised CIGAR is a proposed solution that minimises capital and operational expenditures. This involves redesigning not only the reactor but its operation.

Whilst there is certainly scope for potential CAPEX and OPEX savings given the size and complexity reductions, in this work we did not take into account expenditures involving the settler.

CONCLUSIONS

Vinasse is a high COD, high sulfate, low pH, seasonally produced wastewater and biogas production from it should be done with special attention in planning, design and operation.

A practical demonstration of ADM1 to industrial application is presented here to identify the potential optimisation and behaviour of a large-scale anaerobic reactor processing SV, reducing, consequently, expenditure, risk, and time.

The modelled higher SRT optimised reactor showed higher biogas production and methane per volume of reactor, followed by a reduced HRT from 15 to 7 days.

By adding an external settler with sludge return substantial savings in materials and services associated with a lagoon-type digestor construction and operation costs can be obtained. Moreover, methane yield, methane concentration and biogas production can be higher when SRT is higher.

The predictions of H2S levels in biogas by ADM1, based on sulfate and COD content, is a useful tool to assess biogas composition, especially for projects where the gas is not yet under production, but wastewater composition is available.

The quality of predictions of the model allows practitioners and designers of vinasse-to-energy projects to anticipate with reasonable accuracy the H2S levels in biogas and plan ahead appropriate biogas downstream processing and technology to convert biogas into clean renewable bioenergy. The H2S model presented small differences between the averages of modelled results and large-scale reactor measured data. The model can be qualified as medium accuracy based on rAE, although the majority of calculated values were within a +/− 10% error range. The model also improves the accuracy in prediction of energetic value of biogas by reducing errors in biogas flow and CH4 content compared to previous models.

Electricity or fuel (as biomethane) produced from vinasse biogas are not the subject of any premium price or incentives, and treatment of sugarcane vinasse is not compulsory for mills in Brazil. Upon these circumstances, the actual scenario of vinasse-to-biogas projects will only be profitable if special attention is given to mathematical models such as ADM1 for evaluation, optimisation, and design of existing and planned biogas plants.

REFERENCES

Abbasi
T.
Tauseef
S. M.
Abbasi
S. A.
2011
Biogas Energy
.
Springer Science & Business Media
,
London, UK
.
APHA
2005
Standard Methods for the Examination of Water and Wastewater
, 21st edn.
American Public Health Association
,
Washington, DC, USA
.
Appels
L.
Baeyens
J.
Degrève
J.
Dewil
R.
2008
Principles and potential of the anaerobic digestion of waste-activated sludge
.
Prog. Energy Combust. Sci.
34
,
755
781
.
https://doi.org/10.1016/j.pecs.2008.06.002
.
Barrera
E. L.
Spanjers
H.
Dewulf
J.
Romero
O.
Rosa
E.
2013
The sulfur chain in biogas production from sulfate-rich liquid substrates: a review on dynamic modeling with vinasse as model substrate
.
J. Chem. Technol. Biotechnol.
88
,
1405
1420
.
https://doi.org/10.1002/jctb.4071
.
Batstone
D. J.
2006
Mathematical modelling of anaerobic reactors treating domestic wastewater: rational criteria for model use
.
Rev. Environ. Sci. Biotechnol.
5
,
57
71
.
https://doi.org/10.1007/s11157-005-7191-z
.
Batstone
D. J.
Keller
J.
Angelidaki
I.
Kalyuzhnyi
S. V.
Pavlostathis
S. G.
Rozzi
A.
Sanders
W. T. M.
Siegrist
H.
Vavilin
V. A.
2002a
The IWA anaerobic digestion model No. 1 (ADM1)
.
Water Sci. Technol.
45
,
65
73
.
https://doi.org/10.2166/wst.2002.0292
.
Batstone
D. J.
Keller
J.
Angelidaki
I.
Kalyuzhnyi
S. V.
Pavlostathis
S. G.
Rozzi
A.
Sanders
W. T. M.
Siegrist
H. A.
Vavilin
V. A.
2002b
Anaerobic digestion model no. 1 (ADM1), IWA task group for mathematical modelling of anaerobic digestion processes
.
Water Sci. Technol.
45
(
10
),
65
73
.
Batstone
D. J.
Keller
J.
Steyer
J. P.
2006
A review of ADM1 extensions, applications, and analysis: 2002–2005
.
Water Sci. Technol.
54
,
2002
2005
.
https://doi.org/10.2166/wst.2006.520
.
Cortes Pires
J. R.
Nour
E. A. A.
Barbosa
P. S. F.
Almeida
P. C.
Lopes
L. R. O.
2015
Geração de Eletricidade a Partir da Biodigestão da Vinhaça e o Potencial para o Brasil (Electricity Generation from Vinasse Biodigestion and the Potential for Brazil). In: VIII CITENEL. p. 8
.
Dereli
R. K.
2019
Modeling long-term performance of full-scale anaerobic expanded granular sludge bed reactor treating confectionery industry wastewater
.
Environ. Sci. Pollut. Res.
26
,
25037
25045
.
https://doi.org/10.1007/s11356-019-05739-1.
Dereli
R. K.
Ersahin
M. E.
Ozgun
H.
Ozturk
I.
Jeison
D.
van der Zee
F.
van Lier
J. B.
2012
Potentials of anaerobic membrane bioreactors to overcome treatment limitations induced by industrial wastewaters
.
Bioresour. Technol.
122
,
160
170
.
https://doi.org/10.1016/j.biortech.2012.05.139
.
Donoso-Bravo
A.
Mailier
J.
Martin
C.
Rodríguez
J.
Aceves-Lara
C. A.
Wouwer
A. V.
2011
Model selection, identification and validation in anaerobic digestion: a review
.
Water Res.
45
,
5347
5364
.
https://doi.org/10.1016/j.watres.2011.08.059
.
Elaiuy
M. L. C.
Borrion
A. L.
Poggio
D.
Stegemann
J. A.
Nour
E. A. A.
2018
ADM1 modelling of large-scale covered in-ground anaerobic reactor treating sugarcane vinasse
.
Water Sci. Technol.
77
(
5
),
1397
1409
.
https://doi.org/10.2166/wst.2018.013
.
Elia Neto
A.
Shintaku
A.
Pio
A. A. B.
Conde
A. J.
Gianetti
F.
Donzelli
J. L.
2009
Manual de conservação e reuso de água na agroindustria sucroenergetica (Water Conservation and Reuse Manual in Sugar-Energy Agroindustry)
.
ANA, FIESP, UNICA, CTC
,
Brasília, Brazil
.
Fedorovich
V.
Lens
P.
Kalyuzhnyi
S.
2003
Extension of anaerobic digestion model No. 1
.
Appl. Biochem. Biotechnol.
109
,
33
45
.
https://doi.org/10.1385/ABAB:109:1-3:33
.
Feldman
H.
Alsina
X. F.
Kjellberg
K.
Jeppsson
U.
Batstone
D. J.
Gernaey
K. V.
2018
Model-based analysis and optimization of a full-scale industrial high-rate anaerobic bioreactor
.
Biotechnol. Bioeng.
115
,
2726
2739
.
https://doi.org/10.1002/bit.26807
.
Kazadi Mbamba
C.
Flores-Alsina
X.
John Batstone
D.
Tait
S.
2016
Validation of a plant-wide phosphorus modelling approach with minerals precipitation in a full-scale WWTP
.
Water Res.
100
,
169
183
.
https://doi.org/10.1016/j.watres.2016.05.003
.
Kleerebezem
R.
Loosdrecht
M. C. M.
2006
Critical analysis of some concepts proposed in ADM1
.
Water Sci. Technol.
54
,
51
57
.
https://doi.org/10.2166/wst.2006.525
.
Knobel
A. N.
Lewis
A. E.
2002
A mathematical model of a high sulphate wastewater anaerobic treatment system
.
Water Res.
36
,
257
265
.
https://doi.org/10.1016/S0043-1354(01)00209-3
.
Moraes
B. S.
Zaiat
M.
Bonomi
A.
2015
Anaerobic digestion of vinasse from sugarcane ethanol production in Brazil: challenges and perspectives
.
Renew. Sustain. Energy Rev.
44
,
888
903
.
https://doi.org/10.1016/j.rser.2015.01.023
.
Nandy
T.
Shastry
S.
Kaul
S. N.
2002
Wastewater management in a cane molasses distillery involving bioresource recovery
.
J. Environ. Manage.
65
,
25
38
.
https://doi.org/10.1006/jema.2001.0505
.
Ozkan-Yucel
U. G.
Gökçay
C. F.
2010
Application of ADM1 model to a full-scale anaerobic digester under dynamic organic loading conditions
.
Environ. Technol.
31
,
633
640
.
https://doi.org/10.1080/09593331003596528
.
Reichert
P.
1994
Concepts Underlying A Computer Program for the Identification and Simulation of Aquatic Systems
.
EAWAG
Dübendorf, Switzerland
.
Rosen
C.
Jeppsson
U.
2006
Aspects on ADM1 implementation within the BSM2 framework
.
Tech. Rep. TEIE-7224
,
1
37
.
Ruiz
L. M.
Arévalo
J.
Parada
J.
González
D.
Moreno
B.
Pérez
J.
Gómez
M. A.
2011
Respirometric assays of two different MBR (microfiltration and ultrafiltration) to obtain kinetic and stoichiometric parameters
.
Water Sci. Technol.
63
,
2478
2485
.
https://doi.org/10.2166/wst.2011.578
.
Sander
R.
1999
Compilation of Henry's Law Constants for Inorganic and Organic Species of Potential Importance in Environmental Chemistry (Version 3.0)
.
Silva Neto
J. V.
Gallo
W. L. R.
Nour
E. A. A.
2019
Production and use of biogas from vinasse: implications for the energy balance and GHG emissions of sugar cane ethanol in the brazilian context
.
Environ. Prog. Sustainable Energy
Solon
K.
2015
IWA Anaerobic Digestion Model No. 1 Extended with Phosphorus and Sulfur Literature Review
.
Lund University
,
Lund, Sweden
.
Tauseef
S. M.
Abbasi
T.
Abbasi
S. A.
2013
Energy recovery from wastewaters with high-rate anaerobic digesters
.
Renew. Sustain. Energy Rev.
19
,
704
741
.
https://doi.org/10.1016/j.rser.2012.11.056
.
Turovskiy
I. S.
Mathai
P. K.
2006
Wastewater Sludge Processing
.
John Wiley & Sons
,
Hoboken, NJ, USA
.
Van Lier
J. B.
Mahmoud
N.
Zeeman
G.
2008
Anaerobic Wastewater Treatment, in: Biological Wastewater Treatment: Principles, Modelling and Design
.
IWA Publishing
,
London, UK
, pp.
401
442
.
Wilkie
A. C.
Riedesel
K. J.
Owens
J. M.
2000
Stillage characterization and anaerobic treatment of ethanol stillage from conventional and cellulosic feedstocks
.
Biomass Bioenerg.
19
,
63
102
.
https://doi.org/10.1016/S0961-9534(00)00017-9
.
Yasar
A.
Ali
A.
Tabinda
A. B.
Tahir
A.
2015
Waste to energy analysis of shakarganj sugar mills; biogas production from the spent wash for electricity generation
.
Renew. Sustain. Energy Rev.
43
,
126
132
.
https://doi.org/10.1016/j.rser.2014.11.038
.
Zaher
U.
Rodríguez
J.
Franco
A.
Vanrolleghem
P. A.
2003
Application of the IWA ADM1 model to simulate anaerobic digester dynamics using a concise set of practical measurements
. In:
IWA Conference on Environmental Biotechnology
.
Advancement on Water and Wastewater Applications in the Tropics
, p.
12
.