CHEMICAL ENGINEERING TRANSACTIONS VOL. 61, 2017 A publication of The Italian Association of Chemical Engineering Online at www.aidic.it/cet Guest Editors: Petar S Varbanov, Rongxin Su, Hon Loong Lam, Xia Liu, Jiří J Klemeš Copyright © 2017, AIDIC Servizi S.r.l. ISBN 978-88-95608-51-8; ISSN 2283-9216 Prediction of Performance of the Moving-Bed Biofilm Pilot Reactor Using Back-Propagation Artificial Neural Network (BP-ANN) Eleonora M.E. Rava, Evans M.N. Chirwa* Water Utilization and Environmental Engineering Division, Department of Chemical Engineering, University of Pretoria, Pretoria 0002, South Africa evans.chirwa@up.ac.za Coal gasification stripped gas liquor (CGSGL) wastewater contains large quantities of complex organic and inorganic pollutants which include phenols, ammonia, hydantoins, furans, indoles, pyridines, phthalates and other monocyclic and polycyclic nitrogen containing aromatics, oxygen- and sulphur containing heterocyclic compounds. Most conventional aerobic systems for coal gasification wastewater treatment are not sufficient in reducing pollutants such as chemical oxygen demand (COD), phenols and ammonia due to the presence of toxic and inhibitory organic compounds. The current paper reports on the degradation of aromatic compounds and the reduction of hard COD in CGSGL using a Moving-Bed Biofilm Reactor (MBBR) system. The inoculum contained a genetically enhanced mixed culture of Pseudomonas putida, Pseudomonas plecoglossicida, Rhodococcus erythropolis, Rhodococcus qingshengii, Enterobacter cloacae, Enterobacter asburiae strains of bacteria, seaweed and diatoma. Consistently high hard COD removal (>88 %) and degradation of targeted phenolic compounds (>93 %) was achieved in the reactor with no loss of biodiversity in the consortium culture. The performance of the reactor outside the observable range was projected using a Back-Propagation Artificial Neural Network (BP-ANN) developed in this study. 1. Introduction South Africa has no known petroleum reserves, but it has large quantities of coal reserves projected to last another 200 years under the current use. Coal can be converted to liquid fuel through a gasification liquification process. Sasol uses the patented Sasol-Lurgi process to convert low ranking coal to carbon monoxide as the first step towards the creation of the liquid fuel (Ginster and Matjie, 2005). Unfortunately, the low ranking coal used in the Sasol-Lurgi process contains large amounts of compounds with polar and hydrophilic functional groups such as OH, COOH, O, NH2, and SH (Molva, 2004). A number of coal gasification plants in China also use lignite coals as a raw material to produce gas via the Lurgi process. Jin et al. (2013) reported the presence of phenols, cyanides, thiocyanates, polycyclic aromatic hydrocarbons (PAHs), nitrogen-, oxygen-, and sulphur-containing heterocyclic compounds in the coking wastewater. In other related studies, coal gasification wastewater was reported to contained significant amounts of hydrogen sulphide (H2S) (Gai et al., 2008), ammonia (NH3) and carboxylic acids (Liu et al., 2013), and long chain alkanes (Ji et al., 2015). Due to the mixed nature and complexity of the compounds in the coal gasification wastewater, treatment of this wastewater using physical chemical processes tends to be expensive and most of the time ineffective. Currently, the non-biodegradable component of coal gasification wastewater is mostly removed by solvent extraction and ammonia stripping. The latter process produces a secondary stream of highly toxic effluent (Yang et al., 2006). Microorganisms containing the monooxygenase and dehydrogenase enzymes such as Pseudomonas putida (Shen and Wang, 1995), Pseudomonas aerigunosa (Oboirien and Chirwa, 2007), and Alcaligenes eutropha (Ornston, 1966) are known to facilitate ring cleavage of aromatic compounds leading to degradation of a range of aromatic compounds. Selection and packaging of such compounds into optimised culture soups could DOI: 10.3303/CET1761196 Please cite this article as: Rava E.M.E., Chirwa E.M.N., 2017, Prediction of performance of the moving-bed biofilm pilot reactor using back- propagation artificial neural network (bp-ann), Chemical Engineering Transactions, 61, 1189-1194 DOI:10.3303/CET1761196 1189 provide opportunities for treating highly toxic wastewater at a low cost. In this study, the removal of chemical oxygen demand (COD), phenols and ammonia-nitrogen in a hybrid fixed-film bioreactor (H-FFBR) inoculated with a mixed culture of bacteria containing aromatic compound organisms was investigated. To achieve the above objective, the compounds in the wastewater were characterised. Furthermore, the genetic makeup of organisms in the reactor was determined using PCR followed by genetic sequencing and 16S rRNA genotype fingerprinting. The performance data collected over a period of 6 months was used to calibrate a self-learning Back-Propagation Artificial Neural Network (BP-ANN) to be used for control purposes. The BP-ANN used in this study was originally adopted from Hajmeer and Basheer (2002). A version of the trainable neural network model was tested by Jacobs and Chirwa (2015) for the evaluation of parameters in a phenol-degrading/Cr(VI)-reducing culture. 2. Materials and methods 2.1 Moving-bed biofilm reactor setup The moved-bed hybrid fixed-film bioreactor (H-FFBR) consisted of 3 aeration compartments in series designated as Zone 1, Zone 2 and Zone 3 (Figure 1). The specific respective volumes for the Zones 1, 2 and 3 were 250 L, 150 L and 600, respectively, resulting in a total bioreactor volume 1,000 L. The bioreactor was acclimatised to the feed over a period of time such that the reactor was subjected to at least three complete bacterial sludge ages. Figure 1: Configuration of the moving-bed, hybrid fixed-film bioreactor pilot plant (H-FFBR) used in the study 2.2 Inoculum culture characterisation DNA primers 27F (5’-AGAGTTTGATCMTGGCTCAG-‘3) and 518R (5’-ATTACCGCGGCTGCTGG-‘3) were used in the amplification of the 16S rRNA gene using the Q5® Hot Start High-Fidelity 2X Master Mix (New England Biolabs, UK) PCR system following the manufacturer’s instructions. The amplicon libraries were purified using the Agencourt® Ampure® XP bead protocol (Beckman Coulter, USA). The library concentration was measured using the NEBNext® Library Quant assay kit for Illumina® (New England Biolabs, UK). The library pool was sequenced on a MiSeqTM (Illumina, USA) using the MiSeqTM Reagent kit v3, 600 cycles PE (Illumina, USA). The final pooled library was at 10 pM with 20 % PhiX as control. The DNA sequence for each pure colony was then uploaded to the Basic Local Alignment Search Tool (BLAST) of the National Center for Biotechnology Information (NCBI) (http://www.ncbi.nlm.nih.gov). A phylogenetic tree was constructed from the identified 16S rRNA sequences using the neighbor-joining method in the MEGA Version 6 software (Tamura et al., 2013). 2.3 Analytical methods 2.3.1 COD measurement Chemical oxygen demand (COD) was determined by oxidation of oxidizable COD using a sodium dichromate solution. Using readily oxizable substance, sodium dichromate is readily reduced to the trivalent (III) state while oxidising organic compounds (APHA, 2005). 2.3.2 Measurement of metals Samples for soluble metal analyses (sodium, calcium, iron, silica, potassium, copper, zinc, lead, vanadium, manganese, chromium, cobalt, nickel, aluminium, molybdenum and magnesium) were preserved with 10 mL 1190 concentrated nitric acid per litre sample and stored at 6±2 ºC. Samples were filtered through 0.45 µm membrane filters (Merck Millipore, South Africa) and analysed by inductively-coupled plasma-optical emission spectrometry (ICP-OES) (Agilent Technologies, USA) following the U.S. EPA Method 200.7 (1979) (U.S. EPA, 1979). The ICP-OES was operated and controlled using the installed Agilent ICP Expert II software (Agilent Technologies, USA). 2.3.3 Semi-volatile compounds Aromatic and semi-volatile compounds in water samples were characterised using the Perkin-Elmer Clarus 500 GC-MS equipped with Clarus 600T equipped with headspace (Perkin Elmer, Connecticut, USA). Separation of compounds in the GC was performed in a Perkin–Elmer Elite—5MS capillary column (30 m×0.25 mm ID ×0.5 µm fixed phase) with helium as a carrier gas with the GC operating in a split-less mode. The oven temperature was kept initially at 60 °C for 5 min, followed by an increase to 300 °C at a rate of 15 oC/min. 2.3.4 Polar aromatic compounds Hydrophilic aromatic compounds were measured using the Waters 2695 high performance liquid chromatograph (HPLC) (Waters Corporation, Massachusetts, USA) equipped with the Waters Photo Diode Array (PDA) detector operated at = 254 nm, and a Waters PAH C18 column (4.6 mm × 25 cm with 5 µm packing) operated at a column temperature of 25 oC. Extraction was done in a 1:1 ratio of sample and acetonitrile. 3. Results and discussion 3.1 Microbial culture composition The dominant kingdom classification for the biofilm was bacteria (99.8 %) with 0.2 % consisting of archaea, fungi and protozoa. The dominant phyla were Proteobacteria (75.0 %), Firmicutes (3.66 %), Bacteriodetes (5.16 %) and Actinobacteria (1.65 %). The dominant classes were Beta-proteobacteria (32.9 %), Alpha- proteobacteria (34.8 %), Gamma-proteobacteria (13.6 %), Bacteriodetes (4.14 %) and Actinobacteria (1.65 %). Uncultured bacteria comprised 14.53 % and 12.91 % of the uncultured phylum and class classifications respectively (Figure 2). The bacterial BLAST nucleotides were 99 % homologous to three strains of Pseudomonas putida, two strains of Enterobacter cloacae, one strain of Enterobacter asburiae and one strain of Rhodococcus erythropolis. These strains are tolerant and effective in high conductivity wastewaters and the hypothesis is that these strains will improve the biodegradation of petroleum hydrocarbons (oils and grease), alkanes and aromatic compounds typically found in CGSGL.The core genera and specific species in the actual bioengineered inoculum contained several Pseudomanads which are intended for the biodegradation of aromatic compounds. Several species from the Rhizobial sources were also identified (Table 1). Figure 2: Relative abundance of bacterial phyla in the H-FFBR (excluding unidentified phyla) 3.2 Moving-bed bioreactor performance Phenol (C6H6O); isoquinolinones (C9H7N) and saturated carboxylic acids (C4H8O2; C6H12O2; C7H14O2; C8H15O2; C16H32O2) were detected in the effluent when phenol (C6H6O), isoquinolinones (C9H7N), substituted cyclopentanones (C6H6O), substituted benzoic acids (C8H8O2) and phenylbutenones (C10H10O) were detected in the feed. Thus, these compounds were degraded slower when phenylbutenones (substituted aromatic alkenone) were present in the feed. Hydantoins (C5H8N2O2; C6H10N2O2) were only detected in the effluent 1191 when phenol (C6H6O), aniline (C6H5NH), substituted cyclopentanones (C6H6O), furans (C6H6O) and pyridines (C6H7N9) were detected in the feed. In this phase of the project, phenol was used as a surrogate for the majority of aromatic compounds confirmed to be present using GC-MS. The removal of total phenols ranged between 62 % and 93 % with an average of 78 %, thus 22 % of the total phenols were soluble, but non- biodegradable (Figure 3). The removal of phenols was affected by factors such as substrate inhibition effect, pH, temperature, biomass concentration, microbial community and their metabolic potential, and nutrient concentration. Biodegradation was also affected by the positions of the methyl groups of methyl phenols. The p-substituted phenols were more readily biodegradable than the m- or the o-substituted phenols since they are weaker electron donors (Rava et al., 2015). Table 1: Core genera and species across the H-FFBR Feed (RAS) Biofilm Suspended biomass (SB) Effluent (CA) Pseudomonas aeruginosa (AF440523.1) (6-8 %) Thauera butanivorans (NR_040797.1) (5-10 %) Ochrobactrum anthropi (AB120120.1) (2-4 %) Rhodoplanes cryptolactis (AB087718.1) (1-8 %) Rhodoplanes cryptolactis (AB087718.11) (2-5 %) Pseudaminobacter salicyclatoxidans (NR_028710.1) (3-5 %) Thauera butanivorans (NR_040797.1) (1-4 %) Pseudomonas putida (AE015451.1) (0.3-5 %) Xanthobacter polyaromaticivorans (AB106864.1) (1-3 %) Pseudomonas aeruginosa (AF440523.1) (1-3 %) Pseudomonas aeruginosa (AF440523.1) (1-3 %) Pseudomonas aeruginosa (AF440523.1) (0-4 %) Diaphorobacter nitroreducens (AB076856.1) (0.6-2 %) Diaphorobacter nitroreducens (AB076856.1) (1-2 %) Ancylobacter polymorphus (NR_04279.1) (0.6-2 %) Diaphorobacter nitroreducens (AB076856.1) (0-3 %) (a)Pooled data from sample set 1 and sample set 2 Figure 3: Figure showing removal rate of phenol across the H-FFBR system 3.3 Calibration and Application of BP-ANN (TANN) The neural network was trained using a back-propagation algorithm (Figure 4) with the goal of achieving a minimum net error on the validation data set while preventing overtraining or memorisation (Sewsynker and Kana, 2016). The experimental data set were randomly divided into two sets (i) 75 % of the data were used for training and (ii) 25 % of the data were used for the validation and testing process. A net error value on the validation data set of 0.018 was achieved after 3700 training epochs. The accuracy of the developed model 1192 was assessed on fourteen novel process conditions (validation data set). With this data set, the regression analyses on predicted and observed process outputs, and, the coefficients of determination (R2) were calculated for each model output. Figure 4: Back-propagation training algorithm used for the model ANN training, where MSE = mean square of differences and E = target minimum error between model data and measured values The validation was achieved by using the Artificial Neural Network (ANN) model to predict the process output values of Zone 1 Biofilm thickness, Zone 2 Biofilm thickness, Zone 3 Biofilm thickness, Zone 1 OUR, Zone 2 OUR, Zone 3 OUR, Suspended Biomass (MLSS), COD removal (%) and Phenol removal (%) based on fourteen process conditions not previously exposed to the model (Figure 5). The output values gave varied coefficients of determination (R2) up to 0.96 (average 0.85, excluding outliers). High coefficient of determination (R2>0.7) suggests a higher reproducibility and accuracy in the model when subjected to the novel H-FFBR operational conditions. Thus, the higher predicted outputs (R2>0.7) accounted for more than 70 % variation in the observed data. The only coefficients below 0.7 were for o-Phosphates removal (R2 = 0.06) and for phenol removal (R2 = 0.57 %). The removal of o-Phosphates was not considered significant enough to be estimated using the developed ANN model since R2 outliers have a negative effect on ANN based model development as was earlier observed by Rorke et al. (2017). Figure 5: Comparative regression (R2) values for each output using the Trained Artificial Neural Network (TANN). 4. Conclusions Diluted CGSGL was successfully treated in a pilot H-FFBR reactor using a bioengineered inoculum of phenol degrading organisms. A diverse microbial community in the inoculum removed most of the hard COD including phenolic compounds. However, nitrogen compounds were resistant to degradation due to the 1193 absence of autotrophic ammonia-oxidising bacteria in the inoculum. The microbial community diversities in the reactor were significantly dissimilar thus indicating that microbial communities were affected by the different growth environments in the H-FFBR. Part of the observed data was used to train a predictive Artificial Neural Network developed in this study which successfully predicted biofilm growth rate in phases not used during the training phase of the model. Acknowledgments The project was partially funded by National Research Foundation (NRF) through the Incentive Funding for Rated Researchers Grant No. IFR170214222643 awarded to Prof Evans M.N. Chirwa of the Department of Chemical Engineering, University of Pretoria. Buckman Africa provided funding for chemical and microbiological analyses. References APHA, 2005, Standard Methods for the Examination of Water and Wastewater - Centennial Edition. American Public Health Association, American Water Works Association, Water Environment Federation, USA. Gai H., Jiang Y., Qian Y., Kraslawski A., 2008, Conceptual Design and Retrofitting of the Coal-Gasification Wastewater Treatment Process, Chemical Engineering Journal, 138, 84–94. Ginster M., Matjie R.H., 2005, Beneficial Utilisation of Sasol Coal Gasification Ash. In: World of Coal Ash (WOCA), 11-15 April 2005, Lexington, Kentucky, USA, , accessed 26 26.03.2016. Hajmeer M., Basheer I., 2002, A Probabilistic Neural Network Approach for Modeling and Classification of Bacterial Growth/No-Growth Data, Journal of Microbiological Methods, 51, 217–226. Jacobs I., Chirwa E.M.N., 2015, Estimation of Reaction Parameters for Phenol Biodegradation Using Trainable Artificial Neural Networks, Chemical Engineering Transactions, 45, 1243–1248. Ji Q., Tabassum S., Yu G., Chu C., Zhang Z., 2015, Determination of Biological Removal of Recalcitrant Organic Contaminants in Coal Gasification Waste Water, Environmental Technology, 36, 2815–2824. Jin X., Li E., Lu S., Qui Z., Sui Q., 2013, Coking Wastewater Treatment for Industrial Reuse Purpose: Combining Biological Processes with Ultrafiltration, Nanofiltration and Reverse Osmosis, Journal of Environmental Science, 25, 1565–1574. Liu Y.S., Han H.J., Fang F., 2013, Application of Bioaugmentation to Improve the Long-Chain Alkanes Removal Efficiency in Coal Gasification Wastewater, Fresenius Environmental Bulletin, 22, 2448–2455. Molva M., 2004, Removal of Phenol from Industrial Wastewaters using Lignitic Coals, MSc dissertation, Izmir Institute of Technology, Turkey. Oboirien B.O., Chirwa E.M., 2007, Bioremediation of Organic Pollutants in a Radioactive Wastewater: Evaluation of Feasibility of Treatment, Chemical Engineering Transactions, 12, 673–678. Ornston L.N., 1966, The Conversion of Catechol and Protocatechuate To -Ketoadipate by Pseudomonas putida. Journal of Biological Chemistry, 241, 3776–3786. Rava E., Chirwa E., Allison P., van Niekerk M., Augustyn M.P., 2015, Removal of Hard COD, Nitrogenous Compounds and Phenols from a High-Strength Coal Gasification Wastewater Stream, Water SA, 41, 441– 447. Rorke D.C.S., Suinyuy T.N., Kana E.B.G., 2017, Microwave-Assisted Chemical Pre-Treatment of Waste Sorghum Leaves: Process Optimization and Development of an Intelligent Model for Determination of Volatile Compound Fractions, Bioresource Technology, 224, 590600. Sewsynker Y., Kana E.B.G., 2016, Intelligent Models to Predict Hydrogen Yield in Dark Microbial Fermentations using Existing Knowledge, International Journal of Hydrogen Energy, 41, 12929–12940 Shen H., Wang Y.-T., 1995, Simultaneous Chromium Reduction and Phenol Degradation in a Coculture of Escherichia coli ATCC 33456 and Pseudomonas putida DMP-1, Applied and Environmental Microbiology, 61, 27542758. Tamura K., Stecher G., Peterson D., Filipski A., Kumar S., 2013, MEGA6: Molecular Evolutionary Genetics Analysis version 6.0, Molecular Biology and Evolution, 30, 27252729. U.S.EPA, 1979, Determination of Metals and Trace Elements in Water and Wastes by Inductively Coupled Plasma-Atomic Emission Spectrometry - Method 200.7, Revision 4.4, EMMC Version; Environmental Monitoring Systems Laboratory; U. S. Environmental Protection Agency: Cincinnati, Ohio. Yang C., Qian Y., Zhang L., Feng J., 2006, Solvent Extraction Process Development and On-Site Trial-Plant for Phenol Removal from Industrial Coal-Gasification Wastewater, Chemical Engineering Journal, 117, 179–185. 1194