Integrated Biorefinery of Empty Fruit Bunch from Palm Oil Industries to Produce Valuable Biochemicals

Empty fruit bunch (EFB) utilization to produce valuable bio-chemicals is seen as an economical and sustainable alternative to waste management in palm oil industries. This work proposed an integrated biorefinery configuration of EFB valorization considering sustainability pillars—namely, economic, environmental, and safety criteria. Techno-economic analysis, life cycle assessment, and hazard identification ranking methods were used to estimate annual profit, global warming potential (GWP), fire explosion damage index (FEDI), and toxicity damage index (TDI) of the proposed integrated biorefinery. A multi-objective optimization problem was then formulated and solved for simultaneous maximization of profit and minimization of GWP, FEDI and TDI. The resulting Pareto-optimal solutions convey the trade-off among the economic, environmental, and safety performances. To choose one of these optimal solutions for implementation, a combined approach of fuzzy analytical hierarchy process and a technique for order preference by similarity to ideal solution was applied. For this selection, the economic criterion was more preferred, followed by the safety and environmental criterion; thus, the optimal solution selected for integrated biorefinery configuration had the highest annual profit, which was at the maximum capacity of 100 ton/h of EFB. It can fulfill the global demand of xylitol (by 55%), levulinic acid (by 98%), succinic acid (by 25%), guaiacol (by 90%), and vanillin (by 12%), and has annual profit, GWP, FEDI, and TDI of 932 M USD/year, 284 tonCO2-eq, 595, and 957, respectively.


Introduction
Today, the issue of sustainability has been a major concern associated with palm oil production. The palm oil industry has brought a positive impact on the economic growth in Malaysia since it created jobs as well as downstream activities for national development [1]. It has been the major The proposed integrated biorefinery model is depicted in Figure 1, where EFB generated from palm oil mill activity is collected and sent to a centralized integrated biorefinery area. The potential of converting other palm oil wastes to valuable products is not considered in this work. A flowchart of the methodology of the present study is outlined in Figure 2. EFB composition is taken from Chiesa et al. [47]. Dilute acid pretreatment, enzyme production, and saccharification processes are selected and simulated in Aspen Plus V10 according to Humbird et al. [48]. The number of conversion steps for this work is limited to five biochemical products according to the potential revenues, and they are xylitol [49,50], levulinic acid [51], succinic acid [52], guaiacol [53], and vanillin [54]. The process simulation results are then used for equipment sizing and sustainability assessment. and biomass allocation lead to maximum profit, the minimum of both global warming potential and hazard level. The obtained results (the so-called Pareto-optimal solutions) reveal the trade-off among the objectives. In the subsequent step, a combined approach of Fuzzy Analytical Hierarchy Process (FAHP) and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) is applied to select one of the Pareto-optimal solutions.

Methodology
The proposed integrated biorefinery model is depicted in Figure 1, where EFB generated from palm oil mill activity is collected and sent to a centralized integrated biorefinery area. The potential of converting other palm oil wastes to valuable products is not considered in this work. A flowchart of the methodology of the present study is outlined in Figure 2. EFB composition is taken from Chiesa et al. [47]. Dilute acid pretreatment, enzyme production, and saccharification processes are selected and simulated in Aspen Plus V10 according to Humbird et al. [48]. The number of conversion steps for this work is limited to five biochemical products according to the potential revenues, and they are xylitol [49,50], levulinic acid [51], succinic acid [52], guaiacol [53], and vanillin [54]. The process simulation results are then used for equipment sizing and sustainability assessment. For sustainability assessment of the integrated biorefinery, TEA is used to estimate the annual profit, LCA to quantify GWP, and HIRA to measure the hazard potential at initial design in terms of FEDI and TDI. These multiple criteria lead to a MOO problem to simultaneously maximize annual profit and minimize GWP, FEDI, and HIRA. The decision variables involved are biomass availability and product yields, and the constraints are related to mass and energy balances and global products demand. Prior to solving MOO problems, the developed process simulation models are simplified as surrogate models relating decision variables with the objectives [56]. The surrogate (regression) models are developed from rigorous simulation results obtained through a central composite design (CCD) for later use in solving the MOO problem. The solution of MOO problems produces Paretooptimal solutions, which reveal the trade-off among the objectives. To choose one of these solutions, FAHP and TOPSIS are used; FAHP takes into account the subjective inputs of experts with profound knowledge [57].
The proposed superstructure for biorefinery based on EFB is shown in Figure 3. The underutilized EFB supply is collected and transported to a central processing facility. Transportation costs and supply chain issues are not considered in this work. The EFB then undergoes a pre- Figure 1. Palm-oil-based integrated biorefinery concept [55].
For sustainability assessment of the integrated biorefinery, TEA is used to estimate the annual profit, LCA to quantify GWP, and HIRA to measure the hazard potential at initial design in terms of FEDI and TDI. These multiple criteria lead to a MOO problem to simultaneously maximize annual profit and minimize GWP, FEDI, and HIRA. The decision variables involved are biomass availability and product yields, and the constraints are related to mass and energy balances and global products demand. Prior to solving MOO problems, the developed process simulation models are simplified as surrogate models relating decision variables with the objectives [56]. The surrogate (regression) models are developed from rigorous simulation results obtained through a central composite design (CCD) for later use in solving the MOO problem. The solution of MOO problems produces Pareto-optimal solutions, which reveal the trade-off among the objectives. To choose one of these solutions, FAHP and TOPSIS are used; FAHP takes into account the subjective inputs of experts with profound knowledge [57].
The proposed superstructure for biorefinery based on EFB is shown in Figure 3. The underutilized EFB supply is collected and transported to a central processing facility. Transportation costs and supply chain issues are not considered in this work. The EFB then undergoes a pre-treatment process to break down its major components into cellulose, hemicellulose, and lignin, which will be used as the feedstock to subsequent processes to produce xylitol, levulinic acid, succinic acid, guaiacol, and vanillin.     Figure 3. Schematic of the integrated biorefinery from the empty fruit bunch (EFB).

Techno-Economic Analysys
TEA of the processes is estimated using Aspen Process Economic Analyzer (APEA) V.10. The basis for this estimation is the simulated results and equipment sizing. Investment factors to calculate project capital expenditure are taken from Peters and Timmerhaus [58], as tabulated in Table S1 in the Supplementary Materials. The estimation of annual operating cost is based on Turton et al. [59], and details of this estimations are reported in Table S2 in the Supplementary Materials. The annual profit is calculated as the difference between the product annual sales and total cost, which is the sum of annualized capital expenditure (CAPEX) and annual operating expenditure (OPEX). The economic objective is to maximize annual profit ($/year), as shown in Equations (1)

Life Cycle Assessment
The environmental impact assessment of the integrated biorefinery adopts the LCA technique using the ISO standard [60]. All emissions are taken from the developed process simulation such as direct emissions of the process, electricity and heat consumption. The emission data related to

Techno-Economic Analysys
TEA of the processes is estimated using Aspen Process Economic Analyzer (APEA) V.10. The basis for this estimation is the simulated results and equipment sizing. Investment factors to calculate project capital expenditure are taken from Peters and Timmerhaus [58], as tabulated in Table S1 in the Supplementary Materials. The estimation of annual operating cost is based on Turton et al. [59], and details of this estimations are reported in Table S2 in the Supplementary Materials. The annual profit is calculated as the difference between the product annual sales and total cost, which is the sum of annualized capital expenditure (CAPEX) and annual operating expenditure (OPEX). The economic objective is to maximize annual profit ($/year), as shown in Equations (1)- (3).
Here, B prod m (kg/h) is the mass production rate of products (xylitol, levulinic acid, succinic acid, guaiacol, and vanillin) and C prod m ($/kg) is the product price. Ca jk ($/year) is annualized CAPEX of the pretreatment and final processes. The CAPEX is the sum of total direct plant cost, total indirect plant cost, contractor's fee, contingency and fixed capital investment. Op jk ($/year) is equal to sum of direct manufacturing cost, fixed manufacturing cost and general manufacturing cost.

Life Cycle Assessment
The environmental impact assessment of the integrated biorefinery adopts the LCA technique using the ISO standard [60]. All emissions are taken from the developed process simulation such as direct emissions of the process, electricity and heat consumption. The emission data related to electricity and heat consumption are acquired from commercial LCA databases such as Ecoinvent 3.5 [61] and Simapro 8.5.3 [62]. Then, the total Life Cycle Inventory (LCI) can be formulated as a function of the direct emission, electricity and heat consumption, as stated in Equation (4).
The LCI is then translated into the corresponding environmental impact (GWP). The emissions considered are carbon dioxide (CO 2 ), methane (CH 4 ), and nitrous oxide (N 2 O). The GWP is calculated as the total of GWP from each of these emissions as shown in Equations (5) and (6).
In this equation, m i is the damage factor that accounts for each greenhouse emission, which is retrieved from Guinee [63]. GWP j denotes the GWP in the pretreatment process while GWP k indicates that of the chemical production process.

Inherent Safety
Two hazards potential, quantified in this work, are FEDI and TDI [46]. FEDI is based on thermodynamic data, where the energy factor is the main aspect of the calculation. Another aspect is penalty, which is given based on the operating range of the process. The penalties involve five categories-namely, storage units, units involving physical operations, units involving chemical reactions, transportation units, and other hazardous units. Finally, the damage potential of FEDI is found by the multiplication of the penalties and the energy factors. For TDI calculation, toxic load contained in the processing unit is the major factor. Physical and chemical reaction units are considered. Penalties typically assigned for the location of the nearest hazardous unit and space occupied by the unit are neglected because of lack of required data during this conceptual design phase. The formulation of FEDI and TDI is stated in Equations (7)-(16) Here, F 1 , F 2 , F 3 , and F 4 are chemical, physical, and reaction energy, respectively. M is the mass flow rate of the chemical (kg/s), Hc is the heat of combustion of the chemical (kJ/kg), and K is a constant (3.148). In Equations (8) and (9), Pp (kPa) and Vp (kPa) are the process pressure and vapor pressure of the chemical at process temperature. V is the volumetric flow rate of chemical (m 3 /h). TDI involves a G factor and several penalties. G factor is obtained from A (phase condition), and m is the anticipated release rate in kg/s.

Constraints
Biomass availability constraint is shown in Equation (17).
This shows that EFB consumed in the production process is within a minimum and maximum range of EFB supplied. EFB is delivered to plant using a truck with a capacity of 25 ton/truck [64]. The maximum allowable number of trucks to transport EFB is taken to vary from 2 trucks/h to 4 trucks/h. This range is set based on the assumed biomass unloading time of 15-30 min/truck and only one unloading facility within the plant. The assumption of 15 min/truck is based on about 5-10 min of unloading time, and the remaining time is for traveling in and out of the plant. This number becomes the logistic limit within the plant, which in turn limits the maximum production to 4 trucks/h. The same reasoning applies to 2 trucks/h.
Chemical mass yield information is gathered from the experimental data of references and they are modelled in min and max scenarios, which are used in the process simulation, as stated in Equation (18).
The mass yield of each process X k is set within the range of X kmin and X kmax , which are determined from the experimental data of xylitol [49,50], levulinic acid [51], succinic acid [53], guaiacol [53], and vanillin [54]. The yield information is modelled in min and max scenarios which are used in the process simulation. The data are tabulated in Table 1. The biochemical products are expected to fulfill the global demand for each product, as shown in Equation (19) and tabulated in Table 2.
The products obtained from the integrated biorefinery P m must be equal to or greater than the global demand.

Genetic Algorithm
Genetic algorithm (GA) is a directed random search technique that is modeled on the natural evolution/selection process toward the survival of the fittest [70]. In briefly, this algorithm begins with an initial population of chromosomes or trial solutions or individuals, each characterized by a set of values of decision variables in MOO problem. The individuals are generated randomly within the lower and upper bounds of decision variables, and are supplied to calculate objective functions (annual profit, GWP, FEDI, and TDI). Individuals from one generation are used to create a new population, based on selection and reproduction by crossover and mutation. The crossover is carried out on randomly selected parent individuals and generates offspring by swapping parts of the parent chromosomes. Afterward, mutation occurs by randomly altering the offspring created by crossover. This drastic change helps to prevent solutions from being stuck near a local optimum. Thus, the algorithm generates different individuals for the next generation (i.e., new population) from the previous population. These steps are repeated until some termination criterion (e.g., maximum number of generations or improvement of the best solution) is satisfied.
One adaptation of GA for MOO is gamultobj code in MATALAB, which was used in this study. Table 3 lists the values of GA parameters used in the present study. Bounds on the decision variables are tabulated in Table 4. The developed model equations used in multi-objective genetic algorithm are given in Section B in the Supplementary Materials.

Multi-Criteria Decision Making
To choose one preferred solution from the Pareto-optimal frontier, FAHP and TOPSIS are applied. Both these are outlined in this section.

Fuzzy Analytical Hierarchy Process
FAHP is a combination of fuzzy logic and linguistic variables [71]. In this work, such a method is applied to overcome the uncertainty in the judgment of decision makers in their selection of the best optimum solution [72]. The steps involved in the computation of criterion weights using FAHP are described in this sub-section.

Generating a hierarchy system evaluation
This work is to find the best optimal biorefinery superstructure configuration incorporating the sustainability criteria, which are comprised of economic, environment and safety aspects. The Pareto-optimal solutions generated by solving the MOO problem are the alternatives. The detailed decision model is presented in Figure 4. best optimum solution [72]. The steps involved in the computation of criterion weights using FAHP are described in this sub-section.

Generating a hierarchy system evaluation
This work is to find the best optimal biorefinery superstructure configuration incorporating the sustainability criteria, which are comprised of economic, environment and safety aspects. The Paretooptimal solutions generated by solving the MOO problem are the alternatives. The detailed decision model is presented in Figure 4.

Create a fuzzified pairwise comparison matrix
In this work, the decision maker's judgment is gathered according to the questionnaire, which was taken from Yan [73] and is summarized in Table S6 in the Supplementary Materials. Afterward, the pairwise comparison of the evaluation criteria is made based on a linguistic scale. The calibrated linguistic fuzzy scale suggested by Promentilla et al. [74] is implemented in this work, as shown in Table 5. The result of fuzzy pairwise comparison judgments can be expressed in the form of a matrix. The matrix A is an m x m real matrix, where m is the number of evaluation criteria. Each entry ajk of a matrix A represents the importance of the criterion. If ajk > 1, the j th criterion is more important than the k th criterion; if ajk < 1, the j th criterion is less important than the k th criterion. Then, the weight of the criteria and the rating of alternatives concerning can be calculated using Equation (20). Very Strongly More VS (6.0, 8, 9.5) 3. Aggregation of DM's judgment Then, fuzzy judgments of all DMs are aggregated using Equation (21):

Create a fuzzified pairwise comparison matrix
In this work, the decision maker's judgment is gathered according to the questionnaire, which was taken from Yan [73] and is summarized in Table S6 in the Supplementary Materials. Afterward, the pairwise comparison of the evaluation criteria is made based on a linguistic scale. The calibrated linguistic fuzzy scale suggested by Promentilla et al. [74] is implemented in this work, as shown in Table 5. The result of fuzzy pairwise comparison judgments can be expressed in the form of a matrix. The matrix A is an m x m real matrix, where m is the number of evaluation criteria. Each entry a jk of a matrix A represents the importance of the criterion. If a jk > 1, the jth criterion is more important than the kth criterion; if a jk < 1, the jth criterion is less important than the kth criterion. Then, the weight of the criteria and the rating of alternatives concerning can be calculated using Equation (20).

Aggregation of DM's judgment
Then, fuzzy judgments of all DMs are aggregated using Equation (21):

Computation of criteria weights
The criteria weights are then deduced through the non-linear programming model proposed by Tan et al. [75], as stated in Equations (22) and (23). This model approximates the criteria weights by constraining the consistency index (λ) within the fuzzy bound. A positive λ indicates a consistent fuzzy pairwise judgment given by the expert; hence, the criteria weights elicited are acceptable and can be applied for further computation. Note that λ = 1 suggests perfect consistency in preserving the order of preference intensities. Maximize λ Subject to

. Technique for Order Preference by Similarity to Ideal Solution
The methodology of TOPSIS is briefly explained below. Application and assessment of TOPSIS to many problems are available in Wang et al. [76].

1.
Calculate a normalized matrix The decision matrix of X ij is normalized using the following Equation (24): 2.
Calculate weighted normalized matrix The normalized decision matrix is multiplied by the weight, w i , obtained from Equations (22) and (23). Calculation is performed by the following Equation (25): 3.
Calculate the ideal best and ideal worst value The positive ideal solution A* and the negative ideal solution A − are determined using the following Equations (26) and (27): 4. Calculate Euclidean distance from the positive/negative ideal The distance from the positive and negative ideal for each alternative can be computed by the Euclidean distance as given in Equations (28) and (29).
5. Calculate performance score The relative closeness of the ith alternative is defined as The ranking priority of alternatives is determined based on the higher score of CC * j .

Results and Discussion
Prior to MOO, a sustainability assessment of each process was conducted to see its economy, safety, and environment impact. These results are summarized in Tables S4 and S5 and Figure S1 in the Supplementary Materials. Three cases of MOO are explored in this work. They are two cases of bi-objective (maximize annual profit and minimize GWP, and maximize annual profit and minimize hazard potentials) and one case of tri-objective (maximize the annual profit and minimize GWP, FEDI, and TDI), as shown in Table 6. The considered constraints are applied in all cases. Selection of one optimal solution from the non-dominated solutions is employed in case of tri-objective optimization as it includes all three pillars of sustainability. This step of MOO was adapted from Rangaiah et al. [77], and the MS Excel-based program developed by Wang and Rangaiah [76] was employed as a tool for multi-criteria decision-making. The MOO problem was formulated in MATLAB and solved using the multi-objective genetic algorithm code (gamultiobj).  Figure 5 shows the Pareto-optimal solutions for Case 1 bi-objective optimization, where the x-axis represents the annual profit and y-axis is GWP. Each Pareto-optimal point represents the optimal design of the biorefinery superstructure with a unique combination (trade-off) between the annual profit and GWP. Three solutions (marked S1, S2, and S3 in Figure 5) on the Pareto-optimal front are chosen as examples of optimal solutions for further analysis. Table 7 shows the decision variables at the maximum (S1), intermediate (S2), and minimum (S3) points of the Pareto-optimal front, and Table 8 presents the main results at these optimal solutions. It can be seen that the environmental impact is decreasing from S1 to S2 to S3 at the expense of decreasing annual profit. Ideally, the profit has to be maximized and the GWP minimized simultaneously. However, as depicted in the Pareto-optimal front, the improved performance of one objective function is obtained only at the expense of deteriorating another objective function. In this manner, Figure 5 has 35 optimal solutions, and each of them is equally good.
Point S1 denotes the solution with the highest annual profit and GWP. In this solution, the supply of EFB and mass yield of process reaches the maximum values, as well as the mass allocation of cellulose, hemicellulose, and lignin. In particular, cellulose is allocated to producing levulinic acid (10%) and succinic acid (90%), while lignin is to produce guaiacol (70%) and vanillin (30%). Here, the allocation of cellulose to succinic acid is higher due to greater global demand for it, even though levulinic acid production shows better performance in economy and environmental impact due to having a lower capital investment and annual production cost and higher product yield and selling price as well as lower GWP score. The same goes to producing more guaiacol than vanillin because of higher global demand. There is 30% of lignin used for vanillin production due to having better performance in economy and environmental impact. Xylose (obtained by conversion of xylan contained in hemicellulose by dilute acid pretreatment) is allocated 100% to produce xylitol since it has no other option. Figure 6 shows the integrated biorefinery configuration of solution S1 where 100 ton/h of EFB is supplied to the process, which corresponds to the maximum EFB supplied to the plant. This amount of EFB produces 18.9 ton/h xylitol, 1.77 ton/h levulinic acid, 22.1 ton/h succinic acid, 5.04 ton/h guaiacol, and 0.5 ton/h vanillin. Total resulting annual profit and GWP scores are 923 M USD/year and 285 tonCO 2 -eq. Point S2 is an intermediate solution, where a slight decrease of capacity (99 ton/h) results in lower profit by 1.95% and also GWP by 2%. The corresponding biorefinery configuration has annual profit and GWP of 905 M USD/year and 283 tonCO 2 -eq, respectively. Like in point S1, 10% cellulose is sent to produce levulinic acid while the remaining goes to produce succinic acid, and 70% of lignin is to produce guaiacol and the rest goes to vanillin. Finally, point S3 represents the lowest of both profit and GWP. Here, the allocation of cellulose and lignin are similar to points S1 and S3. The S3 solution gives an annual profit of 880 M USD/year and GWP of 281 tonCO 2 -eq, where the required EFB is supplied at 97 ton/h to produce xylitol (18.4 ton/h), levulinic acid (1.37 ton/h), succinic acid (22 ton/h), guaiacol (5.01 ton/h), and vanillin (0.45 ton/h), as depicted in Figure 7. It is clear that the allocation of cellulose and lignin show similar percentage in all Pareto-optimal solutions. With slightly higher capacity, both profit and higher environmental impact are higher. The plant capacity, on the other hand, has EFB supply truck as its bottleneck. Hence, to increase profit, more EFB supply trucks have to be organized, which leads to a logistic issue within the plant. results in lower profit by 1.95% and also GWP by 2%. The corresponding biorefinery configuration has annual profit and GWP of 905 M USD/year and 283 tonCO2-eq, respectively. Like in point S1, 10% cellulose is sent to produce levulinic acid while the remaining goes to produce succinic acid, and 70% of lignin is to produce guaiacol and the rest goes to vanillin. Finally, point S3 represents the lowest of both profit and GWP. Here, the allocation of cellulose and lignin are similar to points S1 and S3. The S3 solution gives an annual profit of 880 M USD/year and GWP of 281 tonCO2-eq, where the required EFB is supplied at 97 ton/h to produce xylitol (18.4 ton/h), levulinic acid (1.37 ton/h), succinic acid (22 ton/h), guaiacol (5.01 ton/h), and vanillin (0.45 ton/h), as depicted in Figure 7. It is clear that the allocation of cellulose and lignin show similar percentage in all Pareto-optimal solutions. With slightly higher capacity, both profit and higher environmental impact are higher. The plant capacity, on the other hand, has EFB supply truck as its bottleneck. Hence, to increase profit, more EFB supply trucks have to be organized, which leads to a logistic issue within the plant.   Table 8. Optimal objective values corresponding to the selected Pareto-optimal points in Figure 5. S1 S2 S3 Annual profit (M$/year) 923 905 880 GWP (tonCO2-eq) 285 283 281 S1 S2 S3 Figure 5. Pareto-optimal front for profit versus global warming potential. Table 7. Decision variables that correspond to the selected Pareto-optimal points in Figure 5.  Figure 6. The biorefinery configuration of the optimal point S1 for profit and global warming potential objectives.

Case 2: Maximize Profit and Minimize Hazard Potential
This scenario considers the maximization of profit and minimization of FEDI and TDI. Figure  8a,b represents the Pareto-optimal curve for annual profit versus FEDI and annual profit versus TDI from the MOO results. Three solutions on each curve are identified for comparison. Tables 9 and 10 give the corresponding sets of optimal decision variables and objectives, respectively. For profit and Figure 6. The biorefinery configuration of the optimal point S1 for profit and global warming potential objectives.  Figure 6. The biorefinery configuration of the optimal point S1 for profit and global warming potential objectives.

Case 2: Maximize Profit and Minimize Hazard Potential
This scenario considers the maximization of profit and minimization of FEDI and TDI. Figure  8a,b represents the Pareto-optimal curve for annual profit versus FEDI and annual profit versus TDI from the MOO results. Three solutions on each curve are identified for comparison. Tables 9 and 10 give the corresponding sets of optimal decision variables and objectives, respectively. For profit and

Case 2: Maximize Profit and Minimize Hazard Potential
This scenario considers the maximization of profit and minimization of FEDI and TDI. Figure 8a,b represents the Pareto-optimal curve for annual profit versus FEDI and annual profit versus TDI from the MOO results. Three solutions on each curve are identified for comparison. Tables 9 and 10 give the corresponding sets of optimal decision variables and objectives, respectively. For profit and FEDI scenario, point S1 represents the solution with the highest of both annual profit and FEDI score of 923 M USD/year and 592, respectively. In this solution, the maximum of 100 ton/h of EFB is necessary with the same allocation of cellulose as in Case 1 (all three solutions in Figure 5). Interestingly, lignin allocation is now different, where 80% is sent to guaiacol and the remaining goes to vanillin. This is because guaiacol production has lower FEDI score than vanillin. The use of ethyl acetate in extraction step increases FEDI score of vanillin process. Figure 9 shows biorefinery configuration for solution S1 in Figure 8a. This biorefinery fulfills global demand of xylitol by 57%, levulinic acid by 95%, succinic acid by 25%, guaiacol by 100% and vanillin by only 9%.
Point S2 as a middle solution in Figure 8 yields annual profit and FEDI of 864 M USD/year and 577, respectively. The same allocation of cellulose is obtained here, while all lignin is allocated to produce guaiacol due to lower FEDI score than that of vanillin process. In point S3, the lowest annual profit and FEDI score are obtained, which are 789 M USD/year and 570, respectively. The cellulose allocation is only 3% to levulinic acid, and the rest goes to succinic acid because levulinic acid production has a higher flammability and explosion risk compared to the production of succinic acid. This is due to severe operating conditions and hazardous chemicals involved in levulinic acid production. On the other hand, lignin is fully allocated to guaiacol production as it results a lower FEDI score than vanillin production. The lower solution (S3 in Figure 8a) on the Pareto-optimal curve has the lower FEDI score of guaiacol. Its configuration in Figure 10 shows that 96 ton/h of EFB is supplied to the biorefinery to produce xylitol (18 ton/h), levulinic acid (0.43 ton/h), succinic acid (22.5 ton/h), and guaiacol (6.6 ton/h). FEDI scenario, point S1 represents the solution with the highest of both annual profit and FEDI score of 923 M USD/year and 592, respectively. In this solution, the maximum of 100 ton/h of EFB is necessary with the same allocation of cellulose as in Case 1 (all three solutions in Figure 5). Interestingly, lignin allocation is now different, where 80% is sent to guaiacol and the remaining goes to vanillin. This is because guaiacol production has lower FEDI score than vanillin. The use of ethyl acetate in extraction step increases FEDI score of vanillin process. Figure 9 shows biorefinery configuration for solution S1 in Figure 8a. This biorefinery fulfills global demand of xylitol by 57%, levulinic acid by 95%, succinic acid by 25%, guaiacol by 100% and vanillin by only 9%. Point S2 as a middle solution in Figure 8 yields annual profit and FEDI of 864 M USD/year and 577, respectively. The same allocation of cellulose is obtained here, while all lignin is allocated to produce guaiacol due to lower FEDI score than that of vanillin process. In point S3, the lowest annual profit and FEDI score are obtained, which are 789 M USD/year and 570, respectively. The cellulose allocation is only 3% to levulinic acid, and the rest goes to succinic acid because levulinic acid production has a higher flammability and explosion risk compared to the production of succinic acid. This is due to severe operating conditions and hazardous chemicals involved in levulinic acid production. On the other hand, lignin is fully allocated to guaiacol production as it results a lower FEDI score than vanillin production. The lower solution (S3 in Figure 8a) on the Pareto-optimal curve has the lower FEDI score of guaiacol. Its configuration in Figure 10 shows that 96 ton/h of EFB is supplied to the biorefinery to produce xylitol (18 ton/h), levulinic acid (0.43 ton/h), succinic acid (22.5 ton/h), and guaiacol (6.6 ton/h).  Table 9. Decision variables that correspond to the selected optimal solutions in Figure 8.  Figure 8. Pareto-optimal fronts for (a) profit versus fire explosion damage index and (b) profit versus toxicity damage index. Table 9. Decision variables that correspond to the selected optimal solutions in Figure 8.  Figure 9. The optimal biorefinery configuration of the optimal point S1 for profit and fire explosion damage index in Figure 8a.  . The optimal biorefinery configuration of the optimal point S1 for profit and fire explosion damage index in Figure 8a.  Figure 9. The optimal biorefinery configuration of the optimal point S1 for profit and fire explosion damage index in Figure 8a.  The optimal biorefinery configuration of the optimal point S3 for profit and fire explosion damage index in Figure 8a. Table 10. Optimal objective values corresponding to the selected Pareto-optimal points in Figure 8.

Profit and FEDI
Profit and TDI S1 S2 S3 S1 S2 S3 Annual profit (M$/year)  923  864  789  923  889  850  FEDI  592  577  570  ---TDI  ---959  950  946 For-profit and TDI scenario (Figure 8b), point S1 presents the solution with the highest annual profit of 923 M USD/year and TDI score of 959. In this solution, maximum of 100 ton/h of EFB is required and mass allocation of sugars to levulinic acid and succinic acid are 10% and 90%, respectively. In this regard, succinic acid has major allocation because of higher global demand. TDI score of succinic acid process is higher than that of levulinic acid process due to high operating conditions in the purification section and succinic acid is categorized as a toxic/corrosive chemical. However, sustainability assessment of levulinic acid in terms of annual profit and TDI shows better result than succinic acid. Thus, a small percentage (10%) of cellulose goes to levulinic acid. Furthermore, 70% of lignin is allocated to guaiacol and the rest to vanillin. Like succinic acid, guaiacol has higher global demand but lower annual profit and TDI score. Hence, vanillin has gained 30% of lignin allocation due to better economics and safety in terms of TDI. Figure 11 shows the selected biorefinery configuration (S1 in Figure 8b) to produce 18.8 ton/h of xylitol, 1.78 ton/h of levulinic acid, 22 ton/h of succinic acid, 5.16 ton/h of guaiacol, and 0.49 ton/h of vanillin. It can fulfill the global demand of xylitol by 56%, levulinic acid by 95%, succinic acid by 25%, guaiacol by 92%, and vanillin by 11%.
Point S2 in Figure 8b is an intermediate solution, where the total profit and TDI are 889 M USD/year and 950, respectively. The same trend of materials allocation (as for point S1) is seen here. Point S3 in Figure 8b gives the lowest of both profit and TDI of 850 M USD/year and 946, respectively; it also has the same mass allocation of materials. Figure 12 shows the configuration for point S3, where 97 ton/h of EFB is supplied to the biorefinery producing xylitol (18 ton/h), levulinic acid (1.11 ton/h), succinic acid (22 ton/h), guaiacol (5 ton/h), and vanillin (0.45 ton/h). It fulfills the global demand of xylitol by 54%, levulinic acid by 59%, succinic acid by 25%, guaiacol by 89%, and vanillin by 10%. It is interesting to note that all solutions in this case are similar to those in Case 1 since levulinic acid and vanillin have better performance in both economy and TDI.

S1
S2 S3 S1 S2 S3 For-profit and TDI scenario (Figure 8b), point S1 presents the solution with the highest annual profit of 923 M USD/year and TDI score of 959. In this solution, maximum of 100 ton/h of EFB is required and mass allocation of sugars to levulinic acid and succinic acid are 10% and 90%, respectively. In this regard, succinic acid has major allocation because of higher global demand. TDI score of succinic acid process is higher than that of levulinic acid process due to high operating conditions in the purification section and succinic acid is categorized as a toxic/corrosive chemical. However, sustainability assessment of levulinic acid in terms of annual profit and TDI shows better result than succinic acid. Thus, a small percentage (10%) of cellulose goes to levulinic acid. Furthermore, 70% of lignin is allocated to guaiacol and the rest to vanillin. Like succinic acid, guaiacol has higher global demand but lower annual profit and TDI score. Hence, vanillin has gained 30% of lignin allocation due to better economics and safety in terms of TDI. Figure 11 shows the selected biorefinery configuration (S1 in Figure 8b) to produce 18.8 ton/h of xylitol, 1.78 ton/h of levulinic acid, 22 ton/h of succinic acid, 5.16 ton/h of guaiacol, and 0.49 ton/h of vanillin. It can fulfill the global demand of xylitol by 56%, levulinic acid by 95%, succinic acid by 25%, guaiacol by 92%, and vanillin by 11%.
Point S2 in Figure 8b is an intermediate solution, where the total profit and TDI are 889 M USD/year and 950, respectively. The same trend of materials allocation (as for point S1) is seen here. Point S3 in Figure 8b gives the lowest of both profit and TDI of 850 M USD/year and 946, respectively; it also has the same mass allocation of materials. Figure 12 shows the configuration for point S3, where 97 ton/h of EFB is supplied to the biorefinery producing xylitol (18 ton/h), levulinic acid (1.11 ton/h), succinic acid (22 ton/h), guaiacol (5 ton/h), and vanillin (0.45 ton/h). It fulfills the global demand of xylitol by 54%, levulinic acid by 59%, succinic acid by 25%, guaiacol by 89%, and vanillin by 10%. It is interesting to note that all solutions in this case are similar to those in Case 1 since levulinic acid and vanillin have better performance in both economy and TDI.  Figure 11. The optimal biorefinery configuration of the optimal point S1 for profit and toxicity damage index in Figure 8b. Figure 11. The optimal biorefinery configuration of the optimal point S1 for profit and toxicity damage index in Figure 8b.

Case 3: Maximize Profit, Minimize Environmental Impact, and Minimize Hazard Potential
From the Pareto-optimal solutions for Case 3 (Figure 13), one preferred solution is selected through the combination of FAHP and TOPSIS. The subjective weight of different evaluation criteria for this study was from Yan [75], where five experts were asked to give the weight of each criterion. Each weight was compiled to make the fuzzy pairwise comparison. Then, fuzzy geometric mean was used to calculate the weights/preferences of decision-makers for each criterion as tabulated in Table  11. The weights obtained for each criterion were finally 0.5 for the economy (C1), 0.2 for the environment (C2), and 0.3 for safety (C3), as listed in Table 12. It can be seen that the economic criterion would be the main factor in the decision making compared to the other two. It is interesting to note that safety has relatively more weightage than environment. The weights (Table 12) determined using the fuzzy-AHP method were then used for ranking the alternative solutions by TOPSIS. Calculations of TOPSIS are summarized in Tables S7 and S8 in the Supplementary Materials.

Case 3: Maximize Profit, Minimize Environmental Impact, and Minimize Hazard Potential
From the Pareto-optimal solutions for Case 3 (Figure 13), one preferred solution is selected through the combination of FAHP and TOPSIS. The subjective weight of different evaluation criteria for this study was from Yan [75], where five experts were asked to give the weight of each criterion. Each weight was compiled to make the fuzzy pairwise comparison. Then, fuzzy geometric mean was used to calculate the weights/preferences of decision-makers for each criterion as tabulated in Table 11. The weights obtained for each criterion were finally 0.5 for the economy (C1), 0.2 for the environment (C2), and 0.3 for safety (C3), as listed in Table 12. It can be seen that the economic criterion would be the main factor in the decision making compared to the other two. It is interesting to note that safety has relatively more weightage than environment. The weights (Table 12) determined using the fuzzy-AHP method were then used for ranking the alternative solutions by TOPSIS. Calculations of TOPSIS are summarized in Tables S7 and S8 in the Supplementary Materials.   Figure 13a shows the Pareto-optimal solutions for annual profit, GWP, and GEDI, whereas Figure 13b shows them for annual profit, GWP, and TDI. The optimal values for the chosen solution (shown by filled circle in both these plots) by FAHP and TOPSIS are summarized in Tables 13 and 14. Since the main factor is economy, the preferred solution out of many Pareto-optimal solutions is the one with the highest profit. The corresponding optimal biorefinery configuration is depicted in Figure 14.
For the chosen optimal configuration, annual profit, GWP, FEDI, and TDI are 932 M$/year, 284 tonCO 2 -eq, 595, and 957, respectively. This configuration is at the maximum capacity of 100 ton/h of EFB and the mass allocation of glucose is 10% to levulinic acid and the remaining to succinic acid, whereas 70% of lignin goes to guaiacol and the rest to vanillin. The xylitol production is 18.9 ton/h, which fulfills the global demand by 55%. The production of 1.83 ton/h of levulinic acid fulfills 98% of global demand, 22 ton/h of succinic acid fulfills 25% of demand, 5 ton/h of guaiacol fulfills~90% of demand and 0.55 ton/h of vanillin fulfills~12% of demand. As expected, maximum profit corresponds to maximum EFB that can be supplied to the plant.  Figure 13a shows the Pareto-optimal solutions for annual profit, GWP, and GEDI, whereas Figure 13b shows them for annual profit, GWP, and TDI. The optimal values for the chosen solution (shown by filled circle in both these plots) by FAHP and TOPSIS are summarized in Tables 13 and  14. Since the main factor is economy, the preferred solution out of many Pareto-optimal solutions is the one with the highest profit. The corresponding optimal biorefinery configuration is depicted in Figure 14. For the chosen optimal configuration, annual profit, GWP, FEDI, and TDI are 932 M$/year, 284 tonCO2-eq, 595, and 957, respectively. This configuration is at the maximum capacity of 100 ton/h of EFB and the mass allocation of glucose is 10% to levulinic acid and the remaining to succinic acid, whereas 70% of lignin goes to guaiacol and the rest to vanillin. The xylitol production is 18.9 ton/h, which fulfills the global demand by 55%. The production of 1.83 ton/h of levulinic acid fulfills 98% of global demand, 22 ton/h of succinic acid fulfills 25% of demand, 5 ton/h of guaiacol fulfills ~90% of demand and 0.55 ton/h of vanillin fulfills ~12% of demand. As expected, maximum profit corresponds to maximum EFB that can be supplied to the plant.  Figure 13.

Objective
Profit, GWP and FEDI Profit, GWP and TDI Profit (M$/year) 932 931 Figure 13. Pareto-optimal fronts for (a) profit, global warming potential, and fire explosion damage index and (b) profit, global warming potential, and toxicity damage index. Table 13. Decision variables that correspond to the selected Pareto-optimal solutions in Figure 13.  Figure 14. The optimal biorefinery configuration for simultaneously maximizing annual profit and minimizing global warming potential, fire explosion damage index, and toxicity damage index.

Conclusions
This work proposed an integrated biorefinery concept to produce value-added products (namely, xylitol, levulinic acid, succinic acid, guaiacol, and vanillin) from EFB. It formulated three cases of MOO: one for maximization of profit and minimization of GWP, another for maximization of profit and minimization of both FEDI and TDI, and the last one considering all objectives simultaneously, and solved them using surrogate models and a multi-objective genetic algorithm. The resulting Pareto-optimal solutions show the trade-off among the economic, environmental, and safety performances. One of these optimal solutions was selected using FAHP and TOPSIS. The preferred optimal solution revealed that the supply of EFB was at 100 ton/h, i.e., at the maximum supply of EFB possible to the plant. The optimal mass allocation of glucose was 10% to levulinic acid and the remaining to succinic acid, whereas 70% of lignin went to guaiacol and the rest to vanillin. The higher global demand of succinic acid led to more glucose allocation to it, while 10% for levulinic acid was for higher profit and lower GWP and TDI. Similarly, 70% of lignin went to producing more guaiacol than vanillin; 30% of lignin was used for vanillin production due to its better economics and lower environmental impact and TDI. All of xylose was allocated to produce xylitol since it had no other option. Succinic acid (22 ton/h), xylitol (18.8 ton/h), and guaiacol (5 ton/h) were the mostproduced chemicals due to their higher global demand; others are levulinic acid (1.83 ton/h) and vanillin (0.55 ton/h). The optimal integrated biorefinery has an annual profit of $932 M USD/year, GWP of 284 tonCO2-eq, FEDI of 595, and TDI score of 957. Further work on the utilization and optimization of other palm oil biomass to produce marketable bioenergy and biochemicals should be pursued to enhance the sustainability of palm oil industry. Also, cumulative energy cost should be included in MOO to enhance the sustainability of the integrated biorefinery.
Supplementary Materials: The following are available online at www.mdpi.com/xxx/s1, Figure S1: total global warming potential, Table S1: the estimation of capital investment, Table S2: the estimation of operating costs, Table S3: raw material and product selling price, Table S4: economic results for 75 ton/hour of dry empty fruit bunch, Table S5: inherent safety results summary, Table S6: decision maker attribute, Table S7: TOPSIS key Figure 14. The optimal biorefinery configuration for simultaneously maximizing annual profit and minimizing global warming potential, fire explosion damage index, and toxicity damage index.

Conclusions
This work proposed an integrated biorefinery concept to produce value-added products (namely, xylitol, levulinic acid, succinic acid, guaiacol, and vanillin) from EFB. It formulated three cases of MOO: one for maximization of profit and minimization of GWP, another for maximization of profit and minimization of both FEDI and TDI, and the last one considering all objectives simultaneously, and solved them using surrogate models and a multi-objective genetic algorithm. The resulting Pareto-optimal solutions show the trade-off among the economic, environmental, and safety performances. One of these optimal solutions was selected using FAHP and TOPSIS. The preferred optimal solution revealed that the supply of EFB was at 100 ton/h, i.e., at the maximum supply of EFB possible to the plant. The optimal mass allocation of glucose was 10% to levulinic acid and the remaining to succinic acid, whereas 70% of lignin went to guaiacol and the rest to vanillin. The higher global demand of succinic acid led to more glucose allocation to it, while 10% for levulinic acid was for higher profit and lower GWP and TDI. Similarly, 70% of lignin went to producing more guaiacol than vanillin; 30% of lignin was used for vanillin production due to its better economics and lower environmental impact and TDI. All of xylose was allocated to produce xylitol since it had no other option. Succinic acid (22 ton/h), xylitol (18.8 ton/h), and guaiacol (5 ton/h) were the most-produced chemicals due to their higher global demand; others are levulinic acid (1.83 ton/h) and vanillin (0.55 ton/h). The optimal integrated biorefinery has an annual profit of $932 M USD/year, GWP of 284 tonCO 2 -eq, FEDI of 595, and TDI score of 957. Further work on the utilization and optimization of other palm oil biomass to produce marketable bioenergy and biochemicals should be pursued to enhance the sustainability of palm oil industry. Also, cumulative energy cost should be included in MOO to enhance the sustainability of the integrated biorefinery.
Supplementary Materials: The following are available online at http://www.mdpi.com/2227-9717/8/7/868/s1, Figure S1: total global warming potential, Table S1: the estimation of capital investment, Table S2: the estimation of operating costs, Table S3: raw material and product selling price, Table S4: economic results for 75 ton/hour of dry empty fruit bunch, Table S5: inherent safety results summary, Table S6: decision maker attribute, Table S7: TOPSIS key calculation information for profit, global warming potential, and fire explosion damage index, Table S8: TOPSIS key calculation information for profit, global warming potential, and toxicity damage index.