In order to extract the information hidden in the complex NIR spectra of paper, we need a computational rather than an analytical approach. This means that we cannot directly compare absorption at a certain wavelength with a certain material property. This is because the composition of the material leads to NIR absorption at many different wavelengths in the spectrum, which consists of a large number of overlapping bands. Therefore, we compare the absorption at many different wavelengths (actually, we usually take a whole spectrum) with the property of interest. To do so, we use the Partial Least Squares (PLS) method.
Most chemical measurements are multivariate. This means that more than one measurement can be made with a single sample. An example is spectroscopy where light absorption at hundreds of wavelengths using a single sample can be measured, which leads to very complex spectra containing a lot of information. Chemometrics helps us to understand complex spectra and extract the useful information from them. Partial Least Squares has proven to be a popular and effective approach to prediction of different chemical and physical material properties from spectra of unknown samples.
It is close to impossible to adequately describe how PLS works in short. These are the necessary work steps:
- Collect a sufficient number (at least several ten) of representative samples – a well characterized and a very diverse sample set is needed
- Perform reference analyses for the parameters of choice (e.g. pH of paper)
- Measure the spectra of the same samples
- Divide the sample set into two subsets: - One subset for method development - Another set for method validation
- Method development consists of finding the best correlation between the spectral and reference data using PLS (e.g. using commercial software) and the method development sample set. On the basis of a set of actual measurements (e.g. paper pH) and spectra of the same samples, predictions of pH are calculated and compared with actual (real) values.
- Method validation consists of checking the validity of predictions using the developed method on an independent set of samples – method validation sample set. On the basis of the developed model, pH is predicted on the basis of spectra of a new set of samples and compared with the real (actual) values.
- The higher the quality of validation, the better the PLS method.
- Chemometrics provides speed in obtaining information from complex spectral data
- It enables the extraction of chemical information from complex spectra of samples of known composition
- Successful prediction of properties from spectra of unknown samples
- Can be used for qualitative and quantitative analysis
The quality of a PLS method critically depends on:
- the number of samples
- the quality of analytically measured values
- the diversity of the calibration and validations sets
- the quality and optical configuration of the spectrometer
One should appreciate that PLS methods can only be used for characterisation of the same type of objects as used in method development. It is critical to understand that if dissimilar samples are evaluated, the results are likely to be incorrect. E.g., a method developed for rag paper will probably give wrong results for rosin-sized paper. A critical and discerning approach is important.
Chemometrics: A Practical Guide,
Chemometrics, Data Analysis for the Laboratory and Chemical Plant,
Applied Chemometrics for Scientists,
Chemometric Techniques for Quantitative Analysis,