Skip to main content
NCBI home page
ACS Omega logoLink to ACS Omega
. 2024 Mar 6;9(11):12941–12955. doi: 10.1021/acsomega.3c09186

Optimization and Deep Learning Modeling of the Yield and Properties of Baobab-Derived Biodiesel Catalyzed by Waste Banana Bunch Stalk Biochar

Collins Chimezie Elendu †, Chang Liu †, Rao Danish Aleem †, Yaqi Shan †, Changqing Cao †, Naveed Ramzan ‡,*, Pei-Gao Duan †,*
PMCID: PMC10955699  PMID: 38524430

Abstract

graphic file with name ao3c09186_0012.jpg

The integration of optimization techniques and deep learning models, which offer a promising avenue for improving the efficiency and sustainability of biodiesel production processes from baobab seed oil (BSO), is rare. This study utilized a multi-input-multioutput (MIMO) deep learning technique and the most recent central composite design (CCD) optimization tool to model and optimize the yield and properties of biodiesel produced from BSO. First, the baobab seed oil was extracted using a solvent extraction method. BSO was subsequently analyzed and converted to biodiesel by reacting CH3OH catalyzed by waste banana bunch stalk biochar activated by KOH. Multiobjective optimization and prediction of the biodiesel yield (Y) and several key fuel properties, including the cetane number (CN), kinematic viscosity (VS), and purity (P), were achieved. With better correlation coefficients of 0.9709, 0.9464, and 0.9714 for response training, response testing, and response validation, respectively, and a root-mean-square error of 0.00755, the MIMO model on the logsig transfer function accurately predicted the biodiesel yield and properties more than did the MISO and response surface methodology models. The optimum Y (96 wt %), CN (48), VS (3.3 mm2/s), and P (98.3%) were concurrently accomplished at a reaction temperature of 56 °C, a reaction time of 115 min, a CH3OH/BSO molar ratio of 15:1, a catalyst dosage of 6 wt %, and a stirring speed of 400 rpm with 98% optimal validation accuracy. CCD sensitivity analysis revealed that the CH3OH/BSO ratio was the most sensitive (50.9%) input predictor among the other input variables studied.

1. Introduction

In recent years, concerns about the environmental effects of petroleum diesel and energy issues related to population growth and ongoing exploration have intensified.1 This has led to extensive research on developing biofuels as substitute fuels to meet increasing energy demands and prevent an impending energy crisis. However, the growth of biofuel production has not met expectations, with an average increase of 5% between 2010 and 2019, compared to the needed growth rate of 14% per year until 2030.2,3 The International Energy Agency (IEA) has recognized bioenergy as a viable option in its roadmap for achieving net-zero emissions by 2050, suggesting that biofuels could account for 18% of all of the energy supply by that time.4 To achieve carbon neutrality in the global energy system, biofuels must be used either directly to replace fossil fuels or indirectly to offset emissions through the use of bioenergy.5 Given the limited availability of low-carbon options in the transportation sector, particularly for trucking, shipping, and aviation, biofuels are expected to play a larger role in the economy, transportation, and heat and power generation in the coming years.6

The use of biodiesel as an excellent replacement for fossil fuels has spread worldwide.7 This is because it has many benefits over fossil fuels, such as being renewable, biodegradable, and nontoxic, having a high flash point and emitting fewer greenhouse gases.8 Fatty acid methyl esters (FAME), often known as biodiesel, can be obtained from various sources, both consumable and inedible oils.9 Pyrolysis, transesterification, microemulsification with alcohols, and catalytic cracking are only a few of the methods used to create FAMEs from varieties of triglycerides.5 One of the easiest, least constrained, most productive, and crucial procedures for creating a more commercially promising and environmentally acceptable FAME fuel is transesterification.10 Transesterification is a type of reaction that transforms triglycerides present in feedstocks into FAME that may be used as biofuel.11 Because of its significantly lower viscosity, FAME synthesized by the transesterification process is appropriate for use in diesel engines.12 This resource faces difficulties despite its value. These challenges include expensive FAME feedstocks, their scarcity, low oil yield, poor quality, and poor storage; all of these factors significantly impact FAME costs.13 However, among all of these bottlenecks, one of the most crucial factors is the price of feedstocks. According to previous studies, feedstock expenses account for more than 76% of the overall cost of synthesizing FAME fuel.4 Therefore, selecting a low-cost and uncompetitive feedstock is essential for reducing the cost of FAME.

The majority of research on biofuels has been devoted to creating FAMEs using edible vegetable oils as feedstock. The following edible oils have been transesterified to create FAME: corn oil,14,15 groundnut oil,16 palm oil,17,18 soybean oil,13,19 vegetable oil,20 sunflower oil,21 rapeseed oil,22 and cottonseed oil,23 as a few examples utilizing homogeneous or heterogeneous catalysts. However, questions have been expressed about the prolonged usage of edible oils to create FAMEs, thereby raising the price of edible food oil and making FAMEs produced from it relatively expensive; this strategy also reduces the amount of food readily available.

To solve this issue, nonedible oils, such as those produced from nonedible seed crops, must be used as feedstocks for the synthesis of FAMEs. To meet projected energy needs in the coming years, it is crucial to take advantage of the enormous potential of energy realization via sustainable nonedible feedstocks such as baobab seed oil (BSO) and biomass waste (banana bunch stalk) in accordance with the IEA’s outlook on bioenergy security, climate change control and mitigation, and energy independence. As a result, significant research is being conducted to discover and assess biomass waste and locally accessible sustainable nonedible oils as suitable feedstocks and catalysts for FAME synthesis. Tobacco seeds, Chinese tallow, waste vegetable oil, rubber seeds, desert dates, jatropha oil, tobacco oil, castor, and other inedible feedstocks have been studied for FAME synthesis.5,24,25 Nevertheless, these feedstocks are limited by the unavailability of seed oil in commercial quantities and the low oil content of the feedstock available for commercialization. However, baobab seeds, which have a relatively high oil content of approximately 30–40% and are widely available ref (26) are an underexploited nonedible feedstock for massive or pilot manufacturing of FAMEs.

The baobab tree (Adansonia digitata L.) is a deciduous plant that can grow in many regions of West and Southern Africa, as well as certain parts of Asia.26 It is a member of the Bombacoideae subfamily of seed crops. The baobab tree grows in hot semiarid areas with little or no rainfall. It is a nontimber forest product (NTFP). Baobab seeds, which have a relatively good nonedible oil content of approximately 30–40% and a good content of saturated and unsaturated fatty acids, have rarely been employed in FAME production and optimization studies.

Banana is the second most widely grown tropical fruit in the world, and as such, a considerable amount of its waste byproducts, such as banana peels, banana trunks, and banana bunch stalks, are generated.27 Banana peels, which are nonedible biomass waste products, have been put to use in a number of ways, including the creation of biosorbents, bioethanol, and other energy-related processes.27−29 The banana bunch stalk was chosen for use in this study as a raw material for the synthesis of nanocatalysts for the production of FAME due to the abundance of nutrients and minerals, including the metal cations calcium (Ca), sodium (Na), potassium (K), silicon (Si), magnesium (Mg), and other elements that enhance catalysis. Additionally, this raw waste can be thermally and chemically activated to improve its porosity and alter its structure to drive transesterification reactions.

Most biodiesel research has traditionally focused on optimizing the FAME yield, neglecting other important fuel properties, such as cetane number, viscosity, and purity.30−33 This single objective approach may compromise the product quality and hinder the effective operation of diesel engines. To address this limitation, researchers are now shifting toward a multiobjective optimization approach that considers both the maximum yield and the optimum product quality. This study investigates the less commonly explored multiobjective optimization problem for the FAME production process, using inedible oil as a feedstock. By considering multiple attributes simultaneously, this approach aims to overcome the limitations of solely focusing on yield and ensuring the production of a high-quality FAME that meets ASTM requirements and can efficiently power internal combustion engines.

The complexity and nonlinear nature of the transesterification reaction in FAME production pose challenges in data processing and forecasting. To address these challenges, researchers are exploring data-based technologies, such as machine learning (ML), which involves creating, analyzing, and using techniques to enable machines to learn and perform AI-related tasks through a learning process using response data. Machine learning is considered a promising approach for resolving issues related to FAME synthesis and forecasting by leveraging its ability to handle complex relationships between input and response variables.34,35 ML involves a learning process that helps a system adapt to its environment and observations. Deep learning neural networks, a subset of machine learning, simulate the functioning of the human brain by using data inputs, weights, and biases. They enable systems to process complex data and make predictions or decisions based on patterns and relationships.36 The use of deep learning neural networks (ANNs) has been applied to model various biofuel products. In some studies,3,37,38 deep learning has been used to predict the yield of methyl esters using different catalysts. However, the focus was solely on yield prediction, disregarding the fuel properties. Previous studies have explored the modeling and prediction of fuel properties in FAME production using various feedstocks such as fried soybean oil and waste goat tallow. These investigations have employed statistical analysis and artificial neural networks (ANNs).39,40 However, the integration of optimization techniques and deep learning models, which could significantly enhance the efficiency and sustainability of FAME production processes from BSO, is currently lacking. Traditional experimental determination of CN, VS, and P of FAMEs can be expensive and time-consuming, especially when multiple process variables are involved. To overcome these challenges and ensure accurate predictions and high-quality products, the use of deep machine learning techniques, such as ANNs, is necessary. Therefore, this study focused on the optimization and deep learning prediction of fuel yield and properties of FAME derived from BSO (inedible seed oil) via KOH-activated waste banana bunch stalk biochar.

2. Materials and Methods

2.1. Materials

Nonedible seeds (Baobab seeds) and biomass waste (banana bunch stalks) were obtained from the Shaanxi Key Laboratory of Energy Chemical Process Intensification, Xi’an Jiaotong University, China. Laboratory-grade CH3OH (97 wt %), n-C6H14 (99 wt %), C2H5OH (97.4 wt %), NaOH (97 wt %), KI (99 wt %), CHCl3 (99.5 wt %), HCl (37.5 wt %), H2SO4 (98 wt %), iodine acetic acid, and starch were utilized for the characterization of the feedstock following the ASTM, EN, and AOAC methods. Figure 1 shows the nonedible baobab seeds and the extracted oil.

Figure 1.

Figure 1

Open in a new tab

Picture of nonedible feedstock (baobab seed and extracted oil).

2.2. Methods

2.2.1. Oil Extraction

The methods employed by Ude et al.41 were used in the extraction of oil from the baobab seeds. The seeds were first dehusked, and to thoroughly remove moisture, the sample was dried at 105 °C for 12 h. The sample was powdered after drying. A Soxhlet extractor’s porous thimble was filled with 100 g of powdered seeds, and the oil was extracted using n-hexane at 69 °C for 3 h. The oil and solvent were separated by using a rotary evaporator, and the oil was dried at 60 °C for 1 h to remove the remaining solvent. Equation 1 was used to determine the oil yield after extraction on a dry basis

2.2.1. 1

2.2.2. Catalyst Preparation

Using a method devised by Jitjamnong et al.,42 activated banana bunch stalks served as waste-derived green catalysts in this study. The banana stalks were first sun-dried, cut into small pieces, and cleaned with distilled H2O to remove dirt and other impurities. The plants were subsequently ground into powder and subsequently sieved through a mesh (≤250 μm) to create a fine powder after being dried overnight in a 100 °C oven; this powder was denoted as UBB. Then, at a heating rate of 10 °C/min, these tiny particles were calcined at 700 °C and kept there for 4 h to create banana bunch stalk biochar. The resultant biochar was agitated for 2 h in a 2 M KOH solution; this process is known as chemical activation. After filtering, distilled water was used to rinse the product many times to achieve a pH of 7, and the product was allowed to dry for 24 h at 105 °C. The resulting catalyst is known as the BBB.

2.2.3. Characterization of the Synthesized Catalyst

The physiochemical characteristics and elemental content or composition of the catalysts were determined. Scanning electron microscopy (SEM-Carl Zeiss Sigma Field Emission) analysis was performed to record the microstructural morphology of the catalyst, and energy-dispersive spectroscopy (EDS) was performed to determine the elemental composition. SEM–EDS is the name of this strategy. The functional groups of the catalyst were identified using Fourier transform infrared (FTIR) spectroscopy (M530 Buck Scientific). Using an X-ray diffractometer (model XRD-7000), we identified the crystalline phase composition of the catalyst.

The evaluation of a catalyst’s basicity is vital in the production of FAME because it directly affects both the yield and quality of FAME. Optimal basicity enhances transesterification reactions, minimizing unwanted byproducts and allowing for the selection of the most suitable catalysts. In this study, the Hammett indicator method was employed to determine the strength of the base. Titration using an anhydrous methanol solution (0.02 M) and Hammett indicator-benzene carboxylic acid enabled the assessment of the overall basicity of the catalysts.

2.2.4. Esterification of the Extracted BSO

Since the determined acid content was greater than the suitable range, the extracted nonedible oil (BSO) was subjected to an esterification reaction. To achieve this, concentrated sulfuric acid was used as a catalyst. First, to remove any leftover H2O, for 10 min, the oil was heated to 110 °C. Subsequently, a flat-bottom flask was filled with 60 mL of methanol and 100 mL of dried extracted oil, along with an established quantity of 2 g of H2SO4 (98 wt %). On the bottom of the flask, a magnetic stirrer was mounted and rotated at 450 rpm. After 1 h of reaction, after heating to 60 °C, the mixture of the reaction products was put into a separatory funnel and allowed to cool and separate. Two unique layers were formed from the mixture: the treated oil composed the upper layer, while the water, unreacted methanol, and contaminants composed the lower layer. The undesired waste layer was removed from the bottom layer. In preparation for its use in the transesterification procedure, the extracted oil that had undergone pretreatments was relocated into a beaker and gently dried at 105 °C until all of the remaining H2O was eliminated.

2.2.5. Transesterification of the Extracted Prepared BSO

The process described by Chimezie et al.4 was used to manufacture FAMEs. A three-neck round-bottom flask with a stir bar was filled with 100 mL of prepared BSO, 42 mL of CH3OH, and 4 g of the waste-derived catalyst. A reflux condenser was attached to one neck, and a thermometer was positioned below the liquid level to connect to the other. Then, 55 °C was applied to the flask, which was aggressively stirred for 60 min at a speed of 400 rpm. A separatory funnel was used to hold the reaction mixture when the reaction time had ended, and the mixture was allowed to sit for 10 min. The separating funnel was utilized to contain the solution, which was left undisturbed for a while after the stirrer was removed. The collected catalyst was obtained through filtration. After eliminating the lower phase, the upper FAME phase was collected and cleaned with hot distilled water at 60 °C to remove contaminants. After that, the FAME mixture was dried on a mantle and heated to 110 °C. Using factors that affect the transesterification reaction, such as the temperature, agitation speed, catalyst loading, reaction time, and CH3OH/BSO ratio, the method was repeated as intended and is presented in Table 1.

Table 1. FAME Production Experimental Range and Levels of Independent Process Variables.
factor (s) units –1 +1 –α +α 0 level
catalyst concentration (A) wt % 4 6 3 7 5
methanol oil molar ratio (B) mol/mol 9 15 6 18 12
temperature (C) °C 55 65 50 70 60
reaction time (D) min 60 120 30 150 90
agitation speed (E) Rpm 200 400 100 500 300
Open in a new tab

The methodology employed by Khethiwe et al.43 involved assessing the purity of the FAMEs using a GC–MS (Agilent 7820A, Agilent Technologies, Santa Clara, CA) apparatus fitted with an HP-5 capillary column (30 m × 0.25 μm × 0.25 μm) and a flame ionization detector (FID). Methyl oleate (reference FAME) was used as an internal standard. Equations 2 and 3 were used to determine the purity of FAME based on the weight of FAME utilized and the yield of FAME.

2.2.5. 2
2.2.5. 3

mF = mass of FAMEs synthesized (g), mO = mass of BSO (g)

2.3. Experimental Design and Optimization

First, the experiment was designed by using the most recent version of Stat-Ease software (CCD). A five-level, five-factor fractional factorial design was used in this study. The following are the process-applied parameters: catalyst loading (A), CH3OH/BSO molar ratio (B), temperature (C), reaction duration (D), and stirring speed (E); the responses chosen were methyl ester yields (Y), kinematic viscosity (VS), cetane number (CN), and purity (P). The biodiesel production process was optimized to obtain a biofuel with good fuel quality under optimal conditions using the multiobjective tool in Stat-Ease software. Several important process variables, including A–E, affected Y, CN, VS, and P in the transesterification process. Therefore, optimizing these parameters is crucial for achieving both maximum production and enhanced product quality; as such, product commercialization can be feasible and profitable. The experimental design matrix utilized in this study is listed in Table 1.

2.4. Reusability of the Catalyst

Before conducting the catalyst reusability test, the catalyst was retrieved through centrifugation, washed with methanol, and dried in an oven at 100 °C for 5 h to eliminate any moisture or impurities. The reusability test was carried out using the optimized reaction conditions, which included a reaction temperature of 56 °C, a reaction time of 115 min, a CH3OH/BSO molar ratio of 15:1, a catalyst dosage of 6 wt %, and a stirring speed of 400 rpm.

2.5. Determination of VS and CN

The VS of the samples was determined using a Glass PSL Viscometer No. 65526, which was measured using reference viscometers stored in the laboratory and can be traced to the ASTM Test Method D445.44 At 40 °C, it was found that the constant C equals 0.4916. A fixed volume (50 mL) of the sample’s gravity flow time was calculated. Equation 4 was used to determine the kinematic viscosity V (mm2/s), which was derived based on an average measured flow time (seconds) during the experiment. The CN was determined according to ASTM D613-10a.45

2.5. 4

2.6. Modeling of the Yield and Properties of FAME

An ANN with a feed-forward neural network architecture was used to estimate the Y, VS, CN, and P of FAMEs produced from baobab seed oil through waste catalysis. A, B, C, D, and P are the multi-input variables, and the desired outputs are the Y, VS, CN, and P of the FAME. The neural network architectures of MIMO systems are shown in Figure 2. Three layers make up the network architecture: the layer at the input, the hidden layer, and the layer at the output. A nonlinear transfer function exists between the hidden layer and the output layer. The variables were selected to track how the input and output variables interacted with one another and to shorten the duration of the experimental procedure. To determine the ability of the multi-input-multioutput MIMO model to forecast the FAME yield, FAME vs CN, and P, from the responses, 128 different data sets were chosen. The logsig and tansig nonlinear transfer functions were used in the hidden layer to process the data, and the purelin function was used in the output layer. Supervised learning was utilized, and the network was trained with 70% of the data sets and tested and validated with 15 and 15%, respectively, of the data sets. To ensure that the training iterations are adequate and to lessen the likelihood of becoming stuck in local minima, the largest epoch chosen was 1000. Furthermore, for each iteration, the input data sets were randomly chosen. The transfer function and number of neurons in the hidden layer were varied to achieve the best network architecture while avoiding overfitting or overtraining. Equation 5 shows the results of the ANN model. The MSE in eq 6 was used to gauge the performance of the model. In terms of prediction accuracy, the neural network design with the lowest MSE is the superior model.

2.6. 5
2.6. 6

where n is the number of data sets, f = threshold function, d = network dimension, l = available layer number, and wlij = network layer mass l with i input and hidden layer j.

Figure 2.

Figure 2

Open in a new tab

Modeling of the FAME yield, kinematic viscosity, cetane number, and purity with ANN architecture using the MIMO model.

3. Results and Discussion

3.1. Oil Yield and Characteristics

The actual oil yield achieved during oil extraction is one of the most important considerations in evaluating the economic viability of any FAME feedstock. Using eq 1, 36.8 wt % of the oil from the Baobab seeds was extracted. The oil yield obtained was considerably greater than the yields observed from other nonedible seed oils, such as cotton seeds (17–23 wt %) and soapnut seeds (23–30 wt %), reported by Chimezie et al.5 and was within the range (30–40 wt %) of yield reported by Dass et al.26 According to Silva et al.,19 the observed oil production of baobab seeds was also better when the yield of various food oils was increased, including that of soybeans (18.5–22.5 wt %) and cotton seeds (15–20 wt %). A comparatively good oil content of Baobab will lead a long way from being able to encourage less reliance on edible oils as feedstocks for FAME synthesis and, as a result, encourage the commercialization of the products. Therefore, the Baobab seed is extremely suitable and cost-effective for industrial applications involving the production of biodiesel on the basis of its high oil content, as any seed that bears oil and can yield up to 30% oil is recognized as a viable feedstock for large-scale FAME industrial production.4

Tables 2(a) and 2(b) give an overview of the physiochemical properties of BSO derived from its characterization. First, the sample-specific gravity, which is the density relative to the volume of water of the same density, was determined. Owing to its connection to cetane rating, heating values, feedstock storage, and transportation, specific gravity has been referred to as one of the most fundamental and important physiochemical properties of feedstock for the manufacture of biodiesel.46 Triglycerides typically have a specific gravity between 0.87 and 0.98, which is within the range of the reported specific gravity of BSO (0.928). Additionally, the specific gravity of BSO is reportedly beneficial because it can result in improved separation and fewer obstacles during purification with regard to reducing emulsion formation during FAME synthesis.4 The acid content of the oil extracted from the baobab seeds was determined to be 6.6 mg of KOH/g of oil, indicating a relatively low value within the observed range. However, it is still recommended and safer to esterify the oil before transesterification when using biomass-derived biochar activated with KOH as a solid catalyst.5 A saponification value of 170.2 mg of KOH/g of oil (obtained from BSO) indicates that the oil has a significant proportion of long-chain fatty acids. This implies that the oil will require less catalyst for transesterification, which can help lower costs and improve yields while maintaining higher oxidative stability. This also suggested that the FAME produced from these materials will have a longer shelf life and be more resistant to oxidation over time.47 The physicochemical characteristics of the raw oil were more favorable than those of a few other oils, such as Pongamia pinnata, Jatropha curcas, Madhuca indica, and neem seed used for biodiesel.48,49 BSO has a high viscosity that prevents it from being atomized in an internal combustion engine. As a result, these materials cannot be used as biofuels immediately and must undergo transesterification to decrease their value to within the acceptable range of ASTM standards. The oil may be stored for a long time because of the low pour point, which indicates that it will scarcely solidify at room temperature. The amount of iodine in the oil can be utilized to ascertain its level of unsaturation.50 The extracted oil sample has an iodine concentration of approximately 67.330 g/iodine/100 g, which is greater than the 60 g/100 g benchmark, which means that the oil can promote FAME synthesis without causing any negative side effects or undesirable reactions as a result of its high unsaturation characteristics, as suggested by this value.

Table 2(a). List of Physiochemical Qualities of the Baobab Seed Oil (BSO).

properties BSO
specific gravity (30 °C) 0.928
acid value (A.V) (mg KOH/g oil) 6.590
free fatty acid (mg KOH/g) 3.295
saponification value (mg KOH/g oil) 170.200
iodine value (g/iodine/100 g) 67.330
kinematic viscosity at 40 °C (mm2/s) 43.200
peroxide value (Meq/kg) 5.512
pour point (°C) 7.600
moisture content (ppm) 50 ± 2
refractive index (at 25 °C) 1.466
density (g/mL) 0.928
Open in a new tab

Table 2(b). Fatty Acid Composition of BSO.

fatty acid structure percentage area (%) molecular weight (g/mol)
stearic acid C18:0 10.03 284.48
palmitic acid C16:0 7.30 256.40
oleic acid C18:1 46.21 304.47
linoleic acid C18:2 22.96 294.48
gadoleic acid C20:1 5.30 310.51
arachidic acid C20:0 1.38 312.53
Open in a new tab

Table 2(b) indicates that BSO was composed of six primary fatty acids: 83.49% unsaturated acids (oleic, linoleic, and gadoleic acids) and 9.712% saturated acids (stearic, palmitic, and arachidic acids). Oleic acid, which made up 46.21% of the oil’s total fatty acid composition, was the predominant unsaturated fatty acid. As a result, the oil falls into the monounsaturated fatty acid category and is an excellent source of fuel for the synthesis of FAMEs.

3.2. Characterization of the KOH-Assisted Banana Bunch Stalk Catalyst

The morphology and elemental makeup of the raw banana bunch stalks (UBBs) and calcined/activated banana bunch stalks (BBBs) were determined by using SEM–EDS analysis, as shown in Figure 3a–d. The samples (from the UBB and BBB) clearly contained potassium (K), silica (Si), magnesium (Mg), calcium (Ca), zinc (Zn), and sodium (Na). The potassium concentrations (weight concentration and atomic concentration) in the UBB were 39.82 and 33.06 wt %, respectively, while those in the BBB improved to 73.26 and 72.09 wt %, respectively. This increase in K content is due to the efficiency of the calcination and activation processes, which effectively recover the minerals present in biomass waste by degrading the lignin–carbohydrate matrix. Similarly, calcination increased the composition of other minerals, such as Ca, Si, and Na, present in the BBB. In contrast to the BBB, where it was discovered to be abundantly present, minerals, such as K and Ca, were not abundant in the UBB. This lack of minerals can be attributed to the concealed effects of the aforementioned matrix. The XRD and EDS results were in agreement. The size reduction procedure used during the catalyst preparation step may be the cause of the cavities, holes, and randomly distorted rough surface visible in the UBB from the SEM image. However, after calcination at 700 °C for 4 h and activation, the porosity increased, which was evidently shown through the great presence of large perforations at the catalyst surface. Therefore, upon calcination and activation, the surface with the fewest perforations transforms into the surface with the most perforations and highest order, which further demonstrates that KOH more easily degrades the UBB carbon matrix during future activation and is more likely to create well-developed micro- and mesopores. Therefore, the uncalcined and calcined/activated waste-derived catalysts exhibited significant morphological variations, demonstrating the efficiency of calcination and activation in changing the morphology of the waste to promote catalyst synthesis.

Figure 3.

Figure 3

Open in a new tab

(a) EDS analysis of UBB; (b) EDS analysis of the BBB; (c) SEM analysis of UBB; (d) SEM analysis of the BBB; (e) FTIR analysis of the UBB and the BBB; and (f) XRD analysis of UBB and the BBB.

Figure 3e shows the UBB and BBB FTIR patterns. In the UBB, a significant absorption peak at 1040 cm–1 was noted. This peak is ascribed to the C–O–C stretching of the lignin matrix and the carbohydrates (cellulose and hemicellulose) found in the untreated waste biomass.27 Additionally, the minor peak in the UBB band at 2915 cm–1 may be explained by C–H stretching, which results in aromatic ring vibration and lignin C–H bond deformation.51 The bending vibration of water molecules in the −OH state is thought to be the cause of the slight absorption band at 3309 cm–1.52 The primary absorption bands of the lignin matrix present in the UBB, however, were nearly completely absent in the BBB, demonstrating the efficiency of calcination and activation, which broke down the complex carbohydrate–lignin matrix and created openings for catalytic action. The presence of potassium, phosphate, and silicon (Si–O–Si) stretching vibrations in the BBB may be related to the low-intensity peak at 700–960 cm–1.27 The ambient CO2 adsorption onto metal oxides found in the activated biomass catalyst can be attributed to the peak at 1173 cm–1.52 The FTIR analyses of the UBB and BBB materials demonstrated that variations in the peak intensity and shift substantiated the structural changes that occurred during the calcination and activation processes. The identification of crystalline substances in both the UBB and BBB was performed using the XRD patterns shown in Figure 3f. The reflection peaks at 2θ = 20–90° can be ascribed to the presence of osumilite, garnet, and Illite compounds, which consist of more K, Na, Ca, Fe, Mg, Al, and SiO2, respectively; these results are in line with the obtained SEM–EDS and FTIR results.

The base strength determined using Hammett indicator-benzene carboxylic acid revealed a catalyst basic strength of 15.0 < H_ < 18.5 and a total basicity of 11.5 mmol per gram (mmol/g). Based on the obtained results, it is evident that the catalyst exhibits a basicity ranging from moderate to strong, indicating its ability to effectively facilitate the transesterification reaction. Moreover, the measured total basicity, quantified at 11.5 mmol/g, signifies that the catalyst possesses a sufficient amount of basic sites per unit mass. These basic sites are expected to significantly enhance the efficiency of the transesterification process.

3.3. GC–MS Analysis of FAMEs Produced from BSO

To determine the quality of the FAMEs produced, gas chromatography–mass spectrometry (GC–MS) was used to chemically analyze the FAMEs synthesized from BSO. A summary of the obtained results indicated that methyl oleate, methyl linoleate, methyl palmitate, and methyl stearate were the most significant components, with retention times of 13.63, 15.59, 16.13, and 16.64 min, respectively, as indicated in the total ion chromatogram presented in Figure 4. Methyl palmitate, methyl linoleate, methyl oleate, and methyl stearate all dominantly showed strong peaks. The GC–MS results that were examined and presented clearly showed that triglycerides were converted to FAME, and this research also showed how effective the synthesized catalyst was.

Figure 4.

Figure 4

Open in a new tab

Total ion chromatogram of BSO FAME ((1) 9-octadecenoic acid (Z)-, methyl ester; (2) pentadecanoic acid, 14-methyl-, methyl ester; (3) 9,12-octadecadienoic acid, methyl ester; (4) hexadecanoic acid, methyl ester; (5) methyl stearate; (6) 9-octadecenoic acid methyl ester; (7) octadecanoic acid; (8) arachidic acid, methyl ester; (9) eicosanoic acid, methyl ester; (10) 9,12-octadecadienoic acid (Z,Z); (11) 9,17-octadecadienal (Z); (12) 9-octadecenoic acid, methyl ester; (13) oleic acid; (14) 9-octadecenoic acid (Z)–, 2-hydroxy-l-(hydroxymethyl)ethyl ester; (15) 9,12-octadecadienoyl chloride (Z,Z)-; (16) oleic acid; (17) 9-octadecenoic acid (Z)-, 2-hydroxy-l-(hydroxymethyl)ethyl ester; (18) 9-oxabicyclo[6.1.0]nonane).

3.4. Modeling and Prediction of Fuel Yield and Its Properties Using Deep Learning (ANN)

3.4.1. Modeling with MISO and MIMO ANN Architectures

Five input process variables and one output process variable (each variable at a time) make up the MISO design, whereas five input variables and four output layers (all variables at the same time) make up the MIMO architecture. The output variables (FAME yield, cetane number, viscosity, and fuel purity) were trained individually by utilizing the MISO model and jointly by utilizing the MIMO model. The Levenberg–Marquardt (LM) algorithm was employed to train the neural networks. The number of neurons in the hidden layer is between 1 and 30.

The MSEs of the MISO and MIMO neural network models are shown in Figures 5 and 6, respectively. The graphs show that for the tansig and logsig transfer functions, the MSE initially decreases as the number of neurons increases to 15 and 20 neurons, respectively, for MISO and up to 10 and 12 neurons, respectively, for MIMO. However, there is an increase in the MSE above these neurons, indicating that a high number of neurons above 20 neurons for MISO and 12 neurons for MIMO does not provide better forecasting ability because it is well-known that the best model design is the one with the least MSE.

Figure 5.

Figure 5

Open in a new tab

MISO plot of model performance in the hidden layer with different transfer functions: (a) tansig function and (b) logsig function.

Figure 6.

Figure 6

Open in a new tab

MIMO plot of model performance with different functions in the hidden layer.

Table 3 presents the models with the lowest mean square errors. The table unequivocally demonstrates that the models of MISO and MIMO using the logsig function have superior accuracy to those with the tansig transfer functions since these functions are present in their hidden layers. Tansig functions, however, contain fewer neurons in the best model than logsig functions, making them simpler than logsig functions; however, in all neurons trained, logsig functions show a better predictive capability than tansig functions; therefore, they can be deployed in both MISO and MIMO systems for the prediction of fuel yield and properties of FAMEs. Table 3 further demonstrates the performance characteristics of the network during training for the prediction of FAME yield and properties by utilizing the optimal hidden layer’s number of neurons (12) for the MIMO model. At epoch 117, which is the best validation performance, the network mean square error decreases to 0.000057 from a very high value. For these optimum validation performance parameters, 123 epochs at most were used to achieve the best result.

Table 3. Training Outcomes for the Top Neural Network Architecture Models.
ANN model response for prediction transfer function neuron numbers MSE
MISO Y tansig 15 0.05
logsiga 20 0.0012
CN tansig 15 0.0129
logsiga 20 0.006
VS tansig 15 0.0112
logsiga 20 0.009
P tansig 15 0.0097
logsiga 20 0.0007
MIMO Y, CN, VS, P tansig 10 0.00988
logsiga 12 0.000057
Open in a new tab
a

Represents the optimal neural network model for predicting the FAME yield and properties.

3.4.2. Prediction of Fuel Yield and Properties Using the MIMO Model

A model from the MIMO neural network was created to forecast the outputs (Y, CN, VS, and P) using a single-layer neural network architecture. Figure 7 shows the experimental versus anticipated output data from the best MIMO model. The graph clearly shows that the MIMO model, with R2 values of 0.9709, 0.9464, and 0.9714 for response training, response testing, and response validation, respectively, is indicative that the model was able to explain the variation in the data set with a high degree of precision and therefore can be applied in the prediction of these responses at any given input variable. The overall model correlation coefficient is approximately 96.4%, and the root-mean-square error is 0.00756. With this low error value observed, the model is said to have high accuracy in predicting the fuel yield and properties of any input variable that can be provided to it in the future.

Figure 7.

Figure 7

Open in a new tab

MIMO model plot analysis for response prediction.

Figure 8 shows the parity plot between the MIMO model response and the experimental response. Each point in the parity plot clearly represents the data set from the experimental model predicted against the run number. A parity plot, as used in our model, helps in model assessment to identify outliers or unexpected results that are not in line with the model values, indicating a problem that needs to be addressed; however, in our model, there were no apparent parity outliers in the yield and properties modeled; therefore, the model performance was impressive. As observed from the plot, the ANN model performed the best. The outcome matches the prediction accuracy of the real model, and the slight difference between the expected response and experimental data confirms this. Therefore, neural network models using MIMO have proven to be capable of making predictions with an equitable distribution of many input variables despite the complexities and nonlinearities between the input and response variables of a transesterification reaction.

Figure 8.

Figure 8

Open in a new tab

Parity plots of the RSM and ANN response models with experimental values: yield (a), cetane number (b), viscosity parity (c), and purity (d).

3.5. Optimization of the BSO FAME Yield and Properties

The design plan presented in Table 1 was utilized to optimize the yield and properties of FAME produced from BSO catalyzed by KOH-assisted banana bunch stalk waste biochar. The yield and resulting properties are displayed in Figure 6. This study used optimization to determine the process parameters that would produce the maximum yield and the best product quality. Table 1 displays the levels and ranges of key process variables with catalyst concentration (A), CH3OH/BSO ratio (B), temperature (C), duration (D), and stirring speed (E) as the independent variables. The experimental findings included the percentage yield (Y), cetane number (CN), kinematic viscosity (VS), and purity (P), which were used as dependent variables. As determined by the latest Stat-Ease software, the responses (Y, CN, VS, and P) and the independent parameters are used to develop a model and are presented in coded form, as shown in eqs 7–10.

3.5. 7
3.5. 8
3.5. 9
3.5. 10

where the letters A through E represent the coded values of the independent variables and Y is the response variable (the % yield of FAME), CN, VS, and P. The equation above provides a quantitative explanation of how the parameters affect the FAME Y, CN, VS, and P. Equations 10–13 suggest that the yield and FAME properties both quadratically and linearly affect the parameters under investigation. While interaction effects reflect coefficients that actually have more than one variable, single effects, as presented in the model, are coefficients with just one variable. In the model terms under consideration, a positive sign denotes the synergistic impacts of the variable, whereas a negative sign denotes their antagonistic effects. The statistical analysis results are shown in Table 4, and the performance indices shown in Table 5 were used to evaluate the suitability of the model that was previously proposed at the 5% significance level. However, because the difference between the predicted and adjusted R2 values is less than 0.2, as shown in Table 5, the model response and the experimental response are reasonably in agreement; therefore, the models generated are adequate for predicting the FAME yield and its properties, as presented in the parity plot.4

Table 4. ANOVA was Used for the Response Model.

Y
 CN
 VS
 P
source F value p-value F value p-value F value p-value F value p-value
model 558.33 <0.0001a 1254.53 <0.0001a 954.23 <0.0001a 4220.25 <0.0001a
A 189.61 <0.0001 482.32 0.0021 47.80 <0.0001 1423.84 <0.0001
B 1174.53 <0.0001 4641.66 <0.0001 1335.84 <0.0001 2464.21 <0.0001
C 234.08 <0.0001 143.46 0.0431 85.69 <0.0001 1479.88 <0.0001
D 599.25 <0.0001 3717.39 <0.0001 113.70 <0.0001 3142.47 <0.0001
E 114.70 <0.0301 65.73 0.5101 0.1236 0.7318 193.76 <0.0001
AB 225.28 <0.0001 43.09 <0.0001 21.53 0.0007 6043.79 <0.0001
AC 224.72 <0.0001 778.01 <0.0001 227.07 <0.0001 1340.78 <0.0001
AD 1013.55 <0.0001 424.26 <0.0001 41.71 <0.0001 1244.46 <0.0001
AE 224.72 0.0601 4.65 0.054 3.48 0.089 376.73 <0.0001
BC 2559.68 <0.0001 3114.68 <0.0001 52.18 <0.0001 1514.98 <0.0001
BD 284.41 <0.0001 1034.17 0.0021 54.98 <0.0001 531.55 <0.0001
BE 3.44 0.0906 306.71 0.0601 0.5149 0.0588 2862.87 <0.0001
CD 424.86 <0.0001 2353.25 0.0001 288.4 <0.0001 60.73 0.0600
CE 126.40 0.0511 24.84 0.0551 3.48 0.089 51.78 0.0800
DE 31.81 0.0502 81.21 0.0501 0.0023 0.9627 1155.50 <0.0001
A2 879.55 <0.0001 6455.68 <0.0001 1.59 0.2337 19657.74 <0.0001
B2 555.45 <0.0001 768.60 <0.0001 273.78 <0.0001 711.24 <0.0001
C2 1316.18 <0.0001 808.24 <0.0001 35.09 <0.0001 998.91 <0.0001
D2 556.65 <0.0001 573 <0.0001 4411.76 <0.0001 2876.42 <0.0001
E2 605.79 <0.0001 96.41 <0.0001 0.087 0.7739 110.66 0.0891
Open in a new tab
a

Significant.

Table 5. Coefficients of Determination for the Response Models for FAME Yield and Properties.

responses unit std. dev. R2 (%) R2 (adjusted) (%) R2 (predicted) (%)
(Y) (%) 0.2668 0.945 0.9572 0.9543
(CN)   0.1055 0.9496 0.9588 0.9491
(VS) mm2/s 0.0523 0.9594 0.9584 0.9555
(P) (%) 0.0686 0.9540 0.953 0.9510
Open in a new tab

The model response ANOVA table (Table 4) was used to portray the models’ significance at the 5% significance level. A model’s p-value (significance probability value) was considered significant if it was less than 0.05. The p-values presented indicate that four variables, catalyst concentration (A), methanol/oil molar ratio (B), reaction temperature (C), and time (D), have important or significant impacts on the FAME yield and properties, as indicated by the F values (Table 4). B had the greatest impact, with F values of (1174.53), (4641.66), (1335.84), and (2464.21), representing 50.79, 51.285, 84.37, and 28.3%, respectively, of the sensitivity impacts of the input variables to the response variables (Y, CN, VS, and P). The agitation speed (E) had a minimally significant effect on the VS and CN of FAME. Significant interaction effects between the factors under investigation were also detected; (BC) and (CD) had the highest interaction effects on the FAME yield (Y), with observed F values of 2559.6 and 424.86, respectively, indicating a lower probability (<0.0001). For the fuel properties, (BC) and (CD) have the greatest impact on the CN response, whereas (CD) and (AC) have the greatest impact on the VS response; ultimately, (AB) affects the response of the purity more than other interacting factors. This is because the low catalyst concentration (A) and methanol-to-oil molar ratio (B) may lead to inadequate amounts of catalyst and alcohol in the reaction system, resulting in incomplete reactions during the conversion of BSO to methyl esters, thereby affecting the purity of FAME. Increasing these parameters toward their optimal values significantly enhances the FAME purity due to the presence of sufficient catalytic active sites in the system, enabling the overcoming of energy barriers. Moreover, a high methanol-to-oil molar ratio above the stoichiometric value contributes to driving high-purity FAMEs. Therefore, it is crucial to monitor and control these two parameters to achieve high-quality FAMEs. The test of significance terms in the models, as shown in Table 4, is carried out at the 5% significance level, and the statistically insignificant terms in the model are eliminated, as shown in eqs 11–14.

3.5. 11
3.5. 12
3.5. 13
3.5. 14

The outcome of the optimization does, in fact, show that it is possible to maximize Y without sacrificing the resulting CN, VS, or P of the FAMEs produced. The optimal values of the process parameters corresponding to the predicted optimal responses are therefore presented in Figure 9. From the results, acceptable kinematic viscosity, cetane number, and purity were achieved simultaneously with a high yield of the FAMEs.

Figure 9.

Figure 9

Open in a new tab

Optimum conditions for Y with the resulting CN, VS, and P of the BSO FAME.

The perturbation plot, which is depicted in Figure 10, displays the relative impact of all of the parameters at a particular point in the design space. The process input variables that have a great sensitivity to the FAME yield and its properties can be identified using this plot. A significant steep slope or curvature of components A, B, C, D, and E demonstrated that the FAME yield and property responses are indeed sensitive to all of the input predictors taken into account in this study, corroborating the optimization experiments previously provided. The higher slope of the methanol:oil molar ratio to the FMAE yield and its properties are indicative of its high sensitivity to the model responses. This might be because, in practice, another simple way to increase product yield, most importantly, is to shift the transesterification reaction equilibrium forward by using more methanol because the transesterification reaction is reversible. The results obtained indicated that the methanol/oil molar ratio has a high sensitivity effect on the BSO FAME yield and properties. A higher molar ratio of alcohol to oil resulted in a higher yield of ester and a better quality as a result of the presence of more alcohol, which drove the conversion of BSO into FAME in the transesterification reaction. When the molar ratio of CH3OH/BSO exceeded the realized optimal point, the yield decreased. This decrease could be attributed to a decrease in catalyst activity as a result of overflooding of the catalytic active site with alcohol as methanol levels increase, which also affects the breakdown of BSO bonds and thus the reduction in FAME yield and quality. The perturbation plots analyzed by CCD help us to compare the sensitivity of all of the factors during this study. A relatively flat straight line indicates insensitivity to changes in a specific factor, which is uncommon in this case.

Figure 10.

Figure 10

Open in a new tab

Perturbation plot of the BSO FAME yield (a), cetane number (b), kinematic viscosity (c), and purity (d).

Catalyst reusability plays a crucial role in the application of catalysts in commercial-scale and industrial processes. To assess reusability, five successive cycles were tested under optimized reaction conditions. These conditions included a temperature of 56 °C, a reaction time of 115 min, a CH3OH/BSO molar ratio of 15:1, a catalyst dosage of 6 wt %, and a stirring speed of 400 rpm, as depicted in Figure 11. The FAME conversion was recorded after each cycle, revealing a decrease in conversion during each subsequent cycle. After the fifth cycle, the FAME conversion reached just 80%. This decrease can be attributed to the slight leaching of active sites from the catalyst during the reaction, which results in a decrease in catalytic activity. Catalysts may lose active sites or undergo changes in their surface structure over prolonged use, leading to a reduction in available sites for the desired reactions and, consequently, lower FAME yields.

Figure 11.

Figure 11

Open in a new tab

Demonstration of the reusability of the catalyst in the transesterification of BSO to FAME. The reaction conditions were as follows: temperature, 56 °C; reaction time, 115 min; CH3OH/BSO molar ratio, 15:1; catalyst dosage, 6 wt %; and stirring speed, 400 rpm.

4. Conclusions

Baobab seeds are abundant and have a high oil content (36.8 wt %), making them a viable feedstock for FAME manufacturing. The physiochemical properties of the oil indicate that it contains a high concentration and properties of triglycerides, with a dominant unsaturated fatty acid content of approximately 83.49% of the total fatty acid content; hence, the oil belongs to the monounsaturated fatty acid category. The XRD and EDS analyses revealed the presence of minerals, including K, Ca, Mg, Si, Na, and Zn, in the BBB. There were noticeable morphological variations and an increase in pores between the UBB and BBB, which validated the effect of the calcination and activation treatments of the used catalyst. The MIMO deep learning back-propagation training algorithm adequately predicted Y, CN, VS, and P under various reaction conditions considered with a strong correlation coefficient and low RMSE of 0.00755. The MIMO artificial neural network, therefore, was demonstrated to be a successful technique for predicting the yield and properties of the biofuel. The optimization results showed that it is possible to maximize yield without sacrificing the resulting CN, VS, or P of FAMEs produced from BSO. Based on the revised model, we achieved a 96 wt % FAME yield with a 48 FAME cetane number, 3.3 mm2/s FAME kinematic viscosity, and 98.3% FAME purity. From the statistical indices obtained, the MIMO logsig transfer function of the ANN model marginally outperformed the MISO and RSM models in predicting the biofuel yield and properties because RSM and MISO may have difficulty accurately capturing the complex interactions and dependencies among FAME production process parameters. On the other hand, MIMO neural network models can learn, adapt, and effectively identify complex relationships and dependencies within the data. This makes them a more suitable choice for modeling FAME production and can be employed in predicting these responses with any given input variable. Overall, we view nonedible baobab seeds as feedstocks with considerable potential for producing high yields and quality FAME products as a replacement for fossil fuels.

Acknowledgments

The authors acknowledge financial support from the National Natural Science Foundation of China (21776063).

Author Contributions

§ C.C.E. and C.L. authors contributed equally to this work and should be considered co-first authors.

The authors declare no competing financial interest.

References

  1. Hosseinzadeh-Bandbafha H.; Nizami A. S.; Kalogirou S. A.; Gupta V. K.; Park Y. K.; Fallahi A.; Sulaiman A.; Ranjbari M.; Rahnama H.; Aghbashlo M.; Peng W.; Tabatabaei M. Environmental Life Cycle Assessment of Biodiesel Production from Waste Cooking Oil: A Systematic Review. Renewable Sustainable Energy Rev. 2022, 161, 112411 10.1016/j.rser.2022.112411. [DOI] [Google Scholar]
  2. Li Y.; Wen Y.; Chen B.; Fu X.; Wu Y. The Dilemma and Potential Development of Biodiesel in China - In View of Production Capacity and Policy. Energy Sustainable Dev. 2023, 75, 60–71. 10.1016/j.esd.2023.05.005. [DOI] [Google Scholar]
  3. Fangfang F.; Alagumalai A.; Mahian O. Sustainable Biodiesel Production from Waste Cooking Oil: ANN Modeling and Environmental Factor Assessment. Sustainable Energy Technol. Assess. 2021, 46, 101265 10.1016/j.seta.2021.101265. [DOI] [Google Scholar]
  4. Chimezie E. C.; Wang Z.; Yu Y.; Nonso U. C.; Duan P. G.; Kapusta K. Yield Optimization and Fuel Properties Evaluation of the Biodiesel Derived from Avocado Pear Waste. Ind. Crops Prod. 2023, 191, 115884 10.1016/j.indcrop.2022.115884. [DOI] [Google Scholar]
  5. Chimezie E. C.; Zhang X.; Djandja O. S.; Nonso U. C.; Duan P. G. Biodiesel Production from Nonedible Feedstocks Catalyzed by Nanocatalysts: A Review. Biomass Bioenergy 2022, 163, 106509 10.1016/j.biombioe.2022.106509. [DOI] [Google Scholar]
  6. Mohite A.; Bora B. J.; Ağbulut Ü.; Sharma P.; Medhi B. J.; Barik D. Optimization of the Pilot Fuel Injection and Engine Load for an Algae Biodiesel–Hydrogen Run Dual Fuel Diesel Engine Using Response Surface Methodology. Fuel 2024, 357, 129841 10.1016/j.fuel.2023.129841. [DOI] [Google Scholar]
  7. Hosseinzadeh-Bandbafha H.; Nizami A. S.; Kalogirou S. A.; Gupta V. K.; Park Y. K.; Fallahi A.; Sulaiman A.; Ranjbari M.; Rahnama H.; Aghbashlo M.; Peng W.; Tabatabaei M. Environmental Life Cycle Assessment of Biodiesel Production from Waste Cooking Oil: A Systematic Review. Renewable Sustainable Energy Rev. 2022, 161, 112411 10.1016/j.rser.2022.112411. [DOI] [Google Scholar]
  8. Hashemzehi M.; Pirouzfar V.; Nayebzadeh H.; Su C. H. Modeling and Optimization of Main Independent Parameters for Biodiesel Production over a Cu0.4Zn0.6Al2O4 Catalyst Using an RSM Method. J. Chem. Technol. Biotechnol. 2022, 97, 111–119. 10.1002/jctb.6916. [DOI] [Google Scholar]
  9. Topi D. Transforming Waste Vegetable Oils to Biodiesel, Establishing of a Waste Oil Management System in Albania. SN Appl. Sci. 2020, 2, 513. 10.1007/s42452-020-2268-4. [DOI] [Google Scholar]
  10. Sumayli A. Development of Advanced Machine Learning Models for Optimization of Methyl Ester Biofuel Production from Papaya Oil: Gaussian Process Regression (GPR), Multilayer Perceptron (MLP), and K-Nearest Neighbor (KNN) Regression Models. Arabian J. Chem. 2023, 16, 104833 10.1016/j.arabjc.2023.104833. [DOI] [Google Scholar]
  11. Mehra K. S.; Pant G. Production of Biofuel From Sesame Oil And Its Characterization As An Alternative Fuel For Diesel Engine. IOP Conf. Ser.: Mater. Sci. Eng. 2021, 1116, 012076 10.1088/1757-899X/1116/1/012076. [DOI] [Google Scholar]
  12. Xavier J. Performance and Emission Characteristics of Diesel Engine Using Watermelon Seed Oil Methyl Ester. J. Biol. Sci. 2020, 20, 88–93. 10.3923/jbs.2020.88.93. [DOI] [Google Scholar]
  13. Silva A.; Farias A. F.; Meneghetti S. M.; dos Santos Filho E. A.; de Melo Costa A. C. Optimization of Biodiesel Production Via Transesterification of Soybean Oil Using A-MoO3 Catalyst Obtained by the Combustion Method. Arabian J. Chem. 2022, 15, 104012 10.1016/j.arabjc.2022.104012. [DOI] [Google Scholar]
  14. Abukhadra M. R.; Adlii A.; Bin Jumah M. N.; Othman S. I.; Alruhaimi R. S.; Salama Y. F.; Allam A. A. Sustainable Conversion of Waste Corn Oil into Biofuel over Different Forms of Synthetic Muscovite Based K+/Na+ sodalite as Basic Catalysts; Characterization and Mechanism. Mater. Res. Express 2021, 8, 065502 10.1088/2053-1591/ac0367. [DOI] [Google Scholar]
  15. Veljković V. B.; Biberdžić M. O.; Banković-Ilić I. B.; Djalović I. G.; Tasić M. B.; Nježić Z. B.; Stamenković O. S. Biodiesel Production from Corn Oil: A Review. Renewable Sustainable Energy Rev. 2018, 91, 531–548. 10.1016/j.rser.2018.04.024. [DOI] [Google Scholar]
  16. Ayoola A. A.; Hymore F. K.; Omonhinmin C. A.; Olawole O. C.; Fayomi O. S. I.; Babatunde D.; Fagbiele O. Analysis of Waste Groundnut Oil Biodiesel Production Using Response Surface Methodology and Artificial Neural Network. Chem. Data Collect. 2019, 22, 100238 10.1016/j.cdc.2019.100238. [DOI] [Google Scholar]
  17. Pannilawithana N. A.; Pathirana H. M. A Green Method to Produce Biodiesel from Palm Olein Oil. J. Oil Palm Res. 2017, 29, 267–277. 10.21894/jopr.2017.2902.11. [DOI] [Google Scholar]
  18. Alkaaby M. H.; Balla H. H.; Al-Zuhairy M. S. Biodiesel Production from Palm Olein Oil Using Acid-Catalyzed Esterification. AIP Conf. Proc. 2021, 2404, 080026 10.1063/5.0068892. [DOI] [Google Scholar]
  19. Silva A. L.; Farias A. F. F.; Meneghetti S. M. P.; Antonio dos Santos Filho E.; Figueiredo de Melo Costa A. C. Optimization of Biodiesel Production via Transesterification of Soybean Oil Using α-MoO3 Catalyst Obtained by the Combustion Method. Arabian J. Chem. 2022, 15, 104012 10.1016/j.arabjc.2022.104012. [DOI] [Google Scholar]
  20. Sun X.; Opulencia M. J. C.; Alexandrovich T. P.; Khan A.; Algarni M.; Abdelrahman A. Modeling and Optimization of Vegetable Oil Biodiesel Production with Heterogeneous Nano Catalytic Process: Multi-Layer Perceptron, Decision Regression Tree, and K-Nearest Neighbor Methods. Environ. Technol. Innovation 2022, 27, 102794 10.1016/j.eti.2022.102794. [DOI] [Google Scholar]
  21. Zahed M. A.; Revayati M.; Shahcheraghi N.; Maghsoudi F.; Tabari Y. Modeling and Optimization of Biodiesel Synthesis Using TiO2–ZnO Nanocatalyst and Characteristics of Biodiesel Made from Waste Sunflower Oil. Curr. Res. Green Sustainable Chem. 2021, 4, 100223 10.1016/j.crgsc.2021.100223. [DOI] [Google Scholar]
  22. Gupta R.; McRoberts R.; Yu Z.; Smith C.; Sloan W.; You S. Life Cycle Assessment of Biodiesel Production from Rapeseed Oil: Influence of Process Parameters and Scale. Bioresour. Technol. 2022, 360, 127532 10.1016/j.biortech.2022.127532. [DOI] [PubMed] [Google Scholar]
  23. Ude C. N.; Onukwuli O. D.; Ugwu B. I.; Okey-Onyesolu C. F.; Ezidinma T. A.; Ejikeme P. M. Methanolysis Optimization of Cottonseed Oil to Biodiesel Using Heterogeneous Catalysts. Iran. J. Chem. Chem. Eng. 2020, 39, 355–370. 10.30492/ijcce.2020.43368. [DOI] [Google Scholar]
  24. Tan S. X.; Lim S.; Ong H. C.; Pang Y. L.; Fitranto K.; Goh B. H. H.; Chong C. T. Two-Step Catalytic Reactive Extraction and Transesterification Process via Ultrasonic Irradiation for Biodiesel Production from Solid Jatropha Oil Seeds. Chem. Eng. Process. 2019, 146, 107687 10.1016/j.cep.2019.107687. [DOI] [Google Scholar]
  25. Biodiesel P. U. P. Synthesis and Characterization of Rubber Seed Shell Impregnated with Calcium Oxide as Catalyst for Biodiesel Production. Malaysian J. Anal. Sci. 2021, 25, 561–568. [Google Scholar]
  26. Dass P. M.; Danbature W. L.; Karu E.; Ibrahim A.; Baba A. Extraction and Biodegradation of Baobab (Adansonia Digitata) Seeds Oil by Fungi (Aspergillus Niger). J. Nat. Sci. Res. 2013, 3, 128–137. [Google Scholar]
  27. Tarigan J. B.; Perangin-Angin S.; Simanungkalit S. R.; Zega N. P.; Sitepu E. K. Utilization of Waste Banana Peels as Heterogeneous Catalysts in Room-Temperature Biodiesel Production Using a Homogenizer. RSC Adv. 2023, 13 (9), 6217–6224. 10.1039/D3RA00016H. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Manimaran P.; Sanjay M. R.; Senthamaraikannan P.; Jawaid M.; Saravanakumar S. S.; George R. Synthesis and Characterization of Cellulosic Fiber from Red Banana Peduncle as Reinforcement for Potential Applications. J. Nat. Fibers 2019, 16, 768–780. 10.1080/15440478.2018.1434851. [DOI] [Google Scholar]
  29. Balajii M.; Niju S. Banana Peduncle – A Green and Renewable Heterogeneous Base Catalyst for Biodiesel Production from Ceiba Pentandra Oil. Renewable Energy 2020, 146, 2255–2269. 10.1016/j.renene.2019.08.062. [DOI] [Google Scholar]
  30. Samuel O. D.; Okwu M. O.; Oyejide O. J.; Taghinezhad E.; Afzal A.; Kaveh M. Optimizing Biodiesel Production from Abundant Waste Oils through Empirical Method and Gray Wolf Optimizer. Fuel 2020, 281, 118701 10.1016/j.fuel.2020.118701. [DOI] [Google Scholar]
  31. Sukpancharoen S.; Hansirisawat P.; Srinophakun T. R. Implementation of Response Surface to Optimum Biodiesel Power Plant Derived from Empty Fruit Bunch. J. Energy Resour. Technol. 2022, 144, 012101 10.1115/1.4050817. [DOI] [Google Scholar]
  32. Pugazhendhi A.; Alagumalai A.; Mathimani T.; Atabani A. E. Optimization, Kinetic and Thermodynamic Studies on Sustainable Biodiesel Production from Waste Cooking Oil: An Indian Perspective. Fuel 2020, 273, 117725 10.1016/j.fuel.2020.117725. [DOI] [Google Scholar]
  33. Corral Bobadilla M.; Fernández Martínez R.; Lorza R. L.; Gómez F. S.; González E. P. V. Optimizing Biodiesel Production from Waste Cooking Oil Using Genetic Algorithm-Based Support Vector Machines. Energies 2018, 11, 2995. 10.3390/en11112995. [DOI] [Google Scholar]
  34. Ude C. N.; Onukwuli D. O.; Okey-Onyesolu F. C.; Nnaji P. C.; Okoye C. C.; Uwaleke C. C. Prediction of Some Thermo-Physical Properties of Biodiesel Using ANFIS and ANN Cum Sensitivity Analysis. Cleaner Waste Syst. 2022, 2, 100006 10.1016/j.clwas.2022.100006. [DOI] [Google Scholar]
  35. Tuan N. V.; Minh D. Q.; Khoa N. X.; Lim O. A Study to Predict Ignition Delay of an Engine Using Diesel and Biodiesel Fuel Based on the ANN and SVM Machine Learning Methods. ACS Omega 2023, 8, 9995–10005. 10.1021/acsomega.2c07186. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Esonye C.; Onukwuli O. D.; Ofoefule A. U.; Ogah E. O. Multi-Input Multi-Output (MIMO) ANN and Nelder–Mead’s Simplex Based Modeling of Engine Performance and Combustion Emission Characteristics of Biodiesel-Diesel Blend in CI Diesel Engine. Appli. Thermal Eng. 2019, 151, 100–114. 10.1016/j.applthermaleng.2019.01.101. [DOI] [Google Scholar]
  37. Ude C. N.; Onukwuli O. D.; Nwobi-Okoye C.; Anisiji O. E.; Atuanya C. U.; Menkiti M. C. Performance Evaluation of Cottonseed Oil Methyl Esters Produced Using CaO and Prediction with an Artificial Neural Network. Biofuels 2020, 11, 77–84. 10.1080/17597269.2017.1345355. [DOI] [Google Scholar]
  38. Gad A. F.; Jarmouni F. E.. Chapter 3 - ANN with 1 input and 1 output. In Introduction to Deep Learning and Neural Networks with PythonTM- A Practical Guide, 1st ed.; Gad A. F.; Jarmouni F. E.. Academic Press: New York, 2020; pp 33–45 10.1016/b978-0-323-90933-4.00003-6. [DOI] [Google Scholar]
  39. Chakraborty R.; Banerjee A. Prediction of Fuel Properties of Biodiesel Produced by Sequential Esterification and Transesterification of Used Frying Soybean Oil Using Statistical Analysis. Waste Biomass Valorization 2010, 1 (2), 201–208. 10.1007/s12649-010-9016-8. [DOI] [Google Scholar]
  40. Chakraborty R.; Sahu H. Intensification of Biodiesel Production from Waste Goat Tallow Using Infrared Radiation: Process Evaluation through Response Surface Methodology and Artificial Neural Network. Appl. Energy 2014, 114, 827–836. 10.1016/j.apenergy.2013.04.025. [DOI] [Google Scholar]
  41. Ude C. N.; Onukwuli O. D.; Uchegbu N. N.; Umeuzuegbu J. C.; Amulu N. F. Evaluation of Engine Performance and Emission of African Pear Seed Oil (APO) Biodiesel and Its Prediction via Multi-Input-Multi-Output Artificial Neural Network (ANN) and Sensitivity Analysis. Biofuels, Bioprod. Biorefin. 2021, 15 (3), 703–718. 10.1002/bbb.2200. [DOI] [Google Scholar]
  42. Jitjamnong J.; Thunyaratchatanon C.; Luengnaruemitchai A.; Kongrit N.; Kasetsomboon N.; Sopajarn A.; Chuaykarn N.; Khantikulanon N. Response Surface Optimization of Biodiesel Synthesis over a Novel Biochar-Based Heterogeneous Catalyst from Cultivated (Musa Sapientum) Banana Peels. Biomass Convers. Biorefin. 2021, 11 (6), 2795–2811. 10.1007/s13399-020-00655-8. [DOI] [Google Scholar]
  43. Khethiwe E.; Clever K.; Jerekias G. Effects of Fatty Acids Composition on Fuel Properties of Jatropha Curcas Biodiesel. Smart Grid Renewable Energy 2020, 11 (10), 165–180. 10.4236/sgre.2020.1110010. [DOI] [Google Scholar]
  44. ASTM (American Society for Testing and Materials) . Standard Test Method for Kinematic Viscosity of Transparent and Opaque Liquids (and Calculation of Dynamic Viscosity); ASTM D445–2021, 2021. [Google Scholar]
  45. Standard AS . D613–8 Standard Test Method for Cetane Number of Diesel Fuel Oil ASTM International, West Conshohocken: PA, 2008. [Google Scholar]
  46. Singh Pali H.; Sharma A.; Kumar N.; Singh Y. Biodiesel Yield and Properties Optimization from Kusum Oil by RSM. Fuel 2021, 291, 120218 10.1016/j.fuel.2021.120218. [DOI] [Google Scholar]
  47. Walia M.; Rawat K.; Bhushan S.; Padwad Y. S.; Singh B. Fatty Acid Composition, Physicochemical Properties, Antioxidant and Cytotoxic Activity of Apple Seed Oil Obtained from Apple Pomace. J. Sci. Food Agric. 2014, 94 (5), 929–934. 10.1002/jsfa.6337. [DOI] [PubMed] [Google Scholar]
  48. Bombo K.; Lekgoba T.; Azeez O.; Muzenda E. Production of Biodiesel from Moringa Oleifera and Jatropha Curcas Seed Oils over a Modified ZnO/Fly Ash Catalyst. Environ. Clim. Technol. 2021, 25, 151–160. 10.2478/rtuect-2021-0010. [DOI] [Google Scholar]
  49. Ajadi I. K.; Oyegoke T.; Geoffrey T. T.; Fasanya O.; Ojetunde A. O.. Biodiesel Production from Neem Seed Oil: Catalyst Synthesis, Effect of Time, and Temperature on Biodiesel Yield. In Conference: Proceedings of the Materials Science & Technology Society of Nigeria 2020; pp 1–13.
  50. Huang Y.; Li F.; Bao G.; Xiao Q.; Wang H. Modeling the Effects of Biodiesel Chemical Composition on Iodine Value Using Novel Machine Learning Algorithm. Fuel 2022, 316, 123348 10.1016/j.fuel.2022.123348. [DOI] [Google Scholar]
  51. Fan M.; Wu H.; Shi M.; Zhang P.; Jiang P. Well-Dispersive K2O–KCl Alkaline Catalyst Derived from Waste Banana Peel for Biodiesel Synthesis. Green Energy Environ. 2019, 4 (3), 322–327. 10.1016/j.gee.2018.09.004. [DOI] [Google Scholar]
  52. Jitjamnong J.; Thunyaratchatanon C.; Luengnaruemitchai A.; Kongrit N.; Kasetsomboon N.; Sopajarn A.; Chuaykarn N.; Khantikulanon N. Response Surface Optimization of Biodiesel Synthesis over a Novel Biochar-Based Heterogeneous Catalyst from Cultivated (Musa Sapientum) Banana Peels. Biomass Convers. Biorefin. 2021, 11 (6), 2795–2811. 10.1007/s13399-020-00655-8. [DOI] [Google Scholar]

Articles from ACS Omega are provided here courtesy of American Chemical Society

RESOURCES