You are currently viewing: Articles


Corresponding factors influencing crude oils assay using low-field nuclear magnetic resonance

In this study, the main factors influencing the measurement by means of the off-line low-field H NMR in the lab were discussed base on a robust calibration model established by the PLS algorithm using 255 crude oil samples.

Feng Yunxia, Chu Xiaoli, Xu Yupeng and Tian Songbai
Research Institute of Petroleum Processing
Viewed : 1499
Article Summary
The preheating temperature had a great influence on the viscosity of oil samples and the resolution of spectral analysis. The repeatability of spectral measurements was impacted by the metal and wax content of the oil samples. For the case of high wax content oils, the wax species began to crystallize in the course of determination that could affect the repeatability of spectral measurements.

These factors have evidenced why the preheating devices and filter unit are necessary when low field NMR system is used in the online analysis process. The investigation is very important for the on-line application of the low field NMR.

In recent years, with crude oil prices increasingly spiralling, refineries are paying great attention to process analysis technology in order to effectively monitor the production process and save raw materials1-2. The physical and chemical properties of crude oils are of major importance to the petroleum refining processes. However, the traditional evaluation methods are often expensive, time-consuming, and requiring a large amount of samples and poisonous solvents. Hence, it is necessary to develop a fast method for the real-time detection in process analysis.

Nuclear magnetic resonance (NMR) is primarily used as an analytical technique for elucidation of molecular structure and identification of chemical species in a wide range of fields such as food3-4, agriculture5, petrochemicals6-7, medical science8, and bioscience9. The more recent applications of NMR are associated with various multi-variate data analysis methods to exploit the chemical and physical properties of many complex multi-component mixtures such as crude oils, asphaltenes, and base oils10.

In quantitative analysis, the chemometric regression techniques, for instance the partial least squares (PLS), the principal component regression (PCR) and the artificial neural network (ANN), are routinely used to establish the correlation between NMR spectra and the properties of complex mixtures11-12. The main advantage of this correlation is that the assessment of several properties can be achieved simultaneously based on the unique NMR spectra13-15.

With the development of new technologies using permanent magnets, the low magnetic field NMR instruments open up new possibilities for process analysis16. Compared with the high field NMR, the application of low magnetic field NMR in various industrial fields has been promoted thanks to its better stability, lower cost, faster analysis, and the non-destructive nature of measurements.

Small, specified low-field (less than 60 MHz) NMR systems have been used in bypass systems to perform the online analysis and reaction monitoring17-19. Since the 1990s the earliest on-line NMR analysers were provided to petrochemical plants and utilized for controlling the production process20. More industrial applications of NMR with chemometric methods to predict all kinds of petroleum properties have already been realized21-22.

This study focuses on discussing the important factors of low-field 1H NMR in the lab, including the effect of viscosity of crude oils on the resolution of spectra and the factors influencing the repeatability of measurements. The factors influencing the test repeatability, such as the metal and wax contents in crude oils, were discussed in this paper.

Low field NMR spectroscopy
All 1H NMR spectra were recorded using a Qualion 58 MHz NMRS230022 spectrometer that is based on a permanent magnet NMR technology. The 1H NMR measurements were performed at a sampling rate of 4 346 Hz, a band width of 2 173 Hz, and a number of loop scanning equating to 8. After storing the free induction decay (FID) signals, the final spectra were obtained through a number of steps such as zero fill, phase correction, removal of direct current (DC), fast Fourier transform (FFT), and baseline correction. All the steps could be completed automatically.

A total of 255 crude oil samples were analysed in the laboratory to obtain the values of chemical and physical properties for each sample according to various industrial or analytical standard test methods.

In the case of low field 1H NMR, the liquidity of oils is very important for the measurements. All oil samples that are introduced into the analyser must be in an entirely liquid form so that the distribution of protons in the entire molecule can be observed. Each oil sample is preheated at 42°C in the water bath before scanning. Figure 1 shows the 1H NMR spectra of 255 crude oil samples.

Multivariate analysis
Multivariate analysis was performed using the software developed independently by the Analytical Chemistry Department of the Research Institute of Petroleum Processing (RIPP). PLS regressions were performed with leave-one-out cross validation. The methods of building the calibration and validation models in this work were the same as those described previously by Li, et al.23-24

All 255 samples were randomly divided into a training set and a validation set using the Kernel-Stone method, in which the training set consisted of 220 samples and the remaining 35 samples were contained in the validation set. The 220 samples in the calibration set were subjected to a PLS regression, in which 4 physicochemical variables (density, Conradson carbon residue, resins content, and asphaltenes content) were included.

The root mean square error of calibration (RMSECV) is used to assess the calibration performance. The calibration model for the oil properties is obtained with the optimum number of latent PLS factors; it is selected based on the predicted residual sum of squares (PRESS) by the cross-validation results. On the other hand, the standard error of prediction (SEP) for the validation set represents an evaluation of the predicted performance of the quantitative model. These values are calculated by the following equations:

where n and m are the amount of oil samples included in the calibration and validation sets, respectively, while yi,actual and yi,predicted are the properties of crude oil samples that are measured by the reference and NMR methods, respectively.

Results and Discussion
The results of calibration and validation of variables models
Table 1 shows the results of calibration and validation of the model. These values are in the required range of accuracy, which can be effectively used for the fast estimation of oil samples.

Effects of viscosity on resolution
Crude oil samples with different viscosity values were tested in order to observe the effect of viscosity of crude oils on the test results. In this study, some heavy and light oil samples were preheated to different temperatures by a water bath, and the heavy oil samples were preheated by an oven when the required temperature was higher than 100°C. The oil property items were predicted by the models which are referred to in Section 3.1. The values of standard deviation (SD) indicated the difference in oil property at various temperatures.

Sponsor : 

Current Rating :  1

Add your rating:

Your rate: 1 2 3 4 5