Predicting Enthalpy of Combustion Using Machine Learning

Abstract

The present work discusses the development and application of a machine-learning-based model to predict the enthalpy of combustion of various oxygenated fuels of interest. A detailed dataset containing 207 pure compounds and 38 surrogate fuels has been prepared, representing various chemical classes, namely paraffins, olefins, naphthenes, aromatics, alcohols, ethers, ketones, and aldehydes. The dataset was subsequently used for constructing an artificial neural network (ANN) model with 14 input layers, 26 hidden layers, and 1 output layer for predicting the enthalpy of combustion for various oxygenated fuels. The ANN model was trained using the collected dataset, validated, and finally tested to verify its accuracy in predicting the enthalpy of combustion. The results for various oxygenated fuels are discussed, especially in terms of the influence of different functional groups in shaping the enthalpy of combustion values. In predicting the enthalpy of combustion, 96.3% accuracy was achieved using the ANN model. The developed model can be successfully employed to predict the enthalpies of neat compounds and mixtures as the obtained percentage error of 4.2 is within the vicinity of experimental uncertainty.

1. Introduction

Thermochemistry is a subfield of chemical thermodynamics that studies the interactions of heat, work, and other forms of energy in the context of chemical and physical processes. When a system undergoes a change of state, its internal energy, enthalpy, and associated properties change to account for the energy transfer between the system and the surroundings. All heat engines around us, including steam power plants, gas power plants, and automobiles, operate by converting the chemical energy of the fuel into thermal energy, which is subsequently converted to the desired end-use form. In this regard, the enthalpy of combustion is defined as the quantum of change in enthalpy when any element/compound undergoes complete oxidation at a given temperature and pressure. Enthalpies of combustion for various substances are typically measured using a bomb calorimeter (see Figure 1). However, measuring the enthalpy of combustion of various fuels, fuel blends, and surrogates is a time-consuming and costly procedure which necessitates the use of alternate approaches. Among the various possible approaches, the use of machine learning to predict the enthalpy of combustion is promising and has been explored extensively in this work.

Machine-learning tools help in predicting the enthalpy of combustion for various fuels once they have been rigorously tested over a predefined dataset. A number of machine-learning algorithms have been developed over the years, such as decision trees, random forests, Naïve Bayes, support vector machines, k-means, etc. Neural networks are a subset of machine learning and are at the heart of deep learning algorithms. Typical machine-learning methods are based on the use of neural networks with one or more layers connecting the desired output to the given input. A neural network is a set of algorithms that attempts to detect underlying relationships in a piece of data using a technique similar to the way that the human brain works. In this context, neural networks are systems of neurons that might be biological or artificial in origin. Due to their excellent adaptability to changing inputs, neural networks are widely being used to obtain the best possible results without the need to change the output criterion. Artificial neural networks (ANNs) have therefore been used in this study to develop the required machine-learning tools.

The use of ANNs for predicting the enthalpy of combustion is rather novel. The few relevant studies that have utilized multiple approaches for developing predictive models for the estimation of the enthalpy of combustion are discussed here. Gharagheizi et al. [1] developed an ANN-based model to predict the enthalpy of combustion of various pure compounds using the group-contribution method. A squared correlation coefficient of 0.99999 with a root-mean-squared error of 12.57 kJ/mol, in comparison with experimental values from the literature, was reported. Albahri [2] also used the group-contribution method for predicting the enthalpy of combustion; however, a multivariable nonlinear-regression-based method was used for this purpose. The enthalpy of combustion was predicted based on the molecular structure, with a reported average error of 0.71%. Recently, Dashti et al. [3] compared three different models, namely a least-squares support-vector machine, genetic programming, and an adaptive neuro-fuzzy inference system, for predicting the enthalpy of combustion of several chemical compounds. It was concluded that the particle swarm optimization–adaptive neuro-fuzzy inference system (PSO–ANFIS) model with four inputs was the most accurate among the three models. The ANFIS structure contains five layers: 1: the fuzzification layer; 2: the IF layer; 3: the normalization layer; 4: the THEN layer; and 5: the summation layer. In this model, MF variables should be estimated immediately after the generation of the initial FIS model and the training of the ANFIS. The MF values are adjusted via PSO algorithms to find the finest structure [3].

This study reports a detailed ANN model to predict the enthalpy of combustion of compounds (neat and mixtures) and real fuels belonging to different chemical families. The functional groups present in the fuels are used as input features to the ANN model. The method of using functional groups for predicting fuel properties has been reported in a number of recent studies [4,5,6]. Functional groups present in the fuels are also relevant for surrogate formulation [7], particulate matter formation [8,9,10,11], and property prediction [12,13,14].

2. Theoretical Background

The enthalpy of combustion is calculated using the data from the bomb calorimeter using Equation (1):

$$\Delta H_c=m\times C_p\times \Delta T$$

(1)

where ∆H_c is the enthalpy of combustion in kJ, m is the mass of water in kg, C_p is the specific heat capacity of water in kJ/kg °C, and ∆T is the temperature change of water in °C. After obtaining these enthalpies of combustion from the literature, each compound was defined with a specific variable according to a convention so that the machine could relate the given enthalpy of combustion to the corresponding compound. To define each compound, 13 variables were assigned to each compound. These variables, along with their definitions, are presented in

Table 1. Variables, along with their definitions.

No.	Variables	Definitions
1	P.CH3 (wt %)	Weight percentage of paraffinic CH3 groups in the compound
2	P.CH2 (wt %)	Weight percentage of paraffinic CH2 groups in the compound
3	P.CH (wt %)	Weight percentage of paraffinic CH groups in the compound
4	Olef (wt %)	Weight percentage of olefin groups in the compound
5	Naph (wt %)	Weight percentage of naphthene groups in the compound
6	Arom (wt %)	Weight percentage of aromatic groups in the compound
7	Alc OH (wt %)	Weight percentage of alcohol groups in the compound
8	Ether O (wt %)	Weight percentage of ether groups in the compound
9	Aldeh CHO (wt %)	Weight percentage of aldehyde groups in the compound
10	Ketone CO (wt %)	Weight percentage of ketone groups in the compound
11	Ester (wt %)	Weight percentage of ester groups in the compound
12	Mol wt	Molecular weight of the compound
13	BI	Branching index of the compound

.

BI refers to the branching index of the molecule or mixture. It represents the degree of branching/linearity of a molecule, while including the impact of paraffinic branches on the longest chain present in the molecule. More information related to BI has been reported here [15,16].

3. Dataset and Machine Learning

The dataset was built using the collection of data from the extensive available literature on the enthalpy of combustion that meets the criteria that all data should be at the same condition (i.e., temperature and pressure) and that the method used to obtain these enthalpies should be consistent (i.e., bomb calorimeter method).

After the collection of all the possible data,

Table 2. Enthalpy of combustion of different compounds from the literature.

Sr. No.	Compound Name	Compound Formula	Enthalpy of Combustion (Kcal/mol)	References
1	Ethane	C2H6	372.82	[26]
2	Propane	C3H8	530.605
3	n-Butane	C4H10	687.982
4	n-Pentane	C5H12	838.8
5	n-Hexane	C6H14	995.01
6	n-Heptane	C7H16	1151.27
7	n-Octane	C8H18	1307.53
8	n-Nonane	C9H20	1463.3
9	n-Decane	C10H22	1620.06
10	n-Undecane	C11H24	1776.32
11	n-Dodecane	C12H26	1932.59
12	n-Hexadecane	C16H34	2557.64
13	2-Methylpropane	C4H10	686.342
14	2-Methylbutane	C5H12	837.3
15	2,2-Dimethylpropane	C5H12	840.49
16	2-Methylpentane	C6H14	993.71
17	3-Methylpentane	C6H14	994.25
18	2,2-Dimethylbutane	C6H14	991.52
19	2,3-Dimethylbutane	C6H14	993.05
20	2-Methylhexane	C7H16	1149.97
21	3-Methylhexane	C7H16	1150.55
22	3-Ethylpentane	C7H16	1151.13
23	2,2-Dimethylpentane	C7H16	1147.85
24	2,3-Dimethylpentane	C7H16	1149.09
25	2,4-Dimethylpentane	C7H16	1148.73
26	3,3-Dimethylpentane	C7H16	1148.83
27	2,2,3-Trimethylbutane	C7H16	1148.27
28	2-Methylheptane	C8H18	1306.28
29	3-Methylheptane	C8H18	1306.92
30	4-Methylheptane	C8H18	1307.09
31	3-Ethylhexane	C8H18	1307.39
32	2,2-Dimethylhexane	C8H18	1304.64
33	2,3-Dimethylhexane	C8H18	1306.86
34	2,4-Dimethylhexane	C8H18	1305.8
35	2,5-Dimethylhexane	C8H18	1305
36	3,3-Dimethylhexane	C8H18	1305.68
37	3,4-Dimethylhexane	C8H18	1307.04
38	2-Methyl-3-ethylpentane	C8H18	1307.58
39	3-Methyl-3-ethylpentane	C8H18	1306.8
40	2,2,3-Trimethylpentane	C8H18	1305.83
41	2,2,4-Trimethylpentane	C8H18	1305.29
42	2,3,3-Trimethylpentane	C8H18	1306.64
43	2,3,4-Trimethylpentane	C8H18	1306.28
44	2,2,3,3-Tetramethylbutane	C8H18	1303.03
45	n-Tridecane	C13H28	2088.85
46	n-Tetradecane	C14H30	2245.11
47	n-Pentadecane	C15H32	2401.37
48	n-Heptadecane	C17H36	2713.9
49	n-Octadecane	C18H38	2870.16
50	n-nonadecane	C19H40	3026.43
51	n-Eicosane	C20H42	3182.69
52	Ethylene	C2H4	337.25	[27]
53	Propylene	C3H6	419.9
54	n-1-Butene	C4H8	649.66
55	n-1-Pentene	C5H10	806.78
56	n-1-Hexene	C6H12	963.9
57	n-1-Heptene	C7H14	1120.9
58	n-1-Octene	C8H16	1277.97
59	n-1-Nonene	C9H18	1434.9
60	n-1-Decene	C10H20	1591.95
61	Benzene	C6H6	3267.49	[28]
62	Toluene	C7H8	3909.9
63	Naphthalene	C9H10	5157
64	2-Methyl-2-butene	C13H12	3362.2	[29]
65	1-Methyl-1-cyclohexene	C7H12	4353
66	1-Pentanol	C5H12O	3329.96	[30]
67	1-Octanol	C8H18O	5292.5
68	1-Butanol	C4H10O	2675.61
69	1-Decanol	C10H22O	10,468.26
70	1,2,3-trimethylcyclohexane	C12H10O2	5837.7	[29]
71	Ethylcyclohexane	C12H10O2	5059.1
72	methylcyclohexane	C12H10O2	4565.9
73	2-Methylheptane	C12H10O2	5464.7
74	Bicyclohexane	C6H10	3818.8
75	1,3-Dimethylcyclopentane	C7H14	4561.3
76	1,2,4-Trimethylcyclopentane	C8H16	5208.3
77	1,1-Dimethylcyclohexane	C8H16	5196.1
78	1,3-Dimethylcyclohexane	C8H16	5177.3
79	1,4-Dimethylcyclohexane	C8H16	5138.8
80	1,2,3-Trimethylcyclohexane	C9H18	5837.7
81	1,3,3-Trimethylcyclohexane	C9H18	5832.6
82	3,3-Dimethylcyclohexene	C9H18	4995.4
83	Bicycloheptane	C7H12	4308.7
84	Ethylenecyclohexane	C10H18	5059.1
85	Isopropyl-1-cyclohexene-1	C9H16	5611.2
86	o-Xylene	C8H10	4578.1
87	m-Xylene	C8H10	4567.7
88	p-Xylene	C8H10	4556.8
89	Diamylene	C10H20	6616.8
90	1-methyl-3-cyclohexene	C7H12	4364.3
91	Phenyl-1-butene-2	C10H12	5702.1
92	Isopropyl alcohol	C3H3O	1985.6
93	Ethylvinyl carbinol	C5H10O	3148.6
94	Pinacolyl alcohol	C6H14O	3925.2
95	1,3-Dimethylcyclohexanol-5	C8H16O	4949
96	Trimethyl carbinol	C4H10O	2631.7
97	Dimethylethyl carbinol	C5H12O	3281.2
98	1,3-Dimethylcyclohexanol-2	C8H16O	5001.7
99	2-Methyl-2-propanol	C4H10O	2631.7
100	Ethylene glycol	C2H6O2	1179.8
101	Propylene glycol	C3H8O2	1802.4
102	Phenol	C6H6O	3064.3
103	Glycerol	C3H8O2	1661.5
104	Anisole	C7H8O	3787.8
105	Phenetole	C8H10O	4423.1
106	m-Cresol methyl ether	C8H10O	4423.6
107	Safrole	C10H10O2	5206.6
108	Isosafrole	C10H10O2	5163.9
109	Acenaphthene	C12H10	6241.1
110	Phenyl-1-butene-2	C12H10	5702.1
111	Phenyl-1-pentene-2	C11H14	6325.4
112	Benzil	C14H10O2	6789.9
113	Benzoin	C14H10O2	6994.8
114	Furfuraldehyde	C5H4O2	2339.8
115	1-Naphthalenol	C10H8O	4960.9
116	Vinyl acetate	C4H5O2	2084.5
117	Ethylcycloheptane	C9H18	5883.2
118	1-Methyl-3-propylcyclohexane	C10H20	6285.1
119	Ethyl-1-cyclohexene-1	C8H14	5042.3
120	1-Isopropyl-1-cyclohexene	C9H16	5611.2
121	Methylenecyclohexane	C7H12	4404.3
122	Propyl benzoate	C10H12O2	5250.1
123	1-Methylcyclohexane-1,2-diol	C7H14O2	4164.5
124	Diphenylstyrene	C20H16	10,493.5
125	Amyl benzoate	C12H16O	6568.4
126	Cycloheptene	C7H12	4390.7
127	1,2-Propadiene	C3H4	1962.05
128	Eugenol acetate	C12H14O3	6268.3
129	Phenyl benzoate	C13H10O2	6321.4
130	Isopropyltoluene	C7H8	5895.9
131	1,2,4-trimethylbenzene	C5H10O	5195.3
132	Cyclohexanol	C6H12	3724.9
133	Cyclopentanone	C7H14	2852.1
134	Isopropenylbenzene	C16H32	5218.7
135	Propylbenzene	C18H36	5214.9
136	Styrene	C13H26	4375
137	Cyclohexane	C6H12	3919.8	[31]
138	Methyl cyclohexane	C7H14	4564.3
139	Ethyl cyclohexane	C8H16	5222.6
140	Propyl cyclohexane	C9H18	5875.8
141	1,1-Dimethyl cyclohexane	C8H16	5216
142	1,2-Dimethyl cyclohexane	C8H16	5216.5
143	Cyclopentane	C5H10	3291.6
144	Propyl cyclopentane	C8H16	5245.6
145	Butyl cyclopentane	C9H18	5899.9
146	Decyl cyclopentane	C15H30	9822.1
147	1,1-Dimethyl cyclopentane	C7H14	4583.3
148	1,2-Dimethyl cyclopentane	C7H14	4561.3
149	2-Hexanone	C6H12O	3754.02	[32]
150	3-Hexanone	C6H12O	3755.9
151	3,3-Dimethylbutan-2-one	C6H12O	3347.49
152	5-Nonanone	C9H18O	5715.81
153	6-Undecanone	C11H22O	7024.6
154	2-Pentanone	C5H10O	3099.41
155	3-Pentanone	C5H10O	3100.19
156	Cyclohexene	C6H10	895.27	[33]
157	1-Methylcyclopentene	C6H10	895.69
158	Cyclohexyl-benzene	C12H16	6922.73	[34]
159	Phenylbenzene	C12H10O2	6245.45
160	Cyclohexyl-cyclohexane	C12H22	7578.83
161	Ethylbenzene	C8H10	4563.9	[35]
162	Propylbenzene	C9H12	5218
163	1,2-Dimethylbenzene	C8H10	4552.6
164	1,3-Dimethylbenzene	C8H10	4551.6
165	1,4-Dimethylbenzene	C8H10	4552.6
166	Heptyl cyclohexane	C13H26	8478.5
167	Methanol	CH4O	726
168	Ethanol	C2H6O	1367.3
169	1-Propanol	C3H8O	2021
170	2-Propanol	C3H8O	2005.8
171	1-Hexanol	C6H14O	3983.8
172	1-Heptanol	C7H16O	4637.6
173	Ethan-1,2-diol	C2H6O2	1179.5
174	Propan-1,2,3-triol	C3H8O3	1655.2
175	2-Methylpropan-2-ol	C4H10O	2643.8
176	Cyclohexanol	C6H12O	3737
177	Methanal	CH2O	570.6
178	Ethanal	C2H4O	1167.1
179	Propanal	C3H6O	1820.8
180	Butanal	C4H8O	2476
181	2-Methylpropanal	C4H8O	2468.3
182	Pentanal	C5H10O	3166
183	Benzaldehyde	C7H6O	3525.1
184	2-Propanone	C3H6O	1816.5
185	2-Butanone	C4H8O	2441.5
186	3-Methylbutanone	C5H10O	3097
187	Cyclohexanone	C6H10O	3519.3
188	Phenylethanone	C8H8O	4148.7
189	Methyl methanoate	C2H4O2	972.6
190	Cyclopropane	C3H6	2091.4
191	Cyclobutane	C4H8	2720.9
192	Cycloheptane	C7H14	4598.4
193	Cyclooctane	C8H16	5266.7
194	Cyclononane	C9H18	5932.5
195	Styrene	C8H8	4395
196	Acetaldehyde	C2H4O	1167	[36]
197	Benzaldehyde	C7H6O	843.2	[37]
198	Octaldehyde	C8H16O	1218.9
199	Ethanedial	C2H2O2	205.76
200	pyrene	C16H10	1873.83
201	1,2-benzanthracene	C18H12	2144
202	perylene	C20H12	2334.6
203	methyl formate	C2H4O2	234.1
204	Heptanal	C7H14O	1062.2

and Tables S1–S10 were generated. Tables S1–S10 are provided in the Supplementary Materials. Table 2 shows the enthalpy of combustion for 204 compounds from the literature. Edwards and Maurice (2001) [17], estimated the enthalpy of combustion for five surrogates (S1–S5), containing a total of 38 compounds, as shown in Table S1. Similarly, Huber et al. (2009) [18] reported the enthalpy of combustion for two surrogates (S6 and S7), consisting of a total of five compounds, as shown in Table S2. Table S3 represents the enthalpy of combustion for six surrogates (S8–S13), containing a total of five components, as studied by Shrestha (2014) [19]. Similarly, Kalghatgi et al. (2011) [20] calculated the enthalpy of combustion for three surrogates (S14–S16), containing a total of three compounds, as shown in Table S4. Table S5 shows the enthalpy of combustion for two surrogates (S17–S18), comprising a total of nine components, as studied by Huber et al. (2010) [21]. In a same way, Eddings et al. (2005) [22] estimated the enthalpy of combustion for two surrogates (S19–S20), containing a total of six compounds, as shown in Table S6. Table S7 shows the enthalpy of combustion for two surrogates (S21, S22), comprising a total of three fuel constituents, as studied by Naik et al. (2010) [23]. Grubinger et al. (2021) [24] estimated the enthalpy of combustion for three surrogates (S23–S25), containing a total of 11 compounds, as shown in Table S8. Similarly, Table S9 shows the enthalpy of combustion for 9 surrogates (S26–S34), containing a total of 17 compounds, as reported by Xu et al. (2015) [25]. Grubinger et al. (2021) [24] reported the enthalpy of combustion for four surrogates (S35–S38), containing a total of six compounds, as shown in Table S10.

Once the dataset had been built, the variables were used as inputs to the ANN were calculated. The procedure for the calculation of 11 variables for specific fuels/chemicals, as applied to a few examples, is available in our previous publications [15,16]. After defining all compounds and mixtures, all the definitions and the corresponding enthalpies were imported to MATLAB to yield a neural network with 14 input layers, 26 hidden layers, and 1 output layer. Finally, to obtain the ANN model for predicting the enthalpy of combustion, the model needed to be trained to study and evaluate the effects and the relationships between each of these variables. Some of these effects and relationship are presented in this study.

4. Results and Discussion

The data used for developing the model was obtained from the literature and imported into MATLAB. The impact of the functional groups on the enthalpy of combustion was evaluated by means of various plots. Figure 2, Figure 3, Figure 4 and Figure 5 show that the enthalpy of combustion increases with an increase in the weight percentage of the paraffinic CH₂ groups present in various chemical classes of the fuels in the dataset. These groups represent the linearity of the hydrocarbon molecule and have been shown to have a significant impact on a number of combustion properties, such as the antiknock rating, ignition quality, flash point, etc.

However, increasing the weight percentage of olefins, alcohols, ketones, and aldehydes results in decreased enthalpy of combustion (see Figure 6, Figure 7, Figure 8 and Figure 9). However, these figures need to be considered carefully as an increase in the weight percentage of olefins is the result of increasing the CH₂ percentage. Overall, increasing the molecular weight of a compound or a surrogate leads to the increased enthalpy of combustion, as shown in Figure 10, Figure 11, Figure 12, Figure 13, Figure 14, Figure 15, Figure 16 and Figure 17. Molecular weight has a significant impact on the physical properties of pure compounds and blends and is known to influence properties, including viscosity, density, and surface tension, and also phenomena, such as heating, vaporization, droplet formation, etc. Figure 18 shows the effect of the branching index on the enthalpy of combustion for all of the collected data. It can be clearly observed that, as the branching index increases, the enthalpy of combustion increases for all classes of compounds in the data. The branching index denotes the linearity/nonlinearity of a compound by considering the carbon framework of the compound. Inclusion of the branching index as an input feature has shown to reduce the error in prediction in a large number of studies.

Subsequently, a neural network model was created and trained in MATLAB, resulting in an overall accuracy of 96.3% (see Figure 19). A regression coefficient of 0.959 was obtained when the developed model was applied to the training data, indicating that the model was able to succeed in learning the impact of the input features on the target (i.e., the enthalpy of combustion). The model was validated with approximately 15% of the dataset and yielded a regression coefficient of 0.97. The final testing of the ANN code was performed using a test set that was randomly selected by the software. The measured and predicted enthalpy of combustion values for the 40 compounds of the test set are shown in

Table 3. Comparison between the measured enthalpies and the predicted ones.

Sr. No.	Compound Name	Measured/Obtained Enthalpy of Combustion (KJ/mol)	Predicted Enthalpy of Combustion (KJ/mol)	Error (%)
1	2-Methylhexane	4811.5	4819.7	0.2
2	2,2,3-Trimethylpentane	5463.6	5285.9	3.3
3	n-1-Decene	6660.7	7139.8	7.2
4	Propylbenzene	5214.9	5187	0.5
5	1-Methylcyclopentene	3747.6	4056.3	8.2
6	Diphenylstyrene	10,493.5	10,378	1.1
7	Heptanal	4444.2	4727.3	6.4
8	Phenol	3064.3	3164.8	3.3
9	Glycerol	1661.5	1587.7	4.4
10	Phenetole	4423.1	4150.6	6.2
11	Pinacolyl alcohol	3925.2	3715.4	5.3
12	Methyl methanoate	972.6	1003.5	3.2
13	Ethanedial	860.9	941.3	9.3
14	Sr#2	6937.1	7359.7	6.1
15	Sr#3	6054.9	6282.3	3.8
16	Sr#4	5755.7	5913	2.7
17	Sr#7	6173.8	6128.9	0.7
18	Sr#8	6205.3	5798.3	6.6
19	Sr#9	6241.7	6388.5	2.4
20	Sr#10	6245.4	6225.6	0.3
21	Sr#11	6671	7016.6	5.2
22	Sr#12	6715.9	6989.8	4.1
23	Sr#13	4950.2	5258.2	6.2
24	Sr#14	3930.1	4128.6	5.1
25	Sr#15	4455.3	4271.8	4.1
26	Sr#16	6550	6076.1	7.2
27	Sr#20	4520	4739.6	4.9
28	Sr#22	4347.7	4471.2	2.8
29	Sr#23	3793.2	3704.6	2.3
30	Sr#25	6334.2	6811.6	7.5
31	Sr#26	6493.8	6804.7	4.8
32	Sr#27	6843.3	6916.7	1.1
33	Sr#28	7795.1	8041.5	3.2
34	Sr#29	7640.7	8319.1	8.9
35	Sr#30	7678.7	7683.5	0.1
36	Sr#31	7033.3	7647.5	8.7
37	Sr#32	5778.1	5646.1	2.3
38	Sr#33	7268.4	7297.5	0.4
39	Sr#34	7223.8	7643.7	5.8
40	Sr#36	9921.4	10,002	0.8

. The average percentage error obtained was 4.2%, which is near the range of experimental error.

5. Conclusions

This study presents an ANN model with 14 input layers, 26 hidden layers, and 1 output layer to predict the enthalpy of combustion of various oxygenated fuels. The ANN model was built in the MATLAB environment, and it used a detailed dataset of enthalpy of combustion values from the literature. The model was trained, validated, and tested using enthalpy of combustion data for various compounds and fuel surrogates, including numerous chemical classes: paraffins, olefins, naphthenes, aromatics, alcohols, ethers, ketones, and aldehydes. The influence of various functional groups on the enthalpy of combustion has been illustrated graphically and discussed. The procedure for the development of the model has been explained in detail. The overall precision of the developed ANN model in projecting the enthalpy of combustion was approximately 96.3%, and the average percentage error of the model, when applied against the test set, was 4.2. This value is close to that of the experimental uncertainties observed when measuring the enthalpy of combustion. This supervised machine-learning model can be used to predict the enthalpies of pure compounds and real petroleum fuels.