Modeling of an Industrial Wastewater Treatment System ......Modeling of an Industrial Wastewater Treatment System Using Historical Process Data IAN KIT NEHEMIAH M. DAP-OG, MM [email protected]

284

Liceo Journal of Higher Education Research

Modeling of an Industrial Wastewater Treatment System Using Historical Process Data

IAN KIT NEHEMIAH M. DAP-OG, [email protected]

Liceo de Cagayan University

Abstract - In this paper, the predictive capacity of a wastewater treatment process model was studied to find out how well the model works and to propose its use as a tool to help improve control of effluent BOD (biochemical oxygen demand) for the wastewater treatment facility being studied. The mathematical process model for the existing facility was obtained from literature with few added modifications to accommodate certain requirements of the actual setup. A Complete Mix Flow Reactor with recycle was assumed together with Monod and Contois growth models for the microbial growth kinetics assumptions. Using historical process data from the daily operations of the plant, model parameters were estimated and then verified using linear regression. The results of the study show that the use of historical process data posed some limitations to modeling that resulted to moderate correlations between observed and predicted values for effluent BOD except for the prediction of average MLSS. Despite these observations, it was found out that Monod-based model works better than the Contois-based model for the wastewater treatment system studied.

Keywords: Historical process data, Monod, Contois, Model, Industrial wastewater

Date Submitted: August 6, 2008Final Revision Accepted: December 14, 2008

Vol. 6 No. 1 December 2009 ISSN: 2094-1064CHED Accredited Research Journal, Category B

Liceo Journal of Higher Education ResearchScience and Technology Section

285

INTRODUCTION

Industrial effluents such as wastewater constitute a major impact on the environmental condition of receiving bodies of water and ultimately human health. It is of prime concern, therefore, for government’s regulatory agencies to ensure that industries comply with effluent standards through technologies of wastewater treatment facilities that reduce the amount and concentration of harmful water pollutants. Today, industries are able to manage the pollutants found in their effluents to safe levels. However, wastewater treatment plant operators and engineers still encounter operational problems due to uncertainties in influent characteristics and operational variations, which affect the quality of effluent. In this regard, a tool can be developed, such as a process model that can be used to consider the effects of process variables on the performance of the biological reactor. Koloini et al. (2001) found the effective use of mathematical modeling for process optimization and troubleshooting for full-scale treatment plants that use activated sludge process provided that model predictions and experimental or field observations agree.

ProblemThis study sought to answer the question: Is the mathematical

process model chosen in the study effective in predicting the effluent BOD and biomass concentration of the actual wastewater treatment facility being modeled?

Importance of the StudyThis study will provide wastewater treatment operators and

engineers with an additional tool to help them in predicting the quality of effluent based on other process variables. Utilization of historical process data, which is readily available to the operators and engineers, will be an alternative to the more commonly used experimental data obtained from pilot or bench scale reactors.

For academic purposes, the study is deemed to provide engineering students with a fuller grasp of the underlying principles in wastewater treatment modeling through participation in data

286


analysis. Local communities will be benefited from more reliable

performances of industrial wastewater treatment facilities installed near the dwellings of residents.

Future researchers, who will be interested in modeling wastewater treatment plants especially for industry sector, will gain insights into the application of field data to modeling exercise.

Literature Review

Industrial process wastewaters vary in terms of volume and pollutants present. The type of treatment applied prior to disposal will depend on these factors. The contaminants may be classified as suspended solids, dissolved solids, inorganic pollutants, organic pollutants, and pathogenic microorganisms. In general, the treatment of these water contaminants may be grouped into physical, biological, and chemical treatment methods (Henry and Heinke, 2000). For organic contaminants, biological treatment is applied that makes use of the ability of microorganisms to decompose organic matters present in wastewater.

There are now several technologies being used to carry out effective biological treatment of wastewater. The most common is the conventional activated sludge process that has several modifications based on treatment requirements and design constraints. The biological reactions that take place during the treatment of wastewater by microorganisms can be described by chemical equations where organic matter is converted by action of microorganisms with oxygen to carbon dioxide, water, new cells, and other by-products (Metcalf and Eddy, 1991).

Using the principle of material balance for constant mix flow reactor (CMFR) type, growth kinetics, and reactor design, the differential equation will be simplified by steady state assumption to come up with the following model equations:

287

The second equation makes use of Monod’s kinetic growth model to describe bacterial activity. When Contois growth model is used, the second equation should be in the form below.

288


The last two equations differ in a way that Contois considers microbial growth inhibition by biomass. Linearization of these equations will simplify the determination of the required parameters.

In the context of mathematical modeling, several works have been published and many deal with pilot-scale experiments for actual wastewater systems. Remarkable agreement between the mathematical model predictions and the real pilot plant quantities on the effluent parameters for steady-state conditions has been obtained on studies with wastewaters (Koloini et al., 2001). A mathematical model for an activated sludge unit treating 2,4 dichlorophenol (DCP) containing synthetic wastewater was also developed where kinetic constants like specific rate constant (k), saturation constant (Ks), and DCP inhibition constants were estimated using experimental data obtained at different detention times and sludge age (Eker and Kargi, 2006). The study was able to produce model predictions using the estimated constants that were in good agreement with experimental data. In another mathematical modeling study, predictions using Monod and Contois kinetic models were conducted and good correlation of the predicted and experimental values in the effluent BOD was obtained (Hu et al., 2001). The Contois model correlation coefficient was 0.9899, which was higher than 0.9797 of Monod.

289

Some other studies especially on municipal wastewater treatment systems show results that favor Monod kinetics with parameters in agreement with those found in the literature (Kapagiannidis et al., 2006). Alvarez et al, (1991) also found that Monod’s equation adequately described the aerobic biodegradation rates of benzene and toluene by microbial population of a sandy aquifer.

OBJECTIVES OF THE STUDY

This study aimed at coming up with a verified a mathematical process model of a biological treatment system using historical process data from actual operations.

Specifically, the study sought to:

1. determine the model parameters µm, Ks, Y, and kd. 2. determine the best-fit kinetic growth model for the biomass. 3. determine whether the mathematical model, with the use of

the model parameters, will have a high predictive capability on effluent BOD and biomass concentration.

Scope and Limitation

1. The model used actual process data of the treatment facility instead of the commonly used experimental data from bench-scale reactors.

2. The actual physical setup (reactor) is modelled using the complete mix flow reactor type.

3. The BOD concentration was estimated from the COD data by a correlation factor.

4. MLSS was used for biomass concentration instead of MLVSS.5. pH, temperature, and concentration of other nutrients were

considered constant.

METHODOLOGY

The study used mathematical modeling process that involves parameter estimation and model verification using actual process

290


data. The model verification process is depicted in the figure below.

Historical process data from daily operations of the wastewater treatment plant was gathered and subjected to data analysis. About a year’s operation of the wastewater treatment plant was retrieved from operations’ log sheets that included the sludge age (SRT or θc) , influent BOD (S0), effluent BOD (S), MLSS in the reactor (X), return activated sludge (RAS), tank volume, hydraulic retention time (HRT), and flow rates in and out of the reactor. Other parameters like pH, nutrients, and temperature were considered constant and not considered in the model.

Initial treatment of raw data was necessary to eliminate incomplete data sets. The entire data collected were split into two by random selection. The data were tabulated as shown in the appendices. The training set (appendix A) with 113 data sets was used to estimate model parameters while the verification set (appendix B) with 112 data sets was used to verify the model. Linear regression was used in both the estimation and verification steps.

For estimation of parameters Y and kd, the quantities 1/SRT and S0-S/XHRT were calculated using the training set, tabulated, and then regressed based on equation 1(Metcalf and Eddy, 1991). The values of the slope and intercept of the regression equation

Figure 1. The modeling process

291

obtained with Microsoft excel software were equated to parameters Y and kd respectively.

The other two model parameters, namely the saturation constant (Ks) and the maximum specific growth rate (µm), were estimated using equation 2 for Monod-based model and equation 3 for Contois-based model. The quantities XHRT/S0-S, 1/S, and X/S were calculated using the training set, tabulated, and then regressed based on the two equations. The values of the slope and intercept were used to compute for parameters Ks and µm.

Verification of the model was accomplished by getting the correlation coefficient between observed values taken from the verification set and the ones calculated using model equations 4, 5, and 6.

od

c

SS1Equation 1: Yk −=−

s

o

KEquation 2( ):SS kS k

Monod =+−

sx

o

KEquation 3( ):SS kS k

Contois =+−

sd c

cm - d

K( 1 kEquation 4(Monod):S µ

+=

sx dc

cm - d

XK (1 kEquation 5(Contois):S µ

+=

c

dc

Equation 6: X =+

292


RESULTS

Regression of the quantity1/SRT with S0-S/XHRT for estimation of parameters Y and kd is depicted in figure 1.

Figure 1. Regression Using Training Set to Estimate Y and kd

The result above shows a slope of – 0.017, intercept of 0.032, and a correlation coefficient of 0.002. This correlation is rather poor with the yield coefficient (Y) negative and endogenous decay (kd) positive. This is contrary to what is normal where yield coefficient must be positive and decay coefficient negative based on the model equation.

Realizing this, data was subjected to an initial treatment before using them as input to the model. The ratio between (1/SRT) and (S0-S/XHRT) was obtained and the mean and standard deviation calculated. Then the relative distance between individual data points and the mean was determined to establish a basis for determining which data set is more useful for the model. Those data whose distances fall within one standard deviation were considered as input for the model. Furthermore, the need to increase correlation among variables in the regression process requires that some data that affect negatively the result be excluded. The result for the

293

estimation of the first two model parameters is shown in figure 2.

Figure 2. Regression With Some Data Excluded to Estimate Y and kd

The slope, corresponding to Y is now positive (0.506), and the intercept corresponding to kd is 0.001 da-1.

For the other two model parameters, namely the saturation constant (Ks) and the maximum specific growth rate (µm), the regression results are shown in figures 3 and 4.

Figure 3. Regression to Estimate Ks and µm (Monod) using

294


Treated Data

Figure 3 shows an intercept of 5.787, which corresponds to 1/k based on equation 2 that is then converted to µm equals 0.0875 da-1 since µm = kY. Monod saturation constant is Ks = 57.3 mg/l, obtained from the product of slope of 331.6 and the value of k.

Figure 4. Regression to Estimate Ksx and µm (Contois) Using Treated Data

Using Contois assumption, figure 5 shows an intercept of 7.211, which corresponds to µm of 0.0688 da-1 and the Contois saturation constant of Ksx = 0.0104 mg/l.

After the four model parameters were determined, verification followed and the results are shown in figures 5, 6, and 7.

295

Figure 5. Correlation of Observed and Predicted Effluent BOD Using Monod Assumption

Figure 5 shows a moderate correlation coefficient of 0.369 for Monod-based model, which is much better compared to that of the Contois-based model with coefficient of 0.296 as shown in figure 6. The correlation for the prediction of the average MLSS is the lowest of all, having only a coefficient of 0.006 as shown in figure 7.

Figure 6. Correlation of Observed and Predicted Effluent BOD Using Contois Assumption

296


Figure 7. Correlation of Observed and Predicted Average MLSS

DISCUSSIONS AND CONCLUSIONS

The results of parameter estimation and model verification are summarized in the tables below.

Table 1. Summary of results for parameter estimation

Model parameters Estimated Values

Y (yield coefficient), mg MLSS/mg BOD5kd (endogenous decay coefficient), da-1

µm (maximum specific growth rate), da-1

Ks (Saturation constant), mg/l BOD

0.506 0.0010.0875 (Monod) , 0.0688 (Contois)57.3 (Monod) , 0.0104 (Contois)

Table 1 shows the values of the model parameters estimated using historical process data. Comparing these with values from literature for municipal wastewater treatment plants, Y is around 0.6 mgVSS/mgBOD5 while endogenous decay coefficient kd is normally in the range of 0.025 to 0.075 per day. Ks will normally be in the range from 15 to 70 mg/l COD and µm is around 0.3 per day. However, the values obtained in this modeling exercise do not all coincide with literature values, which could be attributed to the fact

297

that industrial wastewater which, is studied here, differs much in terms of physical, biological, and chemical characteristics to that of municipal wastewater. These parameters were used in the chosen model to predict the effluent BOD and the average MLSS.

Table 2. Summary of results for model verification

Variable Predicted Correlation coefficient Slope intercept

BOD out

Average MLSS

0.369 (Monod)0.296 (Contois)

0.006

0.6551.0930.136

0.855-10.961850

Table 2 shows that the model used in this study moderately satisfied the requirement for prediction. The Monod-based model got a coefficient of 0.369, much higher than the Contois-based model with just 0.296. Howeveer, this is not as high as compared to results obtained by other studies due to the fact that industrial process may contain large amount of data but less value in terms of information (George et al., 2009). The prediction of average MLSS suffered more than the effluent BOD with a coefficient of 0.006. This aspect needs to be given more attention considering the characteristics of sludge in the reactor, how they are sampled, and determined.

Several other things can be pointed out with respect to these observations. One is the number of assumptions used to simplify the model at the expense of model accuracy. This factor cannot be avoided but may be minimized if the researcher has more control over his data or has a way of obtaining additional information not usually determined during normal plant operations. Schraa et al. (2006) emphasized the necessity of analyzing historical data to determine if additional data is necessary. There may also be a need to characterize influent wastewater that requires information not usually recorded by most industrial wastewater treatment plants. For instance, most modeling requires the use of ultimate BOD or Biodegradable COD for input data, but in this study only a correlation for BOD to COD is used because it is the only one available. George et al. (2009) also pointed out the use of multivariate statistical tools like PCA in detecting fault in process. This factor may also be one of the sources of much variability in the

298


behavior of historical process data where advanced statistical tools are necessary for preparation of data intended for modeling.

In addition, another source of non-ideal behavior in modeling could be the occurrence of non-steady state flow in the process as production rates vary from time to time since lots of models are developed based on steady state assumption. Moreover, the model that was used in the study is based on the complete mix reactor setup, while the actual setup is a complete mix in series that can be better modeled with plug flow analysis but with more complicated mathematics.

Based on results and observations, it is realized that a more effective modeling exercise for a wastewater treatment plant may include several considerations. One is control over the variables ,which puts experimental methodologies of more relevance over the use of plain historical process data especially in processes as complex as wastewater treatment. However, when historical data are readily available, the consideration of cost and practicality may prove its importance to modeling. Another is the use of process faults detection methods using advance statistical tools that may help improve the quality of data. Subjecting data for initial data analysis and characterization will be also of much help in improving inputs for the model. Still another possibility of improving the effectiveness of historical process data is whether the present facility setup and the wastewater system allows for the collection of supplementary information to check the quality of data for a more reliable calibration of the model.

LITERATURE CITED

Alvarez, Pedro J.J., Anid, Paul J., and Vogel, Timothy M. (1991). Kinetics of Aerobic Biodegradation of Benzene and Toluene in Sandy Aquifer Material. Kluwer Academic Publishers. Retrieved August 26, 2009, from http://www.bren.ucsb.edu/academics/courses/214/Lectures/Lecture_2_ESM214_05.ppt. 1991

299

DENR Administrative Order No. 34 Series of 1990. (n.d.). Retrieved October 1, 2008, from http://www.emb.gov.ph/laws/water%20quality%20management/dao90-34.pdf.

Eker, Serkan and Kargi, Fikret. (2006, March 1). Kinetic Modelling and Parameter Estimation for an Activated Sludge Unit Treating 2,4 Dichlorophenol Containing Synthetic Wastewater. Environmental Engineering Science. Retrieved September 20, 2009, from http://www.liebertonline.com.

George, Joval P., Chen, Zheng, and Shaw Philip. (2009). Fault Detection of Drinking Water Treatment Process Using PCA and Hotelling’s T2 Chart. World Academy of Science Engineering and Technology. Retrieved March 7, 2010, from

http://www.waset.org/juornals/waset/v50/v50-166.pdf

Henry, J. Glynn and Heinke, Gary W. (2000). Environmental Science and Engineering, 2nd ed. Pearson Education Asia.

Hu, W.C., Thayanithy, K., and Forster, C.F. (2001, September). Kinetics Study of Anaerobic Digestion of Sulfate-Rich Wastewaters from Manufacturing Food Industries. 7th International Conference on Environmental Science and Technology. Retrieved July 18, 2009, from www.srcosmos.gr/srcosmos/showpub.aspx?aa=4311.

Kapagiannidis, A.G., Vaiopoulou, E., and Aivasidis A. (2006). Determination of Kinetic Parameters in a Pilot Scale BNR System Treating Municipal Wastewater. GOBAL NEST. Retrieved August 26, 2009, from http://www.gnest.org/Journal/Vol8_No1/paper_11_KAPAGIANNIDIS_398.pdf.

Koloini, T, Drolka M., and Plazyl, I. (2001, May 18). The Results of Mathematical Model and Pilot Plant Research of Wastewater. Chemical and Biochemical Engineering Quarterly. Retrieved July 18, 2009, from http://www.fkit.hr/cabeq/pdf/15_2_2001/Koloini.pdf.

300


Metcalf & Eddy. (1991). Wastewater Engineering Treatment Disposal Reuse, 3rd ed. McGraw-Hill, Inc.

Schraa, Oliver, Robinson, Paul, & Selegran, Anette. (2006). Modeling of an IFAS Process With Fungal Biomass Treating Pharmaceutical Wastewater. Water Environment Foundation. Retrieved January 19, 2010, from http://www.environmental-expert.com/Files%5C5306%5Carticles%5C8894%5C135.pdf

Documents

Modeling of an Industrial Wastewater Treatment System ......Modeling of an Industrial Wastewater Treatment System Using Historical Process Data IAN KIT NEHEMIAH M. DAP-OG, MM [email protected]