hydraulics of laboratory and full-scale upflow anaerobic sludge blanket (uasb) reactors

5
COMMUNICATION TO THE EDITOR Hydraulics of Laboratory and Full-Scale Upflow Anaerobic Sludge Blanket (UASB) Reactors D.J. Batstone, J.L.A. Hernandez, J.E. Schmidt Environment and Resources, Technical University of Denmark (DTU), Byg 113, Bygningstorvet, Lyngby 2800DK, Denmark; telephone:þ45 4525 1557; fax: þ45 4593 2850; e-mail: [email protected] Received 14 October 2004; accepted 27 January 2005 Published online 23 June 2005 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/bit.20483 Abstract: Laboratory-scale upflow anaerobic sludge blanket (UASB) reactors are often used as test platforms to evaluate full-scale applications. However, for a given volume specific hydraulic loading rate and geometry, the gas and liquid flows increase proportionally with the cube root of volume. In this communication, we de- monstrate that a laboratory-scale reactor had plug-flow hydraulics, while a full-scale reactor had mixed flow hydraulics. The laboratory-scale reactor could be model- ed using an existing biochemical model, and parameters identified, but because of computational speed with plug- flow hydraulics, mixed systems are instead recom- mended for parameter identification studies. Because of the scaling issues identified, operational data should not be directly projected from laboratory-scale results to the full-scale design. ß 2005 Wiley Periodicals, Inc. Keywords: UASB; hydraulic; model; laboratory; full- scale; ADM1 INTRODUCTION The upflow anaerobic sludge blanket (UASB) reactor is a very popular option for anaerobic treatment of high-strength wastewaters with a soluble organic fraction (Lettinga and Hulshoff-Pol, 1991). Anaerobic granule formation is bene- ficial for UASB operation and is an important basis for examination of anaerobic microbial ecology and molecular biology (Sekiguchi et al., 1999). Principal of operation of UASB reactors is distribution of feed at the base of the reactor, percolation through a naturally forming granular microbial sludge blanket, with integrated solid–liquid–gas separation at the surface of the reactor. The sludge blanket is partially fluidized and mixed by gas flow, which also mixes the liquid volume. Small-scale laboratory reactors (0.2 – 5 L) are often used as a basis to evaluate behavior of full-scale reactors. There are three scaling issues involved with this: (a) For an equivalent diameter/height ratio (D:H), wastewater concentration, and volume-specific loading, superficial gas velocity, and super- ficial liquid velocity are proportional to the inverse cube of volume (q gas / q liq / V 1/3 ). This means that the velocities change less with volume as volume increases. (b) There are laminar flow and wall effects in small reactors. (c) The granule/reactor size is larger in laboratory reactors, which could cause scaling effects. For laboratory to full-scale scaling, issue (a) is the most important as hydraulics may change from plug-flow to stirred tank reactor (STR), as the reactor volume increases. A plug-flow reactor will more effectively remove substrate than a STR (with Monod, or first-order biological kinetics), as there is a high driving force for substrate removal at the start of the reactor. This is of critical importance when making operational interpretations via model-based analysis, and often, STR hydraulics are used (Costello et al., 1991). Hydraulics in smaller UASB reactors have been modeled as plug-flow (Singal et al., 1998), and approximations made using multiple mixed-compartment models (Bolle et al., 1986), but these have not included de- tailed conversion equations. Kalyuzhnyi et al. (1998) pre- sented an integrated plug-flow model including biochemical kinetics, but the focus of their paper was reactor performance and population balancing rather than on the specific influence of plug-flow hydraulics. In this communication, we compare the hydraulics of laboratory- and full-scale reactors, using mathematical models to evaluate the optimal flow configurations. Addi- tionally, the impact of the optimal hydraulic model on overall reactor kinetics and parameter identification is assessed by adding biochemical kinetics. MATERIALS AND METHODS Data Acquisition The laboratory-scale reactor was a cylindrical reactor with 0.5 L total volume, height of 230 mm, and inside diameter of ß 2005 Wiley Periodicals, Inc. Correspondence to: Damien J. Batstone

Upload: dj-batstone

Post on 06-Jun-2016

218 views

Category:

Documents


6 download

TRANSCRIPT

Page 1: Hydraulics of laboratory and full-scale upflow anaerobic sludge blanket (UASB) reactors

COMMUNICATION TO THE EDITOR

Hydraulics of Laboratory and Full-ScaleUpflow Anaerobic Sludge Blanket(UASB) Reactors

D.J. Batstone, J.L.A. Hernandez, J.E. Schmidt

Environment and Resources, Technical University of Denmark (DTU), Byg 113,Bygningstorvet, Lyngby 2800DK, Denmark; telephone:þ45 4525 1557;fax: þ45 4593 2850; e-mail: [email protected]

Received 14 October 2004; accepted 27 January 2005

Published online 23 June 2005 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/bit.20483

Abstract: Laboratory-scale upflow anaerobic sludgeblanket (UASB) reactors are often used as test platformsto evaluate full-scale applications. However, for a givenvolume specific hydraulic loading rate and geometry,the gas and liquid flows increase proportionally with thecube root of volume. In this communication, we de-monstrate that a laboratory-scale reactor had plug-flowhydraulics, while a full-scale reactor had mixed flowhydraulics. The laboratory-scale reactor could be model-ed using an existing biochemical model, and parametersidentified, but because of computational speedwith plug-flow hydraulics, mixed systems are instead recom-mended for parameter identification studies. Because ofthe scaling issues identified, operational data should notbe directly projected from laboratory-scale results to thefull-scale design. � 2005 Wiley Periodicals, Inc.

Keywords: UASB; hydraulic; model; laboratory; full-scale; ADM1

INTRODUCTION

The upflow anaerobic sludge blanket (UASB) reactor is a

very popular option for anaerobic treatment of high-strength

wastewaters with a soluble organic fraction (Lettinga and

Hulshoff-Pol, 1991). Anaerobic granule formation is bene-

ficial for UASB operation and is an important basis for

examination of anaerobic microbial ecology and molecular

biology (Sekiguchi et al., 1999). Principal of operation of

UASB reactors is distribution of feed at the base of the

reactor, percolation through a naturally forming granular

microbial sludge blanket, with integrated solid–liquid–gas

separation at the surface of the reactor. The sludge blanket

is partially fluidized andmixed by gas flow, which alsomixes

the liquid volume.

Small-scale laboratory reactors (0.2–5 L) are often used as

a basis to evaluate behavior of full-scale reactors. There are

three scaling issues involved with this: (a) For an equivalent

diameter/height ratio (D:H), wastewater concentration, and

volume-specific loading, superficial gas velocity, and super-

ficial liquid velocity are proportional to the inverse cube of

volume (qgas/ qliq/V1/3). This means that the velocities

change less with volume as volume increases. (b) There are

laminar flow and wall effects in small reactors. (c) The

granule/reactor size is larger in laboratory reactors, which

could cause scaling effects. For laboratory to full-scale

scaling, issue (a) is the most important as hydraulics may

change from plug-flow to stirred tank reactor (STR), as the

reactor volume increases. A plug-flow reactor will more

effectively remove substrate than a STR (with Monod, or

first-order biological kinetics), as there is a high driving force

for substrate removal at the start of the reactor. This is of

critical importance when making operational interpretations

viamodel-based analysis, and often, STR hydraulics are used

(Costello et al., 1991). Hydraulics in smaller UASB reactors

have been modeled as plug-flow (Singal et al., 1998), and

approximations made using multiple mixed-compartment

models (Bolle et al., 1986), but these have not included de-

tailed conversion equations. Kalyuzhnyi et al. (1998) pre-

sented an integrated plug-flow model including biochemical

kinetics, but the focus of their paper was reactor performance

and population balancing rather than on the specific influence

of plug-flow hydraulics.

In this communication, we compare the hydraulics of

laboratory- and full-scale reactors, using mathematical

models to evaluate the optimal flow configurations. Addi-

tionally, the impact of the optimal hydraulic model on overall

reactor kinetics and parameter identification is assessed by

adding biochemical kinetics.

MATERIALS AND METHODS

Data Acquisition

The laboratory-scale reactor was a cylindrical reactor with

0.5 L total volume, height of 230 mm, and inside diameter of

�2005 Wiley Periodicals, Inc.

Correspondence to: Damien J. Batstone

Page 2: Hydraulics of laboratory and full-scale upflow anaerobic sludge blanket (UASB) reactors

48 mm. Feed to the reactor was basal anaerobic (BA) media

as previously described (Angelidaki et al., 1990), with 6 g

L�1 glucose, 2.5 g L�1 acetate, and 1.5 g L�1 propionate

(all amounts as COD). Inoculum was from a starch-fed

industrial UASB. Operation was over 6 months, with steady

state (as observed in effluent organic acid levels) achieved

after the first month. Loading set-point was 5 kg COD m�3

d�1 for the first 5 months, and 10 kg COD m�3 d�1 for the

last month. Biochemical kinetics were evaluated by 0.8- and

1.6-g COD acetate pulses in the low- and high-loading

periods, respectively (acetate as sodium acetate in 10 mL of

MilliQwater). The full-scale reactor was a cylindrical UASB

with 1,200 m3 design volume (1,400 m3 based on geometry),

fed with slaughterhousewastewater (average 5 kg CODm�3,

50% soluble) at an average loading rate of 2.7 kg COD m�3

d�1. Average hydraulic retention time (HRT) was 24 h. It had

operated for 1 year prior to the tracer test described, and the

average gas flow was approximately the same as the liquid

flow. Organic acids from both reactors were analyzed by gas

chromatography using a flame ionization detector. Hydraulic

analysis in the laboratory reactor was by injection of 8.7 g

NaCl (0.15 mol) in 30 mL of a 5 M NaCl solution. Con-

ductivity was measured on-line. The sensor was inserted

through the top of the reactor, such that it made measure-

ments in the top 10mm of the reactor fluid. Conductivity was

correlated to saline concentration by linear calibration, and

the conductivity was corrected for the rise in organic acids

concentration. Three pulses were made, with average HRTs

of 1.63, 0.89, and 0.44 d. Respective loading rates were

6.1, 11.2, and 22.7 kg COD m�3 d�1. The loading rate was

doubled again after this, and the reactor suffered complete

acid overload. Hydraulic analysis of the full-scale reactor

was by injection of 8 g of rhodamine WT and effluent

sampling (in a glass container) every 2 h (3� 40 min

composite) for 72 h. Samples were filtered and analyzed

using a fluorescence spectrophotometer with an excitation

wavelength of 558 nm, and emission wavelength of 581 nm.

Full data, and further analysis for the full-scale assessment is

given in (Batstone, 2000).

Model Description

The plug-flow (advective–diffusive) model in Aquasim 2.1d

(Reichert, 1994) was used as a base. This includes axial

diffusion, and linear estimates between calculated grid points

along the reactor. Implementation was similar to that of

Singal et al. (1998), except the internal flow was bidirec-

tional, i.e., there could either be a recycle source from 50%

of the reactor link or a bypass directly from the influent

to 50% of the reactor length. This combines an internal

recycle (Bolle et al., 1986) with an internal bypass (Singal

et al., 1998) and allows simultaneous testing for both. Multi-

CSTR hydraulic models were also evaluated. The bio-

chemical model used was the ADM1 (Batstone et al., 2002),

as implemented in (Batstone et al., 2003), except with plug-

flow hydraulics and retained solids as recommended in

(Batstone et al., 2002). Inputs were fully defined from the

synthetic wastewater. Initial conditions were steady state at

the relevant load. The parameter estimation, and parameter

uncertainty evaluation method of (Batstone et al., 2003) was

used, with a 95% confidence level for significance testing

and parameter uncertainty analysis. Where the uncertainty

estimates of parameters are given, these are 95% confidence,

linear, and uncorrelated estimates. Source files for all models

are available from the corresponding author on request.

RESULTS

Hydraulic Tests

The plug-flow model with a high recycle was most effective

in simulating the laboratory-scale UASB (Fig. 1). The best

multiple-STR hydraulic model was with 8 tanks, and was

not as effective as the plug-flow hydraulic model. Para-

meter optimization indicated that the bypass and recycle

were both zero (i.e., crossed the zero vector). Therefore,

the model could be fitted using only effective diffusivity, and

area. The effective area was 13.5� 0.03 cm2, compared

with an actual area of 17 cm2, while effective diffusivity was

75� 1 cm2 d�1 compared to an actual diffusivity for Naþ of

1.8� 10�4 cm2 d�1. Correlation between these parameters

Figure 1. Measured data (^) and model (lines) responses to pulses in

(a) laboratory and (b) full-scale operations. Pulses were 8.7 g NaCl and 8 grhodamine WT, respectively.

388 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 91, NO. 3, AUGUST 5, 2005

Page 3: Hydraulics of laboratory and full-scale upflow anaerobic sludge blanket (UASB) reactors

was low. Model performance degraded, and optimal para-

meters changed significantly below 12 grid points.

The results indicate loss of effective area due to the

granular sludge, as well as high dispersion within the reactor.

The later pulse at double the flow and load of the first pulse

indicated no recycle or bypass, no matter where the source/

sink point, but confirmed the standard plug-flow model.

Effective diffusivity for the later pulsewas 120 cm2 d�1, with

an effective area of 15 cm2. Because the results indicated no

significant bypass or recycle, this element was removed for

biochemical simulation.

For the full-scale system, a two-STRmodelwas best (pulse

data not shown), with volumes of 114� 33 m3 and 1,422 m3

(total 1,536 m3). Because the first tank is less than 10% of

the second, this indicates, effectively, a mixed reactor. When

a three-STR model was tested, one of the volumes was

statistically zero (0.0017� 110 m3).

Impact of Gas Flow

In the laboratory-scale reactor, three NaCl pulses were

applied, at averages HRTs of 1.63, 0.89, and 0.44 day, with a

corresponding increase in gas flow.Response to the first pulse

is shown in Figure 1. A dimensionless degree of mixing was

defined, which is the hydraulic retention time over the time to

concentration peak:

DM ¼ HRT=tpeak

A higher DM indicates more STR-type hydraulics. This was

used instead of estimated apparent diffusion in a plug-flow

model, as it better represents transition between plug-flow

and CSTR in systems that display the characteristics of both.

It also lumps any correlation influences between apparent

area and diffusivity. Results of the three tests indicated a

downward trend in the degree of mixing with increased gas

flow (Fig. 2). This is not a significant correlation at the 95%

confidence because of the small number of samples, but it

does indicate a trend.

A hydraulic test was also done on a full-scale reactor with

the same design as that assessed above, except with total

volume of 300 m3 with no gas flow. NaOHwas used as tracer

and showed plug-flow hydraulics with a DM of 1 (Batstone,

2000).

Biochemical Response and Model

The laboratory reactor showed a slight decrease in effluent

acetate to a small pulse (0.8 g COD acetate), at anHRTof 2 d.

As found during modeling, this was probably because the

increased acetate at the start of the reactor allowed faster

kinetics in that region, which meant that final effluent levels

decreased in response to the pulse (i.e., an influent increase

caused an effluent decrease). Because of the low response,

this pulse could not be fitted using the model. A higher

pulse (1.6 g COD acetate) at an HRT of 1 d showed a mea-

surable response, and this could be modeled (Fig. 3).

Initial parameter estimation was attempted with four

parameters: uptake/growth and diffusivity/area (km,ac, KS,ac,A, D). This was over-correlated, and diffusion was removed.

The number of grid points was reduced to 9, to simulate

diffusion (Reichert, 1994). After this, the maximum uptake

rate (km,ac), the half-saturation concentration (KS), and area

(A) could be simultaneously identified. The parameter

uncertainty region for these three parameters is shown in

Figure 4. The apparent uptake rate (km) is much lower than

found in other CSTR systems using the ADM1 (Batstone

et al., 2002, 2003); �8–10 kg CODS kg COD X�1 d�1 in

these), and half-saturation concentration higher (KS; �0.1–

0.5 kg CODS m�3 in above). This indicates slower specific

kinetics. Because of the plug-flow hydraulics, however,

overall performance is far better.

Figure 2. Apparent degree of mixing against average superficial gas

velocity in the laboratory reactor, with the trend line fitted by linear

regression. Hydraulic retention times for the three points (left to right) were

1.63, 0.89, and 0.44 d.

Figure 3. Laboratory-scale reactor (^) andmodel (—) acetate response to

a 1.6 g COD acetate pulse with an HRT of 1 day.

BATSTONE ET AL.: HYDRAULICS OF LABORATORY AND FULL-SCALE UASB REACTORS 389

Page 4: Hydraulics of laboratory and full-scale upflow anaerobic sludge blanket (UASB) reactors

DISCUSSION

Hydraulics

This study showed that the laboratory-scale UASB reactor

was almost completely plug-flow, while the full-scale UASB

reactor with gas flow was almost completely CSTR. The

results in the full-scale system are emphasized, since the

loading rate in that systemwas relatively low (2 kgCODm�3

d�1, compared to normal systems at 10 kg COD m�3 d�1).

This was a comparison between two single systems, but other

studies have confirmed the plug-flow behavior in laboratory

reactors, as observed here (Singal et al., 1998). Bolle et al.

(1986) evaluated a full-scale reactor, similar in size to that

found here, and recommended a multi-mixed tank model,

with short-circuiting past the tanks. On the basis of our data,

we support Bolle’s concept for representation of solids and

hydraulics in a kinetic model for full-scale systems. Due to

the differences in hydraulic behavior, it is very perilous to

make any performance-based projections from laboratory-

scale systems to full-scale UASB systems. This is because

if Monod, or first-order, kinetics dominate, plug-flow reac-

tors can be much more effective, with completely different

responses to disturbances. There were also indications in the

laboratory-scale reactor that gas flow had an inverse impact

on the degree of mixing (Fig. 2). This is the opposite of the

expected response, and we believe it is due to the follow-

ing reasons: in the reactor, we observed that gases did not

form bubbles except in static regions and that most of the

stripping occurred at the surface. Microbubbles did form,

and this could cause channeling, with the effect of increas-

ing the retention time. At some unknown geometry or gas

flow, macrobubbles would presumably form, with increased

mixing effects.

We also observed different results with respect to recycle/

bypass compared to others (Bolle et al., 1986; Singal et al.,

1998), who found that recycle and bypass, respectively, were

needed to adequately model their systems. Singal et al.

(1998) used a 2.4-L UASB reactor with the same cross-

sectional area as the reactor used here, and the longer reactor

and higher upflow velocity may have caused some bypass.

Adding a recycle flow to our model gave marginal and

inconsistent improvement in performance, while bypass was

never recommended by the parameter estimation procedure.

Laboratory-Scale Reactors as a Basis for KineticParameter Estimation

We have shown that it is possible to simultaneously estimate

area (A) as well as the governing biochemical kinetic para-

meters (km, KS). However, the biochemical parameters are

correlated with active retained sludge, which is very difficult

to estimate, and there are correlation effects associated with

diffusion. Therefore, althoughwe obtained parameters with a

level of certainty, actual parameters are probably higher due

to these correlation effects. Additionally, solving a plug-flow

model is time consuming compared with a complete mixed

reactor, especially if long periods of reactor operation are

used for model convergence (and to remove the effects of

initial conditions). For these reasons, a full-scale UASB,

anaerobic sequencing batch reactor, or complete mixed re-

actor is a better basis for parameter estimation.

Appropriate Applications for Laboratory-ScaleUASB Reactors

Performance comparison with full-scale systems is not an

appropriate use for laboratory-scale UASB reactors. Labora-

tory-scale systems are very effective for growing granular

sludge for studying granulation itself. However, the con-

centration will profile down the reactor, as shown from the

model results in Figure 5. Because of variation of bulk

substrate concentrations with height, microbial ecology can

also vary. We also assessed this using fluorescence in-situ

hybridization (Hugenholtz et al., 2001) and foundmuchmore

developed acidogenic outer layers in granules sampled from

the base compared to granules sampled from the top of the

Figure 5. Simulated acetate (—), propionate (- - -), and glucose (— —)

profiles at steady state with 10 kg COD m�3 d�1 loading, and a 1-d HRT.

Figure 4. Parameter uncertainty region using the data shown in Figure 3.

Area is on the z axis. Linear, uncorrelated uncertainty estimates are also

shown for reference.

390 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 91, NO. 3, AUGUST 5, 2005

Page 5: Hydraulics of laboratory and full-scale upflow anaerobic sludge blanket (UASB) reactors

sludge bed (results not shown). Another application involves

study of anaerobic digestion under plug-flow conditions.

There are also other plug-flow systems, in which anaerobic

digestion is important, including anoxic and anaerobic

clarifiers

The laboratory-scale work and paper preparation were funded by the

EU 5th Framework Program under the project BIOWASTE (contract

no. QLK5-CT-2002-01138). Mark Newland from ESI Ltd and Jurg

Keller from the University of Queensland are thanked for their

assistance and advice with the full-scale experimental and modeling

work.

References

Angelidaki I, PetersenSP,AhringBK. 1990.Effects of lipids on thermophilic

anaerobic digestion and reduction of lipid inhibition upon addition of

bentonite. Appl Microbio Biotechnol 33:469–472.

Batstone DJ. 2000. High rate anaerobic treatment of complex wastewater.

Ph.D. thesis. St Lucia, Queensland: University of Queensland.

Batstone DJ, Keller J, Angelidaki I, Kalyuzhnyi SV, Pavlostathis SG, Rozzi

A, Sanders WTM, Siegrist H, Vavilin VA. 2002. Anaerobic digestion

model no. 1 (ADM1). IWATask Group for Mathematical Modelling of

Anaerobic Digestion Processes. London: IWA Publishing.

Batstone DJ, Pind PF, Angelidaki I. 2003. Kinetics of thermophilic,

anaerobic oxidation of straight and branched chain butyrate and

valerate. Biotechnol Bioeng 84(2):195–204.

Bolle WL, van Breugel J, van Eybergen G, Kossen N, van Gils W. 1986.

An integral dynamic model for the UASB reactor. Biotechnol Bioeng

28:1621–1636.

Costello DJ, Greenfield PF, Lee PL. 1991. Dynamic modelling of a single-

stage high-rate anaerobic reactor—II. Model verification. Water Res

25(7):859–871.

Hugenholtz P, Tyson GWB, Blackall LL. 2001. Design and evaluation of

16S rRNA-targeted oligonucleotide probes for fluorescence in situ

hybridization. In: Lieberman BA, editor. Methods in molecular biology.

Totowa, NJ: Humana Press. p 29–42.

Kalyuzhnyi S, Fedorovich V, Lens P, Hulshoff-Pol LW, Lettinga G. 1998.

Mathematicalmodelling as a tool to study population dynamics between

sulfate reducing and methanogenic bacteria. Biodegradation 9:187–

199.

Lettinga G, Hulshoff-Pol L. 1991. UASB-process design for various types of

wastewaters. Water Sci. Technol 24(8):87–107.

Reichert P. 1994. AQUASIM—a tool for simulation and data analysis of

aquatic systems. Water sci Technol 30:21–30.

Sekiguchi Y, Kamagata Y, Nakamura K, Ohashi A, Harada H. 1999.

Fluorescence in situ hybridization using 16S rRNA-targeted oligonu-

cleotides reveals localization of methanogens and selected uncultured

bacteria in mesophilic and thermophilic sludge granules. Appl Environ

Microbiol 65:1280–1288.

Singal A, Gomes J, Praveen VV, Ramachandran KB. 1998. Axial dispersion

model for upflow anaerobic sludge blanket reactors. Biotechnol Prog

14:645–648.

BATSTONE ET AL.: HYDRAULICS OF LABORATORY AND FULL-SCALE UASB REACTORS 391