Octane Prediction from Infrared Spectroscopic Data

Abstract

A model for the prediction of research octane number (RON) and motor octane number (MON) of hydrocarbon mixtures and gasoline-ethanol blends has been developed based on infrared spectroscopy data of pure components. Infrared spectra for 61 neat hydrocarbon species were used to generate spectra of 148 hydrocarbon blends by averaging the spectra of their pure components on a molar basis. The spectra of 38 FACE (Fuels for Advanced Combustion Engines) gasoline blends were calculated using PIONA (Paraffin, Isoparaffin, Olefin, Naphthene, and Aromatic) class averages of the pure components. The study sheds light on the significance of dimensional reduction of spectra and shows how it can be used to extract scores with linear correlations to the following important features: molecular weight, paraffinic CH3 groups, paraffinic CH2 groups, paraffinic CH groups, olefinic-CH=CH2 groups, naphthenic CH-CH2 groups, aromatic C-CH groups, ethanolic OH groups, and branching index. Both scores and features can be used as input to predict octane numbers through nonlinear regression. Artificial Neural Network (ANN) was found to be the optimal method where the mean absolute error on a randomly selected test set was within the experimental uncertainty of RON, MON, and octane sensitivity.

Explainable Predictive Modeling for Limited Spectral Data

“…There is an extensive literature in the petrochemical industry and the chemometrics discipline that use spectroscopic data to predict fuel Cetane or Octane Number. The data is usually collected using the following techniques: Gas Chromatogramy-Mass Spectroscopy [25,53], Fourier-transform Infrared (FTIR) Spectroscopy [54,55,56,57], Quantitative Structure Property Relationships analysis [58], Fatty Acid Methyl Esters (FAME) composition analysis [59], Nuclear Magnetic Resonance (NMR) Spectroscopy [22], as well as Near Infrared (NIR) [49,26,60,61,62], Mid-Infrared (MIR) and Raman Spectroscopy [61,63,57].…”

“…Additionally, fewer topics are dedicated to extracting or explaining (see Interpretability column in Table 1), spectroscopic features. To this end, they mainly applied model-based feature selection techniques to identify and remove noisy features in order to improve prediction accuracy and computational efficiency [64,61,62,65,25,66,57,26,60,67]. In another body of work, popular feature selection methods such as PCA [68], and PLS [69] were employed to discover the correlation between decomposed fuel spectra and fuel sample clustering results [54], help isolate certain chemical groups responsible for the deviation in predicted values [49,25], and correlate certain spectra regions of pharmaceutical tablets to the concentration of antiviral drug [62].…”

“…Balabin et al has also explored the deployment of NN and SVR for analytical chemistry and concluded that not only SVR outperforms PCA and PLS regression methods [67], but that SVR also performs similarly to NN, with SVR having the advantage in producing a more generalized model capable of efficiently dealing with non-linear relationships [74]. Similarly, NN is also capable of capturing unique spatial features and have been shown to perform well on spectroscopy data analysis [41,54,25,53], given a wealthy amount of data. In our work, we investigate the performance of LR, SVR, and NN models to process highdimensional spectra data for prediction and explainability effort.…”

“…While the body of research on spectroscopic analysis is extensive, to the best of our knowledge, there is no single comprehensive work focused on developing both explainable and scalable predictive models for spectroscopic data. Most publications in the field either solely focus on obtaining prediction, for example, applying popular ML methods for octane prediction using infrared spectroscopy [25], or the use of common feature elimination techniques [26] to improve the prediction accuracy. The implementation of explainable black-box models is limited to interpreting functional nearinfrared spectra data in developmental cognitive neuroscience using simple multi-variate analysis [27] and using Local Interpretable Model-Agnostic Explanations (LIME) [28] on optical emission spectroscopy of plasma [29].…”

Optimized Artificial Neural Network for Evaluation: C4 Alkylation Process Catalyzed by Concentrated Sulfuric Acid

“…As the number of motor vehicles increases, the gasoline quality of vehicles is of great significance to the environment. 1 Alkylate oil is considered as a typical clean fuel blending component with a low-sulfur content and high research octane number (RON), 2 which is obtained from the alkylation process of isobutane and butene catalyzed by strong acid. Currently, large-scale industrial plants still adopt concentrated sulfuric acid (H 2 SO 4 ) as the mainstream alkylation catalyst.…”