Predicting Ignition Quality of Oxygenated Fuels Using Artificial Neural Networks

Abstract

Artificial intelligence based computing systems like artificial neural networks (ANN) have recently found increasing applications in predicting complex chemical phenomena like combustion properties. The present work deals with the development of an ANN model which can predict the derived cetane number (DCN) of oxygenated fuels containing alcohol and ether functionalities. Experimental DCN's of 499 fuels comprising of 116 pure compounds, 222 pure compound blends, and 159 real fuel blends were used as the dataset for model development. DCN measurements of sixty new fuels were carried out in the present work and the data for the rest was collected from the literature. Fuel chemical composition expressed in the form of eight functional groups namely paraffinic CH3 groups, paraffinic CH2 groups, paraffinic CH groups, olefinic -CH=CH2 groups, naphthenic CH-CH2 groups, aromatic C-CH groups, alcoholic OH groups and ether O groups, along with two structural parameters namely, molecular weight and branching index (BI), were used as the ten input features of the model. The qualitative and quantitative determination of functional groups present in real fuels was performed using 1 H nuclear magnetic resonance (NMR) spectroscopy. A robust ANN methodology was followed to prevent over fitting using a multilevel grid search and genetic algorithm. The final developed model with two hidden layers was tested with 15 % of randomly generated unseen points from the dataset and a regression coefficient (R 2 ) of 0.992 was observed between the experimental and predicted DCN values. An average absolute error of 0.91 obtained from the test set indicates that the developed ANN model is successful in predicting the DCN of oxygenated fuels and captures the dependence of the fuel's ignition quality (i.e. DCN) on its constituent functional groups.

Predicting Sooting Propensity of Oxygenated Fuels Using Artificial Neural Networks

“…This shows the ability of ANNbased models to predict complex chemical phenomena, such as sooting propensity. This also supports the functional group-based approach employed here, which reiterates the ability of the functional groups to predict fuel properties, as previously demonstrated for cetane number [47,48] and octane number [44]. Another advantage of the functional group approach is the ability to predict the YSI of real fuels, which most other models reported in the literature are not designed to do.…”

“…A number of parameters, such as batch size, epochs, number of hidden layers, and number of neurons in each hidden layer, were optimized to yield the final ANN model. Nearly 260 iterations were carried out and the best performance was ob- ously demonstrated for cetane number [47,48] and octane number [44]. Another ad-vantage of the functional group approach is the ability to predict the YSI of real fuels, which most other models reported in the literature are not designed to do.…”

“…The input features of the model consist of the fuel functional groups, molecular weight (MW), and a structural parameter called the branching index, which quantifies the branching in a molecule. The combination of these input parameters adequately represents the molecular structure of a pure compound, mixture, or real fuel, and these have been successfully used to predict derived cetane number [47,48], octane number [44], and formulate surrogates for a number of fuels using the minimalist functional group (MFG) approach [49][50][51].…”

A Methodology for Designing Octane Number of Fuels Using Genetic Algorithms and Artificial Neural Networks

“…Petroleum gasoline is composed of paraffin, olefins, naphthene, and aromatic compounds that can be disassembled into a finite number of functional groups which define the fuel’s properties. The functional groups in fuel have been used to predict the cetane number of diesel fuel and the sooting propensity of a wide class of neat compounds and also to formulate surrogates for gasoline, diesel, and jet fuels. − Along with functional groups, two additional structural parameters called branching index (BI) and molecular weight (MW) have been included as input features in the ANN model. After each generated mixture is converted into the input features (i.e., functional groups, BI and MW), the ANN model predicts the RON and MON of each.…”

Identification and Quantification of Hydrocarbon Functional Groups in Gasoline Using 1H-NMR Spectroscopy for Property Prediction

“…As is the case with RON and MON measurements, experimental measurement of CN/DCN is expensive, and there are multiple studies [33,38,39] reported in the literature that have developed models for CN/DCN prediction. Recently, an artificial neural network-based technique [40] was developed for predicting of the DCN of fuels containing oxygenated classes like alcohols and ethers by using the functional groups as input parameters. The DCN prediction model was developed by using 499 fuels, and the model consisted of two hidden layers with 442 nodes in the first layer followed by 290 nodes in the second.…”