A new catalyst for the production of furfural from bagasse

Abstract

Furfural is a poisonous, flammable compound that is widely used in the chemical industry. The main raw materials for producing furfural are “pentosan-rich plant” components such as bagasse. In this study, furfural was produced in a pilot plant using sulfuric acid as a catalyst and an inorganic salt (NaHSO₄) as a promoter. The resulting furfural product was experimentally measured and reported in the presence of NaCl + H₂SO₄ or H₂SO₄, which were used as reaction catalysts. The obtained results show that H₂SO₄ plus NaHSO₄ is more effective for producing furfural than the other catalysts investigated. Furthermore, a three-layer feed-forward neural network analysis was applied to predict furfural production from the reactor. It was observed that the neural network systems could effectively predict the experimental results. The model for 385 experimental data under various conditions resulted in a squared correlation coefficient of 0.959 and a mean square error of approximately 10% for this analysis.

---

1. Introduction

Furfural is a commercial, poisonous, and flammable chemical substance. It is water soluble and has a boiling point of 161.8 °C. On the industrial scale, furfural can be produced by chemical hydrolysis performed on farm production residue, such as rice bran and corncob with bagasse, which is rich in pentosan. Furfural can also be produced from hydrolysis in an acidic environment. Just before immersion into the hydrolysis conditions, the prepared bagasse is usually blended with a dilute sulfuric acid solution. In these reactions, sulfuric acid acts as a catalyst. Pentosan-rich plant resources can also be utilized to produce furfural. Initially, pentosan is converted into a monosaccharide. Later, the monosaccharide is dehydrated to form furfural.

One of the issues with this process is that the rate of pentose hydrolysis can be several times higher than the rate of monosaccharide dehydration. In both reactions, furfural, as an effective monomer, reacts strongly with the chemical compounds of plant materials. Moreover, the reverse reaction occurs simultaneously. As a result, the generated furfural should immediately be removed from the reaction medium. Fig. 1 shows the overall diagram of furfural production from the bagasse. In 1821, furfural was first produced by Döbereiner, who aimed to produce formic acid from sugar and manganese dioxide.^1,2 In 1922, the primary industrial-scale production of furfural was achieved by Quaker Oats, who utilized oats and corncob as raw materials.³ Monomeric sugars, such as pentoses and hexoses, could be obtained from hydrolyzing wood and agricultural waste, which is credited to Henri Branconnot in 1810.⁴

Fig. 1 Schematic of furfural production.

The method of generation includes three sections, namely, raw material preparation, acid hydrolysis and the separation of the resulting furfural from residual pentosan. Pentosan is available in agricultural residue, which can be converted to xylose by simply hydrolysis. Xylose can be converted into furfural with the loss of approximately three molecules of water. Synthesized furfural may be purified through distillation and dehydration and it can be mixed within a vacuum.^5,6 Yang⁷ and Root⁸ conducted a comprehensive study on the kinetics of the furfural production reaction. Lamminpaa and coworkers also studied the kinetics of furfural decomposition in an acidic environment.⁹ They used formic acid as the reaction catalyst. They assumed that the reaction rate, with a reaction constant, was a function of the hydrogen ion concentration and temperature. In 2014, Danon and colleagues validated this procedure, as well as the kinetics of furfural production by evaluating pentose in acidic surroundings.¹⁰ They studied the production of furfural from bagasse with the combination of various inorganic salts and diluted H₂SO₄, which was used as a catalyst. Extensive studies were conducted to evaluate the kinetics of pentosan hydrolysis. Rong and coworkers utilized sulfuric acid plus NaCl or FeCl₃ as a catalyst. They showed that the combination of sulfuric acid plus FeCl₃ could be more effective than sulfuric acid plus NaCl.¹¹ In 2014, Hongsiri and coworkers studied the kinetics of xylose dehydration when generating furfural in a dilute acidic environment in the presence of salt. Salts could increase the production of furfural and decrease the formation of byproducts.¹² Liu and colleagues investigated the conversion of xylose to furfural in the presence of various inorganic salts without using an acid catalyst. In their study, FeCl₃ generated the largest increase in conversion rate compared to the other salts.¹³

Hosaka investigated furfural production conditions without a loss and carbonization of cellulose and observed considerable efficiency. According to Hosaka's studies, a temperature between 140 and 200 °C is appropriate for preventing cellulose loss. The utilization of hydrochloric acid, sulfuric acid, nitric acid, and other acids as catalysts in the furfural production process has been reported.³ Singh et al. reported the production of furfural and fermentable sugars using hydrolyzed bagasse and rice bran with dilute sulfuric acid in the presence of high-pressure water. The experiments were conducted using 0.4% sulfuric acid at 200 °C.¹⁴ The maximum efficiencies of furfural production from bagasse and rice bran were 11.5% and 10.9%, respectively.

As a result, from the viewpoint of chemical reaction engineering, the optimum production of furfural depends on various factors such as the initial concentration of pentosan in the feed, the added catalyst level, time of hydrolysis, hydrolysis temperature, and the retention time of the produced furfural in the reaction medium. Many studies have demonstrated the promising efficiency of furfural production per pentose unit in terms of the hydrolysis process, optimum time of hydrolysis, required catalyst, and steam consumption.¹⁵ Theoretically, furfural production from pentosan follows eqn (1):¹⁶

Several methods for efficient furfural production from pentosan in large quantities have been published. Some side reactions or the formation of intermediate compounds can also be effective in producing furfural. Likewise, the obtained results from Schonemann's investigations in multistage continuous reactors with a solvent extraction process demonstrated acceptable results with an increase of production efficiency from approximately 60% to 65%.³

In the current study, sulfuric acid plus sodium hydrogen sulfate (NaHSO₄ + H₂SO₄) were used as the catalyst. To predict the output furfural percent from reactors, a three-layer feed-forward neural network employing temperature, reactor pressure, reaction time, sulfuric acid level, and bagasse humidity was introduced. The application of the model for 385 experimental data under various conditions obtained a squared correlation coefficient of 0.959 and a mean square error of approximately 10%.

2. Experimental section

2.1. Materials

Bagasse with a moisture context of 45–70% was obtained from Karun Argo Industry Inc. All other materials used in this study were analytical grade and were purchased from Merck Co.

2.2. Experimental procedure

The experimental tests were performed in a stainless steel reactor with a diameter of 400 mm, a height of 1750 mm, and a thickness of 5 mm. Diluted sulfuric acid + aqueous solutions of different salts and 40 kg of bagasse were mixed in the reactor and then the temperature and pressure were increased to the set point. The reactor was heated to 160 °C for 7 minutes. After a suitable time, the produced furfural left the reactor from the top as a vapor. At the end of the hydrolysis process, the residue was removed from the reactor under proper pressure conditions. Subsequently, the furfural that was removed from the reactor was sent to the separator to separate the solid particles from any possible condensation. In the next step, aldehyde vapors were cooled and condensed in a condenser of stainless steel 316 (100 cm in length, 20 cm in diameter, and 5 mm in thickness) with 21 tubes (diameter of 1 inch). The remaining condensate was discharged to the tank, which comprised aldehyde. Subsequently, hydroxylamine (50 mL) and bromophenol (1 mL) were reacted with 1.5 g of the sample. The mixture was then treated with NaOH (1 N). Fig. 2 shows the pilot plant used in this study.

Fig. 2 Schematic of the pilot plant. (1) Stainless steel reactor; (2) vapor vessel; (3) inlet feed; (4) drain line; (5) outlet furfural; (6) cyclone; (7) inlet water; (8) condenser; (9) outlet water; (10) tank; (11) insulation of reactor.

3. Furfural production reactions

Furfural is generally obtained from pentosan-rich agricultural products. In this case, pentosan was hydrolyzed to pentose using an acid catalyst. Then, furfural was produced by pentose dehydration. The stoichiometry of the aforementioned reactions is expressed in the following equations:¹⁶

1. Pentosan hydrolysis

(C₅H₈O₄)_n + nH₂O → nC₅H₁₀O₅

2. Pentose dehydration

C₅H₁₀O₅ − 3H₂O → C₅H₄O₂

The overall reaction is as follows:

(C₅H₈O₄)_n − 2H₂O → C₅H₄O₂

Therefore, from a reaction kinetics point of view, furfural production can be explained in a two-step process. In the first step, pentosan (xylan) was hydrolyzed to xylose (pentose) and then xylose was converted to furfural. Some water, according to the number and type of monosaccharides in pentosan, is consumed to dissolve the pentosan (polysaccharide) and convert it into monosaccharides such as xylose and arabinose. In the next stage, each xylose molecule loses three water molecules and is converted into furfural. According to the explained cases, the acid catalyst seems to be involved in both of the above steps. Therefore, both the reaction rate in the two stages and the intensity of furfural production dramatically increase with increasing catalyst consumption. It is worth noting that excess catalyst in the furfural production process reactions results in the generation of unwanted byproducts. These byproducts are produced according to reactions (1) and (2). Thus, the amount of catalyst should be optimized to prevent excess hydrolysis, ultimately leading to a more efficient furfural production.^17–25

4. Artificial neural networks (ANNs)

One of the most well-liked simulation resources involves artificial neural networks, which incorporate a mixture of straightforward features and parallel computational tactic functionality. Through considering these features in the setting of biological nervous systems, these elements could be called neurons.²⁶ Many strategies for ANNs have been applied.^27,28 An ANN system comprises neurons and the weights associated with them. There are at least three layers of neurons in any model, including the input, hidden and output layers. The three-layer feed-forward neural network (3FFNN) is one of the best techniques with the shift purpose of a sigmoid (hyperbolic) type. This set of ANN can be used to generate nonlinear correlations with input and output parameters. The 3FFNN strategy is shown in Fig. 3. With this network, the input layer, the hidden layer (which contains ‘n’ neurons) and the output layer can be visualized. One of the most important parameters within the 3FFNN technique is ‘n’, which should be specified through the optimization system. This may occur immediately after generating the normal design connected with network. A mathematical representation of this course is as follows:

output(i) = W₂ tanh[W₁input(i) + b₁] + b₂ (2)

where i is the input data; W₁ is the weight, which in turn applies the first layer to the second layer; W₂ is the additional weight that helps relate the second layer to the third layer (an output layer); and b₁ and b₂ are the biases of the second and third layers, whose parameters are achieved with training. To resolve the challenge, the input data and result details need to be decided and outlined. Prior to producing a 3FFNN model, it is important to understand the details and resulting challenge. These guidelines usually include the number of inputs (nip), outputs (nop) and objects (nobj), which represent the number of data points with known input parameters (ip) and corresponding output parameters (op). The dimensions of each part in eqn (2) might be determined using n neurons. The input (i) shows a row with nip input parameters of the i^th subject. As a result, it is truly connected with the dimensions for nip × 1; in addition, W₁ is usually connected with the dimensions n × nip and W₁ × input (i), while b₁ should be connected with the dimensions n × nip and n × nip. Therefore, the second layer results would be of dimensions n × 1. Considering the size of W₂ and b₂, W₂ should be of size nop × n, b₂ should be of size nop × 1, and finally, the output (i) will be of size nop × 1. As discussed earlier, by considering the training process and unfamiliar guidelines for the 3FFNN model, W₁, W₂, b₁, and b₂, could well be received. Within this process, an optimization technique must be used to minimize the objective function, which involves the output and experimental data. In this system, the mean square of errors (MSE) is applied as an objective function for the training process of 3FFNN. Parameters, such as W₁, W₂, b₁, and b₂, play important roles when choosing the best optimization method. For starters, the determined process should be compatible with the defined difficulty, as well as correctly sorted. In terms of prior scientific studies, the Levenberg–Marquardt (LM) search engine optimization algorithm²⁶ is among the most suitable algorithms. As a result, this technique is utilized to determine the parameters in the present study. To determine the value of n from the 3FFNN, which usually describes the number of neurons in the hidden layer, there are many empirical correlations. Therefore, post optimization needs to be completed to determine the most effective value for n. To validate the obtained 3FFNN, as well as the prediction capability, the data set was divided into three groups, which are called the training set, test set and validation set. In this study, the test set, training set, and validation set were determined. The percentages of groups from the key files have been previously studied.²⁹

Fig. 3 Schematic of the Artificial Feed-Forward Neural Network.²⁷

5. Results and discussion

The percentage of produced furfural was calculated in the presence of NaHSO₄ + H₂SO₄. The acid percentage, bagasse humidity, temperature, and pressure are important parameters in furfural production. The results of produced furfural in dilute sulfuric acid solutions with different concentrations of NaHSO₄ are shown in Fig. 4. From Fig. 4, it can be observed that by increasing the amount of NaHSO₄, more furfural will be produced in the reactor. The results of furfural production in the presence of diluted sulfuric acid, including (NaHSO₄ + H₂SO₄), (NaCl + H₂SO₄) and H₂SO₄, are shown in Fig. 5. As can be observed in Fig. 5, a significant increase in furfural production is observed in the presence of NaHSO₄ + H₂SO₄. NaHSO₄ causes an interaction or complex formation with xylose, which improves the stabilization of the xylose transition states during dehydration and promotes furfural production. It seems that NaHSO₄ in dilute aqueous sulfuric acid can increase the rate of xylose reaction. In addition, NaHSO₄ increases the xylose reaction rate even at relatively low concentrations in acidic solutions. Therefore, this process increases the acid activity, boiling temperature and salting-out effect. In general, NaHSO₄ has a double positive effect on furfural production. First, it helps salt out the reaction product. Second, it increases the furfural selectivity and formation rate.

Fig. 4 The results of produced furfural in dilute sulfuric acid solutions with different concentrations of NaHSO₄.

Fig. 5 The results of produced furfural with various catalysts (bagasse humidity: 53%, H₂SO₄: 10%, NaHSO₄: 23%, NaCl: 23%).

Furthermore, the percentage of produced furfural versus the time in different conditions of bagasse and at various sulfuric acid in NaHSO₄ + H₂SO₄ solution concentrations is tabulated in Table 1. Table 1 summarizes the various bagasse humidities and acid percentages, ranging from 46% to 68% and from 7 to 13%, respectively. In addition, in Table 1, the temperature, pressure and percentage of sodium hydrogen sulfate in solution were fixed at 160 °C, 8 bar and 23% according to the optimum conditions. Then, the percentage of furfural was calculated at various time points from 10 min up to 110 min after the start of the reaction. The furfural level varied with the bagasse humidity or acid concentration under constant time conditions. This may be due to the presence of pentosane in bagasse, which decreases with increasing bagasse humidity. In fact, the humidity can alter the fermentation of pentosane. In other words, fermentation is an important parameter in furfural production and might change the acid percentage in the reactor. For instance, the acid percentage increases with increasing humidity or fermentation and decreases with decreasing humidity or fermentation. Table 2 compares the percentage of furfural formed in the reactor at various conditions and with various catalysts, including (NaHSO₄ + H₂SO₄), (NaCl + H₂SO₄) and H₂SO₄. The results are shown in Table 2 and demonstrate that the combination of sulfuric acid + sodium hydrogen sulfate can be an efficient catalyst in the furfural production process. Furthermore, the percentage of furfural leaving the reactor in the vapor phase in the presence of NaHSO₄ + H₂SO₄ (as a catalyst) is higher than for the systems using other catalysts. As mentioned earlier, an artificial neural network (ANN) was developed to calculate and predict the furfural percentage during the process. The temperature, reactor pressure, reaction time, sulfuric acid level, and bagasse humidity are input parameters in the artificial neural network (ANN). The results of the ANN for furfural production are listed in Table 1. Moreover, Fig. 6 shows the comparison between the ANN model results and experimental data, wherein the model results coincide well with the produced data.

Table 1 Comparison between the model results and experimental data^a

T (°C)	P (bar)	Bagasse humidity (%)	Dilute sulfuric acid (%)	Sodium hydrogen sulfate solution (%)	Time (min)	Furfural (%)		AAD%
						Exp.	Calc.
a
165	8	58	11	23	10	4.2	4.2434	4.95
165	8	58	11	23	20	5.1	5.0803
165	8	58	11	23	30	5.8	6.1816
165	8	58	11	23	40	6.5	7.2263
165	8	58	11	23	50	8.02	7.7416
165	8	58	11	23	60	7.8	7.1726
165	8	58	11	23	70	6.65	6.8002
165	8	58	11	23	80	6.64	6.5917
165	8	58	11	23	90	6.63	6.2374
165	8	58	11	23	100	5.4	5.706
165	8	58	11	23	110	4.6	5.0231
165	8	53	10	23	10	4.8	4.799	4.19
165	8	53	10	23	20	5.2	5.4978
165	8	53	10	23	30	6.4	6.4757
165	8	53	10	23	40	7.2	7.671
165	8	53	10	23	50	8.8	8.6199
165	8	53	10	23	60	7.3	7.3855
165	8	53	10	23	70	7	6.7489
165	8	53	10	23	80	6.7	6.5657
165	8	53	10	23	90	6.6	6.1589
165	8	53	10	23	100	5.6	5.5753
165	8	53	10	23	110	4.2	4.9012
165	8	46	7	23	10	4.1	4.3037	2.52
165	8	46	7	23	20	5.5	5.6113
165	8	46	7	23	30	7	6.9342
165	8	46	7	23	40	8.4	8.0289
165	8	46	7	23	50	8.8	8.8432
165	8	46	7	23	60	7.5	7.2592
165	8	46	7	23	70	6.4	6.4284
165	8	46	7	23	80	5.6	5.7943
165	8	46	7	23	90	5.1	5.1471
165	8	46	7	23	100	4.7	4.5703
165	8	46	7	23	110	4	4.1635
165	8	67	13	23	10	4.2	4.566	4.28
165	8	67	13	23	20	5.3	5.3542
165	8	67	13	23	30	6.7	6.3023
165	8	67	13	23	40	7.7	7.2393
165	8	67	13	23	50	8.5	7.7801
165	8	67	13	23	60	7.4	6.9918
165	8	67	13	23	70	6.8	6.3199
165	8	67	13	23	80	6	6.03
165	8	67	13	23	90	5.6	5.6513
165	8	67	13	23	100	5.1	5.1743
165	8	67	13	23	110	4.7	4.6284
165	8	49	11	23	10	4.4	4.4999	2.58
165	8	49	11	23	20	5.8	5.4972
165	8	49	11	23	30	6.3	6.6533
165	8	49	11	23	40	7.5	7.7765
165	8	49	11	23	50	8.6	8.5435
165	8	49	11	23	60	8.5	8.4347
165	8	49	11	23	70	7.8	7.7764
165	8	49	11	23	80	7.3	7.3819
165	8	49	11	23	90	6.7	6.6807
165	8	49	11	23	100	5.6	5.6838
165	8	49	11	23	110	4.8	4.4656
165	8	60	12	23	10	5.1	5.2312	3.68
165	8	60	12	23	20	5.8	6.3681
165	8	60	12	23	30	7.2	7.5948
165	8	60	12	23	40	8.8	8.6141
165	8	60	12	23	50	9.2	9.1138
165	8	60	12	23	60	8.1	8.2799
165	8	60	12	23	70	7.6	7.5196
165	8	60	12	23	80	7.3	6.964
165	8	60	12	23	90	6.4	6.2911
165	8	60	12	23	100	5.2	5.5724
165	8	60	12	23	110	4.7	4.832
165	8	58	13	23	10	4.1	4.3127	3.65
165	8	58	13	23	20	5.4	5.3576
165	8	58	13	23	30	6.9	6.6472
165	8	58	13	23	40	7.6	7.8937
165	8	58	13	23	50	8.4	8.6432
165	8	58	13	23	60	8.1	7.9289
165	8	58	13	23	70	7.2	6.9446
165	8	58	13	23	80	6.4	6.3329
165	8	58	13	23	90	5.1	5.6434
165	8	58	13	23	100	4.9	4.9171
165	8	58	13	23	110	4.5	4.2294
165	8	61	12	23	10	5.6	5.1921	6.75
165	8	61	12	23	20	6.3	6.1506
165	8	61	12	23	30	6.9	7.1365
165	8	61	12	23	40	7.5	7.8979
165	8	61	12	23	50	8.7	8.1834
165	8	61	12	23	60	7.4	7.2115
165	8	61	12	23	70	7	6.513
165	8	61	12	23	80	6.8	6.0947
165	8	61	12	23	90	6.1	5.5586
165	8	61	12	23	100	5.9	4.9494
165	8	61	12	23	110	4.1	4.3097
165	8	68	13	23	10	5.8	5.4877	3.83
165	8	68	13	23	20	6.3	6.3513
165	8	68	13	23	30	6.9	7.3239
165	8	68	13	23	40	7.5	8.2273
165	8	68	13	23	50	8.8	8.6855
165	8	68	13	23	60	7.2	7.7666
165	8	68	13	23	70	6.9	6.9922
165	8	68	13	23	80	6.4	6.6124
165	8	68	13	23	90	6.1	6.1401
165	8	68	13	23	100	5.7	5.5689
165	8	68	13	23	110	5.1	4.9332

---

Table 2 Furfural percentage from the reactor under various conditions with various catalysts

T (°C)	P (bar)	Bagasse humidity (%)	Dilute sulfuric acid (%)	Salinity%	Time (min)	Furfural (%)
						Calc. H2SO4	Calc. NaCl + H2SO4	Calc. NaHSO4 + H2SO4
165	8	53	10	23	10	6	7.2	4.8
165	8	53	10	23	20	4.5	8.02	5.2
165	8	53	10	23	30	3.51	7.8	6.4
165	8	53	10	23	40	2.5	6.3	7.2
165	8	53	10	23	50	1.75	5.7	8.8
165	8	53	10	23	60	1.25	5	7.3
165	8	53	10	23	70	1.2	4.1	7
165	8	53	10	23	80	1.2	3.2	6.7
165	8	53	10	23	90	1.1	2.4	6.6
165	8	53	10	23	100	1	1.9	5.6
165	8	53	10	23	110	0.96	1.6	4.2
165	8	67	13	23	10	3.7	4.7	4.2
165	8	67	13	23	20	3.9	5.6	5.3
165	8	67	13	23	30	4.5	6.4	6.7
165	8	67	13	23	40	3.1	4.3	7.7
165	8	67	13	23	50	2.2	3.8	8.5
165	8	67	13	23	60	2	3.1	7.4
165	8	67	13	23	70	1.9	2.9	6.8
165	8	67	13	23	80	1.8	2.3	6
165	8	67	13	23	90	1.5	2	5.6
165	8	67	13	23	100	0.8	1.9	5.1
165	8	67	13	23	110	0.5	1.2	4.7
165	8	60	12	23	10	3.2	2.8	5.1
165	8	60	12	23	20	4.1	4.2	5.8
165	8	60	12	23	30	5.7	4.7	7.2
165	8	60	12	23	40	3.9	5.3	8.8
165	8	60	12	23	50	2.6	5.9	9.2
165	8	60	12	23	60	2.2	6.3	8.1
165	8	60	12	23	70	1.9	5.8	7.6
165	8	60	12	23	80	1.6	4.6	7.3
165	8	60	12	23	90	1	3.7	6.4
165	8	60	12	23	100	0.7	2.9	5.2
165	8	60	12	23	110	0.3	2.1	4.7
165	8	61	12	23	10	1.8	5.9	5.6
165	8	61	12	23	20	2.4	6.2	6.3
165	8	61	12	23	30	4.2	7.3	6.9
165	8	61	12	23	40	3.5	7.4	7.5
165	8	61	12	23	50	3	7.5	8.7
165	8	61	12	23	60	2.8	7	7.4
165	8	61	12	23	70	2.6	6.6	7
165	8	61	12	23	80	2.1	6	6.8
165	8	61	12	23	90	1.8	5.4	6.1
165	8	61	12	23	100	1.5	5	5.9
165	8	61	12	23	110	1.2	4.5	4.1
165	8	68	13	23	10	2.1	3.1	5.8
165	8	68	13	23	20	2.9	4.7	6.3
165	8	68	13	23	30	4.6	4	6.9
165	8	68	13	23	40	4	3.7	7.5
165	8	68	13	23	50	3.7	3	8.8
165	8	68	13	23	60	3.1	2.8	7.2
165	8	68	13	23	70	2.8	2.1	6.9
165	8	68	13	23	80	1.5	2	6.4
165	8	68	13	23	90	0.95	1.8	6.1
165	8	68	13	23	100	0.83	1.3	5.7
165	8	68	13	23	110	0.4	1	5.1

---

Fig. 6 Comparison between the model results and experimental data in 3FFNN.

6. Conclusions

In this study, furfural production in the presence of sulfuric acid and an inorganic salt (NaHSO₄ and H₂SO₄) in a pilot scale were studied experimentally. It is concluded that:

- The addition of NaHSO₄ to sulfuric acid, used as a furfural production reaction catalyst, increases the furfural yield in the system.

- The percentage of furfural and the energy consumption in the presence of NaHSO₄ + H₂SO₄ (as a catalyst) is higher than that for the systems using other catalysts.

- The bagasse humidity, percentage of sulfuric acid, reaction time, temperature, and pressure were the strongest, most influential parameters in the furfural production process.

- Temperature, pressure and the percentage of sodium hydrogen sulfate in solution were fixed at 160 °C, 8 bar and 23% according to optimum conditions.

- A three-layer feed-forward ANN was successfully developed to represent and predict the furfural concentration in the reactor.

- The results demonstrate that the model was suitable to estimate the furfural percentage with a mean square error of about 10%.

Acknowledgements

The authors wish to thank managers of Behran Oil Company of Furfural Factory in Iran for their financial support and also from Islamic Azad University, Science and Research Branch for granting this research project.

References

1. W. Jaggle, Escher Wyss Mitteilungen, 1975, 2, 38–51.
2. J. W. Dunning and E. C. Lathrop, J. Ind. Eng. Chem., 1945, 37, 24–29.
3. E. Hagglund, Chemistry of Wood, Academic Press Inc, New York, 1951, p. 235.
4. E. Lin, An experimental investigation of Furfural production process, M. S. thesis, Mississippi State University, 1990.
5. J. Mcketta, Encyclopedia of chemical processing and design, Taylor & Francis, 1986, vol. 24, p. 40.
6. J. A. Horwath, R. Mutharason and E. D. Grossmann, Biotechnol. Bioeng., 1983, 25, 19–32.
7. W. Yang, P. Li, D. Bo and H. Chang, Carbohydr. Res., 2012, 357, 53–61.
8. D. Root, J. Seaman, J. Harris and W. Neil, For. Prod. J., 1959, 9, 158–165.
9. K. Lamminpaa, J. Ahola and J. Tanskanen, Ind. Eng. Chem. Res., 2010, 12, 6297–6303.
10. B. Danon, G. Marcotullio and W. De Jong, Green Chem., 2014, 16, 39.
11. C. Rong, X. Ding, Y. Zhu, Y. Li, L. Wang, Y. Qu, X. Ma and Z. Wang, Carbohydr. Res., 2012, 350, 77–80.
12. W. Hongsiri, B. Danon and W. De Jong, Ind. Eng. Chem. Res., 2014, 53, 5455–5463.
13. C. Liu and C. Wyman, Carbohydr. Res., 2006, 341, 2550–2556.
14. A. Singh, K. Das and D. K. Sharma, J. Chem. Technol. Biotechnol., 1984, 34, 51–61.
15. R. D. Sproull, P. R. Bienkowski and G. T. Tsao, Biotechnol. Bioeng. Symp., 1985, 15, 561–577.
16. K. J. Zeitsch, The chemistry and technology of Furfural and its many by-products, (Sugar series; 13), Elsevier Science, 1st edn, 2000.
17. Y. Wandian, L. Pingli, B. Dechen and C. Heying, Carbohydr. Res., 2012, 357, 53–61.
18. R. Cunguang, D. Xuefeng, Z. Yanvhao, L. Ying, W. Lili, Q. Yuning, M. Xiaoyu and W. Zichen, Carbohydr. Res., 2012, 350, 77–80.
19. T. Qiliang, F. Shiyu, Z. Huaiyu, C. Xinsheng and A. Lucian, J. Agric. Food Chem., 2008, 56, 3097–3101.
20. M. T. Garcia-Dominguez, J. C. Garcia-Dominguez, D. M. Feria, M. J. Gomez-Lozano, F. Lopez and M. J. Diaz, J. Chem. Eng., 2013, 221, 185–192.
21. Y. Wandian, L. Pingli, B. Dechen, C. Heying, W. Xiaowei and Z. Tao, Bioresour. Technol., 2013, 133, 361–369.
22. C. Branca, C. Di Blasi and A. Galgano, Ind. Eng. Chem. Res., 2010, 49, 9743–9752.
23. V. Choudhary, I. Sandler and D. G. Vlachos, ACS Catal., 2012, 2, 2022–2028.
24. E. Lam, J. H. Chong, E. Majid, Y. Liu, S. Hrapovic, A. C. W. Leung and J. H. T. Luong, Carbon, 2012, 50, 1033–1043.
25. F. Lopez, M. T. Garcia, M. J. Feria, J. C. Garcia, C. M. D. Diego, M. A. M. Zamudio and M. J. Diaz, J. Chem. Eng., 2014, 240, 195–201.
26. M. T. Hagan, H. B. Demuth and M. H. Beale, Neural Network Design, Martin Haugen, 2002.
27. F. Gharagheizi, A. Eslamimanesh, A. Mohammadi and D. Richon, Ind. Eng. Chem. Res., 2011, 50, 221–226.
28. A. Chouai, S. Laugier and D. Richon, Fluid Phase Equilib., 2002, 199, 53–62.
29. L. Piazza, G. Scalabrin, P. Marchi and D. Richon, Int. J. Refrig., 2006, 29, 1182–1194.

---