is an analytical instrument made by the principle of selective absorption of infrared rays by gases or liquids, which has the characteristics of high sensitivity, fast response time, wide analytical range, good selectivity and strong anti-interference ability, etc. It is widely used in petrochemical metallurgy and other industrial production.
Near-infrared spectroscopy (NIR)
is one of the fastest growing and most impressive analytical techniques since the 1990s. With the further application and development of NIR analysis methods, it has gradually gained universal acceptance. In recent years, with the development of chemometrics, optical fiber, and computer technology, online NIR spectroscopy
is being applied to many fields including agriculture and animal husbandry, food, chemical, petrochemical, pharmaceutical, tobacco, etc. at an astonishing speed, providing a very wide use space for scientific research, teaching and production process control.Near Infrared (NIR)
is an electromagnetic wave between visible light (VIS) and mid-infrared light (MIR). The wavelength range of the NIR spectral region defined by ASTM is 780~2526nm (12820~3959cm-1). NIR spectra
are generated when molecular vibrations jump from the ground state to higher energy levels due to the non-resonant nature of molecular vibrations and record the multiplicity and ensemble absorption of X-H (X = C, N, O) vibrations of hydrogen-containing groups. The NIR
absorption wavelengths and intensities of different groups (e.g. methyl, methylene, benzene ring) or the same group in different chemical environments are significantly different. NIR spectra are rich in structural and compositional information and are well suited for the measurement of the composition and properties of hydrocarbon organic substances. However, in the NIR region, the absorption intensity is weak, the sensitivity is relatively low, and the absorption bands are wide and heavily overlapped. Therefore, it is very difficult to rely on the traditional method of establishing a working curve for quantitative analysis, however, the development of chemometrics has laid the mathematical foundation for the solution of this problem. It works on the principle that if the composition of a sample is the same, its spectrum is also the same, and vice versa. If the correspondence between the spectrum and the parameter to be measured is established (called an analytical model), then the desired data on the quality parameters can be quickly obtained by measuring the spectrum of the sample and the above correspondence. The analytical method consists of two processes: calibration and prediction.
In the calibration process, a certain number of representative samples are collected (generally more than 80 samples are required). While measuring their spectrograms, measurements are made using relevant standard analytical methods as needed to obtain various quality parameters of the samples, called reference data. The spectra are processed by chemometrics and correlated with the reference data so that a one-to-one correspondence mapping is established between the spectrograms and their reference data, which is usually called a model. Although the number of samples used for model building is limited, the models obtained by chemometric processing should be highly generalizable. The calibration methods used for model building vary depending on the relationship between the sample spectra and the properties to be analyzed. Commonly used methods include multiple linear regression, principal component regression, partial least squares, artificial neural networks, and topological methods. Obviously, the wider the range to which the model applies, the better. However, the size of the range of the model is related to the calibration method used to build the model, to the property data to be measured, and also to the range of analytical accuracy required to be achieved by the measurement. In practice, model building is achieved with chemometric software and has strict specifications.
In the prediction process, the spectrogram of the sample to be measured is first measured using a near-infrared spectrometer, and then the model library is automatically searched by the software to select the correct model to calculate the mass parameters to be measured.
The main sources of detection errors in NIR spectrometers
are as follows.
a. The number of samples used to establish the calibration equation and the test equation.
b. The particle size and distribution of the sample.
c. The testing environment and the temperature of the sample.
d. The error due to chemical analysis.
An important characteristic of NIR analysis technology is the set of technology itself, which means that the following three conditions must be available at the same time.
a. NIR spectrometer with a long-term stable performance of each item is the basic requirement to ensure good reproducibility of data.
b. Full-featured chemometric software, which is an essential tool for model building and analysis.
c. A model that is accurate and has a wide enough range of applications.
The combination of these three conditions can be truly useful for the user. Therefore, it is important to have sufficient knowledge of the usability of the model provided by the instrument when purchasing it.
The fast speed of NIR analysis is due to the fast speed of spectral measurements and the fast speed of computer calculation of the results. However, the efficiency of NIR analysis depends on the number of models the instrument is equipped with. For example, if a spectrogram is measured with only one model, only one data can be obtained. If 10 data models are built, then 10 analytical data can be obtained simultaneously from just one measured spectrum.