Elsevier

Fuel

Volume 85, Issues 10–11, July–August 2006, Pages 1396-1402
Fuel

The prediction of clay contents in oil shale using DRIFTS and TGA data facilitated by multivariate calibration

https://doi.org/10.1016/j.fuel.2005.12.025Get rights and content

Abstract

The prediction of clay content in oil shale is important for the optimisation of oil shale processing conditions and process feasibility. The multivariate calibration technique of partial least squares regression (PLSR) was implemented in order to predict clay content in oil shale samples taken from the Stuart oil shale deposit, Queensland, Australia. The calibration data used were the diffuse reflectance infrared Fourier transformed spectroscopy (DRIFTS) spectra of 34 oil shale samples. DRIFTS data from another set of 20 oil shale samples were used for model validation. The data pre-processing includes the use of derivatives facilitated by the Savitsky-Golay nine-points’ method. A four components model was constructed and it showed a root mean square error of calibration (RMSEC) of 4.79% and a root mean square error of prediction (RMSEP) of 4.35%. TGA data sets were also used to construct a calibration model, which produced less accurate results than DRIFTS. DRIFTS, when combined with multivariate calibration, provided an accurate in situ method of evaluating clay content in oil shale. Clay content measured using XRD was used as a reference.

Introduction

The constant decrease in petroleum sources and the increase in oil prices encourage the search for alternative energy sources. Finding substitute sources for petroleum-based products has been the main motive for extensive studies. The extraction and production of oil from oil shale is available as an alternative in many places in the world such as Australia. The estimated oil that can be theoretically produced worldwide is about 2.6 trillion barrels. Australia has the third largest oil shale deposit in the world of 32.4 billion tonnes of shale (about 220 billion barrels) with proven recoverable reserves of 1.725 billion tonnes of oil (about 11.7 billion barrels). Researchers are investigating ways to increase the feasibility of oil generating processes using oil shale as a feedstock.

Oil shale is a sedimentary rock containing a complex mixture of minerals and an organic substance called kerogen. Kerogen can be converted into oil through a retorting process. The accurate determination of the mineral content of oil shale is important for the selection and optimisation of retorting process conditions. Clay minerals act as a catalyst for the coking reactions of kerogen. These reactions happen on the surfaces of the minerals and go together with the oil pyrolysis reaction, leading to a decrease in the conversion of kerogen to oil [1], [2], [3], [4]. Accurate prediction of the clay minerals content is useful knowledge for operators and it potentially helps limiting its negative effect on oil yield. The amount of clay minerals present also affects process heat balance, trace element emission, gas composition, oil losses and the recovery of by-products [4]. The investigation of clay in this study is based on the collective results of clay minerals of smectite, kaolinite and illite.

Diffuse reflectance infrared Fourier transformed spectroscopy (DRIFTS) is a cheaper, faster and non-destructive way of evaluating clay minerals and oil content of oil shale [5], [6]. Combining multivariate calibration and DRIFTS by generating a model to predict mineral contents from oil shale samples based on spectral data has the potential to facilitate further the processing of oil shale. Although large numbers of variables are generated from DRIFTS spectra by using multivariate calibration methods, emphasis is often concentrated on just a few major ones.

Many researchers have reported the IR assignment of clay minerals in the MID-IR region. Kaolinite is associated with the structural OH group band near 3700 cm−1 [5]. Clay components are reported to be associated with strong bands near 1040 and 470 cm−1 [7]. Strong intensities in the region between 450–1000 cm−1 are also linked to clay minerals [5], [7]. The band at 800 cm−1 is the vibration of the Si–O group and it is assigned to kaolinite and illite [8]. The absorption band near the region at 900 cm−1 is assigned to the OH–Al group band that relates to montmorillonite, kaolinite and illite [8]. The broad absorption band near the 500 cm−1 region is assigned to the Si–O–Al group, which relates to montmorillonite, kaolinite and illite [8].

The best indication regarding clay content from TGA data is the weight loss in the 40–200 °C temperature range [9], [10]. The weight loss in this region could be attributed to the loses of moisture, structural water from clay minerals and the decomposition of other minerals such as nahcolite and dawsonite [10], [11]. It was also reported that the measurement of weight loss due to the release of structural water in clay minerals could be stretched to 550°C [12]. However, the relatively larger amount of kerogen decomposition of oil shale samples in the 400–600 °C temperature region prevents the accurate detection of free structural water.

Multivariate calibration methods have been used in the past for predicting concentrations of unknown samples using the responses and variables of known samples. These methods proved successful with spectroscopic data in previous studies [13], [14]. Prediction models can be obtained using regression analysis such as principal components regression (PCR) and partial least square regression (PLSR). Both methods are widely used in generating models from spectroscopic data. PLSR assumes a linear relationship between variables and response(s) and it also compensates for any nonlinearity by including more components. Furthermore, PLSR can handle noisy data with numerous variables [15], [16]. Data pre-processing such as smoothing and/or derivatives are normally applied to spectral data to reduce or eliminate the information that is not related to the response. The main information (variance) in the data set is identified and the variables representing those variances are nominated as main components. Models generated using PLSR have already shown high success for interpretation and improved prediction [13]. The PLSR calibration method has also been previously used in predicting oil yield from Australian oil shale using DRIFTS in the near IR region [17].

This study aims to develop a model using a non-orthogonal PLSR method to predict clay content in oil shale samples using DRIFTS measurements in the mid-IR region. The interpretation of model components is also attempted.

Section snippets

Samples

Oil shale samples were obtained from South Pacific Petroleum (SPP) Co., Australia from their Stuart oil shale deposit in Queensland with measured modified Fisher assay (MFA) values in L T−1 and mineral content by XRD. The oil shale samples were mixed thoroughly and 3–4 g were separated from the mix and ground in an orbital-steel ring binder for 30 s. A total of 54 samples were used in this study. 34 samples were selected to cover the range of clay contents, and used for the construction of a

Results and discussion

A typical oil shale spectrum is shown in Fig. 1a. The XRD and MFA data for 54 oil shale samples as received from SPP Company are shown in Table 1, Table 2. Samples are randomly selected to represent the calibration and the validation (prediction) data sets. Table 1 shows the clay composition of the shale samples that were used to construct the calibration model. Table 2 shows the clay components of the shale samples that were used to construct the prediction model.

Infrared spectroscopy is

Conclusions

Multivariate calibration coupled with IR spectroscopy was useful and efficient for the prediction of clay content in oil shale samples. IR data proved to be more accurate in producing calibration models than TGA data. A four-factor model is adequate for producing low RMSEP values for clay content in oil shale. Full spectral data produced more accurate calibration models than the data extracted at different spectral regions

Acknowledgements

The authors wish to thank the Southern Pacific Petroleum and the Australian Research Council for their financial assistance.

References (22)

  • K. El harfi et al.

    J Anal Appl Pyrol

    (2000)
  • A.J. Berkovich et al.

    Thermochim Acta

    (2000)
  • J.H. Patterson et al.

    Fuel

    (1990)
  • J.H. Patterson

    Fuel

    (1994)
  • P.R. Solomon et al.

    Fuel

    (1980)
  • R.W. Snyder et al.

    Fuel

    (1983)
  • S. Bhargava et al.

    Fuel

    (2005)
  • A.K. Burnham et al.

    Fuel

    (1983)
  • S. Sato et al.

    Fuel Process Technol

    (1997)
  • S. Wold et al.

    Chemometr H Intell Lab Syst

    (2001)
  • S. Wold et al.

    Chemometr Intell Lab Syst

    (2001)
  • Cited by (0)

    View full text