Temperature-Based Control of an Anaerobic ... - ACS Publications

Succ., Centre-Ville, Montréal, QC, Canada H3C 3A7. This study presents a temperature-based control strategy for the stabilization of an anaerobic rea...
1 downloads 0 Views 273KB Size
Environ. Sci. Technol. 2007, 41, 978-983

Temperature-Based Control of an Anaerobic Reactor Using a Multi-Model Observer-Based Estimator E M M A N U E L M O R E L , †,‡ B O R I S T A R T A K O V S K Y , †,‡ MICHEL PERRIER,‡ AND S E R G E R . G U I O T * ,† Biotechnology Research Institute, NRC, 6100 Royalmount Avenue, Montre´al, QC, Canada H4P 2R2, and De´partement de Ge´nie Chimique, EÄ cole Polytechnique de Montre´al, C.P. 6079 Succ., Centre-Ville, Montre´al, QC, Canada H3C 3A7

This study presents a temperature-based control strategy for the stabilization of an anaerobic reactor during organic overloads. To prove feasibility of the proposed approach the rate of methane production was followed in batch activity tests and reactor runs during mesophilicthermophilic transitions. Within the first 0.25-6 h of temperature augmentation, an increase in the rate of methane production was observed with higher rates measured under thermophilic (above 40 °C) conditions. However, 24 h after startup both in batch tests and reactor runs, the rate of methane production under thermophilic conditions was inferior to that under optimal mesophilic conditions (35 °C). Following these results, a control strategy based on short-term augmentation of the reactor temperature was proposed and tested in a 10 L UASB reactor. The control strategy employed a multi-model observer-based estimator to stabilize the effluent COD concentration during organic overloads. The temperature-based control resulted in an increased methanization rate and improved reactor stability overall.

1. Introduction Anaerobic treatment of high strength wastewaters is economically attractive for wastewater management because it combines high biotransformation rates with low biomass production and reduced energy requirements. Anaerobic biotransformation of degradable organic matter is a complex multi-stage process, which involves several microbial trophic groups (1). Anaerobic degradation of high strength industrial wastewaters is more efficient than aerobic degradation, but anaerobic reactors are more difficult to operate because of their high sensitivity to operating conditions. Reactor failure can be caused by a number of factors such as abrupt changes of organic loading rates, pH variations, and presence of toxicants. Anaerobic biodegradation of organic matter can be carried out over a wide range of temperatures including psychrophilic (below 20 °C), mesophilic (25-45 °C), and thermophilic (above 55 °C) conditions (2). The Arrhenius equation is often * Corresponding author phone: (514)496-6181; fax: (514)496-6265; e-mail: [email protected]. † Biotechnology Research Institute, NRC. ‡ E Ä cole Polytechnique de Montre´al. 978

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 3, 2007

used to describe the influence of temperature on microbial growth and biodegradation in anaerobic digestion (3-5). The transition from mesophilic to thermophilic conditions generally requires a prolonged adaptation period, and improper reactor operation during this period was found to result in reactor failure (6, 7). Nevertheless, an improved methane production has been reported after a short-term temperature increase from 25 to 45 °C (8). A recovery of methanogenic activity was observed after a short-term temperature increase above 45 °C followed by a return to mesophilic conditions (9). Furthermore, sludge exchange in a two-phase mesophilic-thermophilic digestion process was shown to improve the overall reduction of volatile solids (10). This study presents an experimental demonstration of anaerobic reactor stabilization using short-term temperature increases to avoid reactor failure due to organic overload. Because overexposure of the mesophilic sludge to thermophilic conditions can be harmful to the mesophilic methanogens, timely temperature adjustments were provided by a multi-model adaptive controller whose design was based on the multi-model approach (11, 12). This approach to process modeling is distinctly different from using comprehensive process models as it combines the simplicity of individual submodels with the flexibility of a knowledgebased system. Furthermore, the adaptive controller used multiwavelength fluorometry for on-line measurements of chemical oxygen demand (COD) and volatile fatty acid (VFA) concentrations in the reactor effluent (13).

2. Materials and Methods 2.1. Media. An anaerobic 0.05 M phosphate buffer, pH 7.5, contained (per L) K2HPO4 (2000 mg), Na2HPO4‚7H2O (2550 mg), NaH2PO4‚H2O (550 mg), 0.1% (w/v) resazurin (10 mL), and 1.25% (w/v) cysteine-sulfide solution. Stock solution of synthetic wastewater had a COD content of 315 g COD L-1 and contained (in g L-1): sucrose (99), butyric acid (48), yeast extract (60), ethanol (95%, 35), KH2PO4 (3), K2HPO4 (3.5), and NH4HCO3 (34). A stock solution of trace metals containing (in g L-1) AlK(SO4)‚12H2O( 0.0006); H3BO3 (0.001); Ca(NO3)2‚ 4H2O (0.5351); Co(NO3)2‚6H2O (0.0075); Cu(SO4) (0.0003); Fe(SO4)‚7H2O (0.0546); MgSO4 (0.1973); Mn(SO4)‚H2O (0.0151); Na2(MoO4)‚2H2O (0.0023); NiSO4‚6H2O (0.0007); Na2SeO4 (0.0013); and ZnSO4‚7H2O (0.0035) was used for feeding the anaerobic reactor. A bicarbonate buffer was composed of 1.36 g L-1 of NaHCO3 and 1.74 g L-1 of KHCO3. 2.2. Analytical Methods. VFA in the effluent were determined using a gas chromatograph (Sigma 2000, PerkinElmer, Norwalk, CT) equipped with a 91 cm × 4 mm i.d. glass column packed with 60/80 Carbopack C/0.3% Carbopack 20 NH3PO4 (Supelco, Mississauga, ON). The column temperature was maintained at 120 °C isothermally, and the injector and detector temperature were permanently at 200 °C. The carrier gas was nitrogen. COD and VSS were determined according to Standard Methods (14). For both COD and VFA measurements the standard deviation attributed to the analytical procedures did not exceed 5%. However, errors associated with sampling procedures led to an overall standard deviation of up to 15%, in particular for low levels of COD and VFA. Standard deviation of VSS measurements was estimated at 20-25% because of sample heterogeneity. Biogas composition (H2, CH4, N2, and CO2) was measured by gas chromatography (Sigma 2000, Perkin-Elmer, Norwalk, CT) equipped with a thermal conductive detector and two tandem columns, measuring 61 cm × 3.2 mm i.d. and 1.83 10.1021/es0618043 CCC: $37.00

 2007 American Chemical Society Published on Web 01/03/2007

m × 3.2 mm i.d., respectively, and packed with 60/80 Chromosorb 102, and 60/80 molecular sieve 5A (Supelco, Mississauga, ON), respectively. The carrier gas was argon. 2.3. Activity Tests. Batch activity tests were carried out to evaluate the methane-producing potential of the anaerobic sludge at different temperatures. The tests were carried out in 60 mL serum bottles maintained under anaerobic conditions. All tests were carried out in triplicates to ensure reproducibility of the results. At startup, the bottles were inoculated with anaerobic granular biomass stored at 4 °C and diluted in phosphate buffer to a concentration of 5 g VSS L-1. The synthetic wastewater was added to obtain an initial COD concentration of 4 g L-1. The bottles were flushed with N2/CO2 (80%/20%) and then incubated in a rotary shaker (New Brunswick Scientific Co., Edison, NJ) for a period of at least 24 h at 100 rpm and a constant temperature. The volume of biogas accumulated in the bottle headspace was measured periodically using a burette (gas displacement method). Concentrations of methane and hydrogen in the headspace were measured by gas chromatography immediately following the volume measurements. Methane and hydrogen concentrations were converted into COD equivalents and biogas production rates were calculated using the time interval between the measurements. 2.4. UASB Reactor. A 10 L upflow anaerobic sludge bed (UASB) reactor with an external recirculation line was used for the experiments. The setup is described in more detail in Zeng et al. (15). The reactor was equipped with a water jacket and a water heating system for temperature control. The synthetic wastewater and trace metals were added into the bicarbonate buffer stream at a feeding rate of 0.2 L d-1 each. The total influent flow rate was 20 L d-1, which corresponded to an influent COD concentration of 3 g L-1 at an organic loading rate (OLR) of 6 g COD L-1 R (liter of reactor volume) d-1. The feeding rate of the synthetic water was doubled to 0.4 L d-1 during organic overloads. The reactor was inoculated with an anaerobic granular sludge from a wastewater plant (A. Lassonde Inc., Rougemont, QC) with an average volatile suspended solids (VSS) content of 50 g L-1. Biogas production was measured on-line using an electronic bubble counter and the values were adjusted with respect to the reactor temperature. Methane content of biogas was also measured on-line using a methane analyzer (Nova Analytical Systems, Hamilton, ON). Reactor pH was measured by a pH meter (Cole-Parmer Instrument, Vernon Hills, IL) with the probe inserted in the external recirculation line. TH series temperature sensors (Roctest, Saint-Lambert, QC) were used for on-line measurements of temperature in the reactor, water jacket, and air. A PC equipped with a PC-1200 acquisition board (National Instruments, Austin, TX) was used for data acquisition and pump control. The software for reactor monitoring and control was developed in house using Visual Basic v6 (Microsoft Corporation, Redmond, WA) and MATLAB (MathWorks Inc., Natick, MA). 2.5. Multiwavelength Fluorometry. On-line fluorescence measurements for COD and VFA estimations were performed with a multiwavelength fluorometer (MWF). The probe was placed in the external recirculation line of the reactor. The MWF consisted of a CCD array spectrometer (Ocean Optics Inc., Dunedin, FL), light sources, and a fiber optic probe. Light sources were PX-2 Xenon and LS-450 equipped with a 380 nm UV-LED (Ocean Optics Inc., Dunedin, FL). Fluorescence spectra were acquired in 4 min intervals with background acquisition prior to each fluorescence measurement. The spectrometer integration time was 2000 and 5000 ms for the PX-2 and LS-450 light sources, respectively. To reduce noise, 5 spectra were acquired and averaged for each measurement.

TABLE 1. Kinetic Dependencies of the Multi-Modela process state state variable (s) COD (a) VFA (Qm) CH4 flow rate

organic methanogenic overload -rmax,ss -rmax,aa rmax,ms

-rmax,s rmax,as rmax,m

acidogenic -rmax,ss rmax ,as rmax,m KI/(KI + a)

a Notations: r max,s, rmax,a, and rmax,m are the maximum biotransformation rates for COD, VFA, and methane flow rate, respectively. KI is the VFA inhibition constant.

Calibration of COD and VFA predictions was carried out using a linear regression model between selected areas of the fluorescence spectra and corresponding analytical measurements. Two areas per each emission spectrum were selected, i.e., a total of four areas were used for COD and VFA measurements. In addition, the effect of pH variations on the fluorescence intensity was considered in the regression models by including the pH value as an additional regression variable.

3. Results and Discussion 3.1. Design of a Multi-Model Adaptive Controller. The adaptive controller implemented in this study was based on the multi-model of the anaerobic digestion process developed previously (12, 16). The multi-model was modified to include the effect of temperature on the biotransformation rates and then transformed into an observer-based estimator. The input-output linearization method was then used to design a multi-model adaptive controller. Below is a brief description of the design procedure, while more details can be found elsewhere (11, 17). The multi-model consisted of three submodels describing normal (methanogenic), organic overload, and acidogenic process states. Multi-model outputs were defined by the weighted sums of the submodel outputs. The weights were calculated by a knowledge-based system, which used online measurements of biogas composition and reactor pH for process diagnosis (11). The rates of biotransformation (r) of the multi-model were defined using kinetic dependencies given in Table 1. COD degradation was described by zero- and first-order dependencies, and the influence of VFAs on methane production was described by a nonlinear dependence. The vector of weights (βi) for computation of multi-model outputs was calculated by a knowledge-based system. The latter was tuned by defining spheres in a 3D space (pH, biogas flow rate, methane content) with respect to the expert’s knowledge of the anaerobic digestion process. Each sphere represented a typical process state (methanogenic, organic overload, or acidogenic). The influence of temperature on the microbial activity was modeled using a modified Arrhenius equation, which linked the biotransformation rate with the temperature: f(T) ) eθ(T-T0), where θ is the temperature coefficient (θ ) 0.025) and T0 is the relative temperature (T0 ) 35 °C) (4, 5). With the assumptions of ideal mixing and instantaneous methane transfer from liquid to gas phase, liquid-phase material balances for COD (j ) 1), total VFA (j ) 2), and the output methane flow rates for each submodel (Qi) are defined as follows:

Qi dζi,j F in ) ri,j(ζi,j, t) f(T) + (ζi,j with i ) - ζi,j) dt V V 1, . . . , n and j ) 1, . . . , 2 (1) Qi ) ri,j (ζi,j,t) f(T)V with i ) 1, . . . , n and j ) 3 (2) VOL. 41, NO. 3, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

979

where ζ is the vector of state variables (species concentration, g L-1), ζin is the vector of influent species concentration (g L-1), r is the vector of biotransformation rates (g L-1 d-1), f(T) is the temperature function, F is the volumetric input flow rate (L d-1), V is the reactor volume (L), Qi is the vector of gaseous output flow rate (g d-1), and n is the total number of submodels, n ) 3. The vector of multi-model outputs (y) is defined as n

yj(t, T) )

∑βζ

i i,j

i)1

n

with

∑ β ) 1, β ∈ [0,1], i

i

j ) 1, . . . , m (3)

FIGURE 1. Biogas production rates observed in batch tests 15 min, 2 h, and 24 h after inoculation. The tests were carried out at temperatures of 25, 35, 45, and 55 °C.

i)1

n

where m is the total number of state variables, m ) 3, and βi is the weight of the j-th submodel. The temperature model of the reactor used the assumptions of ideal mixing and isothermal biochemical reactions. The model was simplified by neglecting the contribution of the influent stream :

dT ) K1(Th - T) + K2(T - Tair) dt with K1 ) U1A1 and K2 ) -U2A2 (4) where T, Tair, and Th are the reactor, the air, and the water jacket temperatures (°C), K1 and K2 are the overall heat transfer coefficients (W °C-1), U1, U2 are the heat transfer coefficients (W m-2 °C-1) and A1, A2 are the heat transfer areas (m2). To obtain a multi-model observer-based estimator the multi-model submodels were converted into corresponding sub-estimators (11) and tuned using a systematic tuning approach presented by Perrier et al. (18). This approach reduced the number of design parameters to a single parameter (wi) per state variable. The multi-model adaptive controller was designed by linearizing the multi-model observer. The input/output linearization method (19) was applied with the tracking error (ζ* - ζ) defined using the following first-order reference model:

d * (ζ - ζ) + λ(ζ* - ζ) ) 0 and λ > 0 dt

(5)

The set point ζ* is defined as a constant, dξ*/dt ) 0, and the reference model can be simplified as follows:

dξ ) λ(ζ* - ζ) dt

(6)

For each sub-controller (i ) 1...3) the COD concentration (ξi,1) and the local reactor temperature (Ti) were defined as the controller set point and the manipulated variable, respectively, and the following adaptive controller was obtained by combining eqs 1 and 6 with Qi ) 0:

Ti ) T0 +

[

]

λi(ζ/i,j - ζi,j) F (ζi,j - ζ/i,j) 1 ln , + θ V ri,j(ζi,j, t) ri,j(ζi,j,t) i ) 1, . . . , n, j ) 1, . . . , m (7)

where ri,j is the estimated reaction rate, and λi is the tuning parameter of the i-th controller corresponding to i-th submodel. The reactor temperature (T*) was calculated as a weighted average of the temperatures calculated for each submodel. The weight vector (βi) was calculated by the knowledge-based system: 980

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 3, 2007

T* )

∑ βT

j j

(8)

j)1

The adaptive temperature controller was obtained using the linearization method, where the previously calculated reactor temperature (T) was used as the set point (T*) for the temperature controller and the water jacket temperature (Th) was the manipulated variable:

Th ) T +

1 [λ (T* - T) + K2(Tair - T)] K1 T

(9)

where λT is the tuning parameter of the controller. 3.2. Effect of Temperature on Biogas Production. The effect of temperature on biogas production was studied both in batch activity tests and in UASB reactor runs. Initial activity tests were carried out using mesophilic anaerobic sludge stored at a temperature of 4 °C. After inoculation, test bottles were incubated at temperatures of 25, 35, 45, and 55 °C. The net rate of biogas production was expressed on a COD basis to include both methane and hydrogen, as small quantities of hydrogen were detected in the thermophilic bottles. The results of these tests are summarized in Figure 1. When biogas production was measured 15 min after the test startup, no biogas production was observed at a temperature of 25 °C and the production rates were low at 35 and 45 °C. At the same time, biogas production reached 0.6 g COD (g VSS d)-1 at 55 °C. However, 2 h after the test startup the highest production rate was observed at a temperature of 35 °C. At t ) 24 h the production rates were similar in all bottles although the rate was slightly higher in the 35 °C bottles. A similar response to the temperature increase was observed in a 10 L UASB reactor. The reactor was operated at a temperature of 25 °C with temperature increases during the tests. Between the tests the temperature was returned to 25 °C to avoid sludge adaptation to thermophilic conditions. The tests were carried out at an organic loading rate (OLR) -1 and then repeated at an OLR of 12 g COD of 6 g COD L-1 R d -1. The latter load corresponded to an organic overload L-1 d R of the reactor. After each temperature increase, an immediate increase in the methane production was observed. Within 6 h of heating the methane production rate reached a maximum value and then slightly declined, as shown in Figure 2. As in the batch tests, small amounts of hydrogen (less than 1%) were observed when the temperature was above 40 °C. The observed peak of methane production was proportional to the temperature increase with the highest production rate at 48 °C. Also, the peak of methane production was more -1 and reached 5.7 pronounced at a load of 12 g COD L-1 R d -1 -1 L CH4 LR d . Improved COD removal during reactor heating was confirmed by COD measurements. At an OLR of 12 g COD -1 and a temperature of 25 °C the COD concentration L-1 R d reached 1.2-1.5 g L-1, while at 48 °C the COD concentration

FIGURE 2. Effect of a temperature increase on the methane production rate at an OLR of 12 g COD L-1 R . In each test at t ) 0 the temperature was changed from 25 °C to 30 °C (light gray line), 37 °C (dark gray line), or 48 °C (black line), while OLR was maintained at 12 g COD L-1 R during and at least 24 h prior to the test. remained below 0.6 g COD L-1. Also, during the organic overload at 25 °C the reactor pH dropped to 6.5, while at 48 °C pH remained above 6.8, which can be explained by a decreased solubility of carbon dioxide in the liquid phase due to the temperature increase. Interestingly, a comparison of methane yields at different reactor temperatures showed that during peak methane production due to heating, the apparent yield exceeded a theoretical value of 0.35 L CH4 (g COD)-1 and then returned to values below 0.35. In particular, a methane yield of 0.31 was estimated before the temperature augmentation shown in Figure 2. Shortly after the temperature increase to 48 °C the apparent yield reached a value of 0.47 and then declined to 0.33. Although methane solubility in water decreases with increasing temperature, a calculation showed that the methane release due to temperature variations can explain less than 2% of the observed increase in methane production. The increase in the yield can be explained by the hydrolysis and biodegradation of organic materials, which were accumulated in the sludge bed or were stored in the intracellular space. The increased hydrolytic activity was combined with increased methanogenic activity under thermophilic conditions thus resulting in increased methane yield. It can be hypothesized that under thermophilic conditions the growth rate of mesophilic methanogenic microorganisms declined, as was evidenced by a drop in the methane production later in the heating phase. Visual inspection of granular sludge after 2 months of reactor operation with intermittent heating showed no signs of granule disintegration. Over this period of time the granular sludge volume increased from 5 to 8 L. However, an attempt to extend thermophilic conditions in the reactor over a 24 h period resulted in a visually observable increase in the amount of solids in the effluent and detection of hydrogen in the biogas (results not shown). Apparently, fermentative microorganisms were more tolerant to the temperature increase, while the mesophilic methanogens sustained only short thermophilic periods. Application of thermophilic temperatures to mesophilic consortium for a sufficiently long period of time would require acclimatization of the microorganisms to thermophilic conditions (20). In addition to batch activity tests described above, sludge samples were withdrawn from the UASB reactor at the end of a 24 h period of operation at a temperature of 48 °C. These samples were used in activity tests carried out at temperatures of 25 and 55 °C. The tests were aimed at studying the recovery of mesophilic activity after sludge exposure to thermophilic conditions. Measurements of the methane production rate

FIGURE 3. Biogas production rates observed 15 min and 2 h after the test startup. The bottles were inoculated with sludge withdrawn from a reactor operating at 50 °C. The incubation temperature was 25 and 55 °C. showed trends similar to those obtained using the inoculum sludge (Figure 3). As in the initial tests, 15 min after test startup methane production was higher in the thermophilic bottles, while 2 h later methane production was higher at 25 °C. Also, at both mesophilic and thermophilic temperatures the biogas production rate was higher than that in the initial tests, which can be attributed to sludge adaptation to the synthetic wastewater composition. Overall, both batch activity tests and reactor runs suggested that a short exposure of mesophilic sludge to thermophilic temperatures (45-55 °C) has no long-term consequences on the biodegradation rates under mesophilic conditions. Meanwhile, a significant increase in the methane production rate was observed during the heating periods. This increase can be attributed both to increased enzymatic activity of the mesophilic methanogenic populations as well as an increased substrate availability due to enhanced hydrolysis. However, the exposure of mesophilic sludge to thermophilic conditions had to be limited in order to avoid changes in the mesophilic anaerobic consortium of microorganisms. Based on these observations, a reactor control strategy that uses temperature as a means for a temporary increase of the COD removal rate was proposed. Notably, in this control strategy the duration of sludge exposure to increased temperature had to be limited to avoid any changes in the mixed anaerobic consortium. 3.3. Temperature-Based Reactor Control. The proposed control strategy was tested in a 10 L UASB reactor operated -1 at a temperature of 25 °C and an OLR of 6 g COD L-1 R d . The reactor was periodically overloaded by increasing the -1 and the reactor temperature was OLR to 12 g COD L-1 R d increased during the overloads. The reactor temperature was controlled by an adaptive controller described above. Several reactor runs with setpoint effluent COD concentrations ranging from 400 to 1000 mg L-1 were carried out. During this experiment the pH control loop was disabled and reactor pH was stabilized by the bicarbonate buffer present in the influent. Figure 4 shows the results of an experiment in which the controller setpoint was 0.6 g COD L-1. Prior to the test, -1 and a COD the reactor produced 1.7 ( 0.1 L CH4 L-1 R d concentration of 0.34 ( 0.04 g COD L-1 was measured in the effluent. Also, the effluent was mostly composed of VFAs, which comprised 87% of the total COD content. The OLR -1 at t ) 0.1 day. In response, was changed to 12 g COD L-1 R d -1 the methane production rate increased to 3.4 ( 0.2 L L-1 R d and pH declined. Consequently, the organic overload process state was diagnosed by the knowledge-based system of the multi-model OBE (Figure 4f) and the temperature control of the reactor was activated. After a short transition period the adaptive controller stabilized the effluent COD concentration at a preset level of 0.6 g COD L-1 and the temperature was stabilized at 38 VOL. 41, NO. 3, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

981

FIGURE 5. Multiwavelength fluorometer calibration using analytical measurements of soluble COD (a) and total VFA (b) concentrations.

FIGURE 4. Dynamics of key process parameters (a-d), temperature (e), and process diagnosis (f) in a 10 L reactor during feedback control (0-1.1 day) and open-loop (1.1-3 day) experiments. COD and VFA analytical measurements are shown by black diamonds. Methane flow rate was normalized to 25 °C. °C while pH increased to 6.7. Similar to the temperature step-change experiments described above, the heating of the reactor resulted in a temporary peak of methane production with the methane production rate reaching a -1 (Figure 4c). The COD load was maximum of 5.2 L CH4 L-1 R d -1 at t ) 0.85 d. Shortly after the returned to 6 g COD L-1 d R OLR change the adaptive controller returned the reactor temperature to 25 °C. To compare reactor performance with and without temperature control, the reactor overload test was repeated without increasing the reactor temperature. As in the previous test, at startup the reactor was operated at a temperature of 25 °C and then the OLR was changed from 6 to 12 g COD L-1 R d-1 at t ) 1.1 d. As a result, the effluent COD concentration -1 reached 1.0 g L and pH declined to 6.5. A state of organic overload with a transition to the acidogenic state was diagnosed by the expert system (Figure 4f). The overload experiment had to be terminated in order to avoid reactor failure and at t ) 1.8 d the OLR was returned to 6 COD L-1 R d-1. The improvement in reactor stability due to temperature augmentation was reflected in the outputs of the knowledgebased system. The first OLR increase caused a transition to the organic overload state, which was promptly detected 982

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 3, 2007

(Figure 4f). The acidogenic state was avoided by increasing the reactor temperature, which increased the rate of COD conversion to methane. However, the knowledge-based system did not classify the reactor state as normal during the heating period, even though the effluent COD concentration did not exceed 0.6 g L-1. A detailed analysis of process outputs used in process state diagnosis showed that the classification was based on an increased carbon dioxide content in offgas. This effect can be attributed to a decreased solubility of carbon dioxide at higher temperatures, which was not considered by the knowledge-based system. The absence of the feedback control during the second OLR increase led to the development of acidogenic conditions in the reactor, which were evidenced by decreasing reactor pH. Consequently, the transition to the acidogenic process state was detected by the knowledge-based system (Figure 4f). While the overload period was short and the reactor performance recovered, a longer overload period would have caused irreversible changes in the mixed anaerobic consortium. Throughout the experiment on-line measurements of soluble COD and total VFA concentrations were carried out by a multiwavelength fluorometer. The regression models, which inferred the fluorescence spectra with the COD and VFA concentrations, were calibrated in the preliminary reactor tests using analytical measurements obtained at similar operating conditions. The fluorescence-based measurements were found to be pH-dependent at pH below 6.8. Consequently, pH values were included in the regression models. A comparison of on-line estimations with corresponding analytical values showed good accuracy of the fluorescence-based measurements thus demonstrating successful on-line monitoring of, at least, COD concentrations (Figure 5). However, VFA and COD concentrations throughout the experiment were highly correlated and the fluorescence of the influent components such as yeast extract was significantly higher than that of VFAs (results not shown). This implies that VFA measurements were based on the correlations with the fluorescent components of the media. A more detailed study would be required to prove the feasibility of on-line VFA measurements by multiwavelength

fluorometry. In conclusion, a combination of advanced process instrumentation with a novel control strategy allowed for successful stabilization of the reactor performance during organic overloads. Because the exposure time of anaerobic sludge to elevated temperature was limited, the imbalance of the mesophilic anaerobic consortium was avoided while the reactor degradation capacity was increased.

Acknowledgments This is NRC paper no 49026.

Notations A

heat transfer area

a

volatile fatty acid concentration

D

dilution rate

F

volumetric input flow rate

Fj

feed flow rate

f(T)

temperature function

K

heat transfer coefficient

n

total number of sub-models

m

total number of state variables

Qi

vector of gaseous output flow rates

r

vector of biotransformation rates

s

substrate concentration

T

reactor temperature

Tair

ambient temperature

Th

water jacket temperature

U

heat transfer coefficient

V

reactor volume

y

vector of multi-model outputs

Greek Letters θ

vector of kinetic parameters

θˆ

vector of estimated kinetic parameters

x

vector of state variables

ξˆ

vector of estimated state variables

ξin

vector of influent species concentration

βi

weight function

Ω, Γ

matrices of OBE design parameters

w, γ

OBE design parameters

l

tunable controller parameters

Literature Cited (1) Hobson, P. N.; Wheatley, A. D. Anaerobic Digestion: Modern Theory and Practice; Elsevier Applied Science: New York, 1993. (2) Pavlostathis, S. G.; Giraldo-Gomez, E. Kinetics of anaerobic treatment: a critical review. Crit. Rev. Environ. Control 1991,

21, 411-490. (3) Batstone, D. J.; Keller, J.; Angelidaki, R. I.; Kalyuzhnyi, S. V.; Pavlostathis, S. G.; Rozzi, A. G.; Sanders, W. T. M.; Siegrist, H.; Vavilin, V. A. Anaerobic Digestion Model No.1 Technical and Scientific Report No.13; IWA Publishing: London, UK, 2002. (4) Siegrist, H.; Vogt, D.; Garcia-Heras, J.; Gujer, W. Mathematical model for meso- and thermophilic anaerobic sewage sludge digestion. Environ. Sci. Technol. 2002, 36, 1113-1123. (5) Hao, X.; Heijnen, J. J.; Van Loosdrecht, M. C. M. Model-based evaluation of temperature and inflow variations on a partial nitrification-ANAMMOX biofilm process. Water Res. 2002, 36, 4839-4849. (6) van Lier, J. B.; Rintala, J.; Sanz Martin, J. L.; Lettinga, G. Effect of short-term temperature increase on the performance of a mesophilic UASB reactor. Water Sci. Technol. 1990, 22, 183190. (7) Visser, A.; Lettinga, G. Effects of short-term temperature increases on the mesophilic anaerobic breakdown of sulphate containing synthetic wastewater. Water Res. 1993, 27, 541550. (8) Speece, R. E.; Kem, J. A. The effect of short-term temperature variations on methane production. J. Water Pollut. Control Fed. 1970, 42, 1990-1997. (9) Ahn, J.-H.; Forster, C. F. The effect of temperature variations on the performance of mesophilic and thermophilic anaerobic filters treating a simulated papermill wastewater. Process Biochem. 2002, 37, 589-594. (10) Song, Y.-C.; Kwon, S.-J.; Woo, J.-H. Mesophilic and thermophilic temperature co-phase anaerobic digestion compared with single-stage mesophilic- and thermophilic digestion of sewage sludge. Water Res. 2004, 38, 1653-1662. (11) Morel, E.; Tartakovsky, B.; Guiot, S. R.; Perrier, M. Design of a multi-model observer-based estimator for anaerobic reactor monitoring. Comput. Chem. Eng. 2006, 31, 78-85. (12) Tartakovsky, B.; Morel, E.; Steyer, J. P.; Guiot, S. R. Application of a variable structure model in observation and control of an anaerobic digestor. Biotechnol. Prog. 2002, 18, 898-903. (13) Morel, E.; Santamaria, K.; Perrier, M.; Guiot, S. R.; Tartakovsky, B. Application of multi-wavelength fluorometry for on-line monitoring of an anaerobic digestion process. Water Res. 2004, 38, 3287-3296. (14) APHA, AWWA, WEF. Standard Methods for Examination of Water and Wastewater, 19th ed.; American Public Health Association/ American Water Works Association/Water Environment Federation: Washington, DC, 1995. (15) Zeng, Y.; Mu, S. J.; Lou, S. J.; Tartakovsky, B.; Guiot, S. R.; Wu, P. Hydraulic modeling and axial dispersion analysis of UASB reactor. Biochem. Eng. J. 2005, 25, 113-123. (16) Tartakovsky, B.; Morel, E.; Guiot, S. R. Application of VSMbased process control to a bench-scale anaerobic digestor. Ind. Eng. Chem. Res. 2005, 44, 106-113. (17) Morel, E.; Tartakovsky, B.; Guiot, S. R.; Perrier, M. On-line estimation of kinetic parameters in anaerobic digestion using observer-based estimators and multiwavelength fluorometry. Water Sci. Technol. 2006, 53, 77-83. (18) Perrier, M.; de Azevedo, S. F.; Ferreira, E. C.; Dochain, D. Tuning of observer-based estimators: Theory and application to the on-line estimation of kinetic parameters. Control Eng. Pract. 2000, 8, 377-388. (19) Bastin, G.; Dochain, D. On-Line Estimation and Adaptive Control of Bioreactor; Elsevier: Amsterdam, 1990. (20) Bouskova, A.; Dohanyos, M.; Schmidt, J. E.; Angelidaki, I. Strategies for changing temperature from mesophilic to thermophilic conditions in anaerobic CSTR reactors treating sewage sludge. Water Res. 2005, 39, 1481-1488.

Received for review July 28, 2006. Revised manuscript received November 17, 2006. Accepted November 22, 2006. ES0618043

VOL. 41, NO. 3, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

983