Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy
Calibration sets selection strategy for the construction of robust PLS models for prediction of biodiesel/diesel blends physico-chemical properties using NIR spectroscopy
Graphical abstract
Introduction
Using near infrared (NIR) spectroscopy for refining processes monitoring is an attractive analytical resource for petroleum industry because it can contribute to reaching the required quality at optimal cost and to increasing the conversion levels and feedstocks flexibility [1]. Due to its features of predominantly CH stretches measurement, ease of use, fast response and non-destructiveness of samples, petroleum has been one of the fields of use of NIR spectroscopy since its industrial application started [2].
The nature of the vibrational rotations of petroleum samples molecules in the NIR spectral region generates bands usually broad and overlapped, which makes difficult to visually take advantage of them. To overcome this situation, application of NIR spectroscopy has been combined with mathematical and statistical tools since the beginning of the second half of the past century. The specific analysis of hydrocarbon mixtures done by Rose in 1938 [3], was probably the first report of use of this combined approach in the crude oils industry [4]. With the impulse provided by informatics developments in 1970, application of such mathematical and statistical tools to spectroscopic and chemical data in general, became into what is nowadays known as chemometrics.
Applications of NIR spectroscopy and chemometrics to petroleum industry have been described for a wide variety of samples and purposes. From the whole crude [5] to finished products [6], [7], [8], [9], [10], [11], [12] as well as additives have been analysed [13], both for qualitative [14] and quantitative motives. Quantitative studies have been mainly related to the use of chemometrical multivariate methods for prediction of physico-chemical properties of intermediate product streams and finished products. In these cases, multivariate methods relate the NIR spectral data with reference values of target parameters by mean of a mathematical model termed calibration. The error of the resulting predictions of the models relative to reference values is generally expressed by the root mean square error of prediction (RMSEP).
Specially for diesel fuels -fraction of petroleum atmospheric distillation 200–300 °C [15]-, numerous efforts on quantitative analysis have been done using NIR and the multivariate method of partial least squares (PLS) regression [9], [16], [17], [18]. Additionally, later politics and economic factors shifted the attention of energy companies to alternative fuels [19], [20], [21], [22], these analytical tools has been also applied to biodiesel/diesel blends. Biodiesel fuel is a mixture of fatty acid alkyl esters produced by transesterification of edible or non-edible vegetable oils and alcohol, with or without the presence of a catalyst [23], [24]. The reason for such interest in these blends has been their successful performance in transportation engines [23]. Researches in analysis of biodiesel/diesel blends based on NIR spectroscopy and diverse multivariate methods have produced interesting results [25], [26], [27], [28]. Particularly, Alves et al. have demonstrated, the simplicity of using PLS versus others multivariate methods, as e.g. Support Vector Machine (SVM) [28]. The PLS method has been used for the prediction of the most relevant specification properties for this kind of samples: alternative fuel content, fatty acid methyl esters (FAME) content, density, flash point, viscosity, pour point, cetane index, temperatures at diverse distillation percentages and sulphur [28], [29], [30], [31], [32].
Nevertheless, an important challenge that still has to be faced is the eventual lack of robustness of the models when they are installed in plant [33]. Due to setting up an optimum NIR multivariate model is costly, once it has been developed it is expected to be valid for a long period of time [11]. Achieving such performance depends on diverse chemometrical calibration settings. Some of them, as prediction intervals, infrared spectral region, calibration algorithm and spectral pre-processing procedure have been studied for petrochemicals [12], [34], [35]. In spite of this, the selection of the samples to be included into the calibration sets remains as a factor that is not easily handled during the construction of PLS models for biodiesel/diesel blends.
Calibration sets are the group of samples employed to calculate the calibration models, so they should be designed appropriately [11]. They are supposed to cover both the concentration range as well as all the other sources of variability of the studied system [36]. Considerations regarding the concentration ranges have been taken into account for the preparation of calibration sets of models for renewable diesel, biodiesel and diesel blends, by mean of design of experiments [28], [37]. Even though, specific considerations for these systems, as the geographical origin and therefore, the hydrocarbon composition of diesel fuels [38] has not been included in any previous proposal of calibration sets selection for this kind of samples. With the aim of contributing in this particular aspect, we present a combination of the typical sample selection criteria (viz. spectral information and reference values) with the crude oil compositions to construct specific calibration sets for each target parameter, by mean of principal components analysis (PCA) and the Kennard-Stones algorithm. This methodology has been applied to on-line simultaneous determination of seven physico-chemical properties of biodiesel/diesel blends.
Section snippets
NIR Operating Conditions
Near-infrared spectra were recorded on a Bartec Benke GmbH FT-NIR Matrix-F spectrophotometer. The spectra were on-line acquired in the transmittance mode, using a probe with an effective path length of 2 mm and a flow cell with two channels, one for background and one for sample. This probe was inserted in the diesel recycling stream from the blending line. All the spectra were recorded at the temperature of the process, 37 ± 2 °C.
Each spectrum was acquired in 10 scans of the 1000–2200 nm wavelength
Calibration Sets Selection
As stated above, because of hydrocarbon composition result in slight differences between NIR spectra, we used the typical sample selection criteria -viz. spectral information and reference values- in addition to the crude oil composition, to construct a specific calibration set for each target parameter, i.e. for each individual model.
Conclusions
In this work, a strategy for calibration sets selection of biodiesel/diesel samples based on PCA and the Kennard and Stones algorithm was presented. Its performance was assessed by the construction of PLS models using NIR spectroscopy for the on-line monitoring of seven physico-chemical properties of the blends: density, cetane index, FAME content, cloud point, T95, flash point and sulphur.
PLS models exclusively constructed from on-line recorded NIR spectra allowed the accurate determination of
Acknowledgements
The authors are grateful to Spain's Ministry of Economy and Competitiveness (MINECO) for funding this research within the framework of Project CTQ2012-34392.
References (42)
- et al.
Characterization of petroleum using near-infrared spectroscopy: quantitative modeling for the true boiling point curve and specific gravity
Fuel
(2007) - et al.
Comparison of PLS algorithms in gasoline and gas oil parameter monitoring with MIR and NIR
Chemom. Intell. Lab. Syst.
(2005) - et al.
Comparison of near-infrared and mid-infrared spectroscopy for the determination of distillation property of kerosene
Vib. Spectrosc.
(1999) - et al.
Comparison of linear and nonlinear calibration models based on near infrared (NIR) spectroscopy data for gasoline properties prediction
Chemom. Intell. Lab. Syst.
(2007) - et al.
Near-infrared spectroscopy and multivariate calibration for the quantitative determination of certain properties in the petrochemical industry
TrAC Trends Anal. Chem.
(2002) - et al.
Using principal component analysis to find the best calibration settings for simultaneous spectroscopic determination of several gasoline properties
Fuel
(2008) - et al.
On the evaluation of the performance of asphaltene dispersants
Fuel
(2016) - et al.
Near-infrared (NIR) spectroscopy for motor oil classification: from discriminant analysis to support vector machines
Microchem. J.
(2011) - et al.
A low cost short wave near infrared spectrophotometer: application for determination of quality parameters of diesel fuel
Anal. Chim. Acta
(2010) Review on analysis of biodiesel with infrared spectroscopy
Renew. Sust. Energ. Rev.
(2012)
Variable selection in near-infrared spectroscopy: benchmarking of feature selection methods on biodiesel data
Anal. Chim. Acta
Critical review on analytical methods for biodiesel characterization
Talanta
Perspectives on biodiesel as a sustainable fuel
Renew. Sust. Energ. Rev.
Inedible vegetable oils and their derivatives for alternative diesel fuels in CI engines: a review
Renew. Sust. Energ. Rev.
Neural network (ANN) approach to biodiesel analysis: analysis of biodiesel density, kinematic viscosity, methanol and water contents using near infrared (NIR) spectroscopy
Fuel
Biodiesel content determination in diesel fuel blends using near infrared (NIR) spectroscopy and support vector machines (SVM)
Talanta
A comparative study of diesel analysis by FTIR, FTNIR and FT-Raman spectroscopy using PLS and artificial neural network analysis
Anal. Chim. Acta
A comparison of methods for estimating prediction intervals in NIR spectroscopy: size matters
Chemom. Intell. Lab. Syst.
Quantification of conventional and advanced biofuels contents in diesel fuel blends using near-infrared spectroscopy and multivariate calibration
Fuel
Physical properties analysis in modern petroleum industry
Recent advances in the use of near-IR spectroscopy in the petrochemical industry
Cited by (40)
Rapid prediction method of ZIF-8 immobilized Candida rugosa lipase activity by near-infrared spectroscopy
2023, Spectrochimica Acta - Part A: Molecular and Biomolecular SpectroscopyRapid determination of the key temperatures in diesel distillation process based on near-infrared spectroscopy
2023, Infrared Physics and TechnologyClassification and determination of sulfur content in crude oil samples by infrared spectrometry
2022, Infrared Physics and TechnologyCitation Excerpt :In this study, a new and simple approach for parallel quantification and qualification of the sulfur content present in crude oils using ATR-FTIR spectroscopy associated with chemometric methods is shown. This new and simple approach will help to estimate the amount of sulfur content in crude oils and classify crude oils based on sulfur content into sweet and sour crude oil, thus to allow better optimization of desulfurization conditions during fuel production [27–42]. The chemometric approaches used for the determination of sulfur content in crude oils are partial least squares regression (PLS-R) and support vector machine regression (SVM-R).
Improvement of NIR prediction ability by dual model optimization in fusion of NSIA and SA methods
2022, Spectrochimica Acta - Part A: Molecular and Biomolecular SpectroscopyCitation Excerpt :Partitioning a representative calibration set is crucial. A representative calibration set can often make the model have appreciated prediction accuracy, and it is easy to update the model [10–11]. The sample division algorithms based on the sample distances are simple and fast, and can provide a reasonable and representative sample set to a certain extent [12].
Near Infrared Spectroscopy: A useful technique for inline monitoring of the enzyme catalyzed biosynthesis of third-generation biodiesel from waste cooking oil
2022, FuelCitation Excerpt :Besides, it avoids the need for sample withdrawal when used inline, waste production, and the need for complex pre-treatments of samples with solvents or other chemicals, all of which makes it a safe, clean, energy-saving choice fully compliant with the principles of green chemistry [28]. NIR spectra are complex and possess broad overlapping bands that require special mathematical procedures to accurately interpret spectra and understand the results, such as principal component analysis (PCA) or partial least-squares (PLS) regression [29]. NIR spectroscopy has so far been successfully used by the biodiesel industry to assess the quality or properties of biofuel/diesel blends [29,30], and also for inline monitoring of chemically catalyzed transesterification reactions [20,31,32].