Application Of Chemometric Methods To The Analysis Of Spectral Data In Complex Systems

The rapid update and development of instrument technology not only improves the analysis of complex system,but also generates large amounts of complicated data.Therefore,researchers have encountered the challenge of processing with these multiple data.Chemometrics is one of the most powerful and effective tool to extract useful information from abundant data for subsequently analyze.This dissertation focuses on developing qualitative and quantitative methods for spectra analysis in complex system.It mainly includes the following contents.1.Pulsed field gradient nuclear magnetic resonance(PFG NMR)is widely used to analyze the diffusion coefficient of particles.Determining the number of chemical species is crucial during analyzing procedures.PFG NMR spectra contains molecular decay and nuclear magnetic resonance information.Due to the similarity of the molecular decay information,PFG NMR data often have severely collinearity,making it difficult to calculate the number of chemical species.Thus some existing methods fails such type of data.In this thesis,a novel method is proposed by taking advantage of collinearity.Severe collinearity makes decay profiles have similar and lower frequency,whereas the noise information usually have higher frequency.The proposed method discriminates decay-profile-dominant eigenvectors from noise-dominant ones based on a novel low-and high-frequency energy ratio(LHFER).The number of chemical species is equal to the number of decay-profile-dominant eigenvectors.The proposed method was validated with both simulated and experimental data.It also has the potential to be applied to other types of data in which collinearity is fairly severe.2.Open-path Fourier transform infrared spectroscopy(OP/FT-IR)plays an important role in the study of environment and many other fields for its advantages of nondestructive,large measuring range,without sample pretreatment and so on.OP-FT/IR spectra are always complicated due to a lot of unpredictable interferences from air.Thus the quality of analyze results is determined by the credibility of spectra data.The type of background spectrum is one of the critical factors affecting the credibility.A desirable background spectrum should contain all other information except that of target analytes.This requirement can be easily satisfied in lab but not in the case of OP-FT/IR spectrometry.We systematically investigate several properties of different types of background spectra.Results show that a short-path background could significantly reduce noise in the calculated absorbance spectra and is more resistant to interferences such as wavenumber shift or resolution alteration that rises from aging hardware or misalignment.In addition,we also discussed a systematic error in quantitative analyses introduced by the short-path background and developed a effective method to correct the error.3.The spectral peak position may be changed by unstable of devices,fluctuate of temperature or other interferences.Such shift situation is common in chromatography or NMR spectra.There are several methods to solve the problems.Although the existing methods are effective in some cases,the problems encountered in OP/FT-IR spectra have been infrequently investigated.We simulate peak shifts through shifting spectra and study the influence of peak shifts in OP-FT/IR spectra on quantitative analyses.It turns out one shift point may cause about 4%of relative error in quantitative analysis for NH3 when using partial least squares(PLS)regression.Thus a method is proposed to determine and correct the peak shifts in spectra.The proposed method determines the shift point of peaks in measurements by minimizing the standard deviation of difference between measured and reference spectrum.Then correct the spectral peaks according to the determined shift points.After correction,the quantitative analysis error is reduced in most measurements.4.The establishment of calibration model is the basis of spectral quantitative analysis.However,the model built on one instrument may be invalid when applied on other instruments.Rather than building new calibration model,transfer the established model is more convenient and less time-consuming.In this dissertation,we establish two methods to perform quantitative model transfer between different OP/FT-IR instruments.Method A is first to calculate reference concentrations of spectra with the high and low resolution conversion algorithm,and then correct the predicted concentrations by the slope bias correction algorithm or quadratic polynomial fitting algorithm.Method B is to select proper standardization sample set by placing two instruments side by side,and then correct spectra by combing wavelet transform baseline removal algorithm and piecewise direct standardization algorithm or spectra space transformation algorithm.Both of two methods could improve the predictive ability of established model when applied on another instrument.Compared to each other,method A is simpler and more convenient,but the transfer parameters need to recalculate when datasets are measured in different time.Method B requires more parameters but leads to better stability.