Chemometric methods for analysis of spectroscopic data from batch process monitoring

Sammanfattning: Understanding of a process is essential for processdevelopment, optimisation and scale-up. In situ spectroscopy isan attractive measurement technique that can increase processunderstanding, since samples need not be retrieved from theprocess mixture. This thesis demonstrates how chemometricmethods can be used to extract information from data obtainedby spectroscopic monitoring of batch processes.If calibration samples are available, multivariatecalibration can be used to relate the spectroscopic data to asought property, e.g. a concentration. Partial least squarescalibration in combination with in situ infrared spectroscopyis used for this purpose in two applications. (1) Measurementsof mandelic acid concentration in aqueous solutions are used todetermine solubility curves of mandelic acid enantiomers andracemate and to monitor batch cooling crystallisations. (2) Theconcentration of the product and an intermediate is monitoredduring hydrolysis of phosphonoformic acid triethyl ester inaqueous sodium hydroxide.It is demonstrated that a self-modelling curve resolution(SMCR) method based on alternating least squares can be used toobtain concentration profiles from infrared spectroscopic datacollected during the hydrolysis of phosphonoformic acidtriethyl ester. The SMCR results agree well with those frommultivariate calibration of the product and the intermediate,but concentration profiles are extracted for several additionalspecies. No calibration samples are needed for SMCR, which isadvantageous in many cases.If a kinetic model of a reaction under study is available,it can be fitted directly to the spectroscopic monitoring data,without calculating species concentrations as an intermediatestep. A novel target testing (TT) method to carry out the fitis developed. The TT method overcomes a weakness of thetraditional non-linear least squares fitting; it allowsadditional systematic contributions to the spectral data, suchas baseline drift or peak shifts. Using infrared, Raman andUV/VIS spectroscopic data, it is demonstrated that reactionrate constants can be estimated reliably in the presence ofdrift and shifts.A typical example of when peak shifts are present is themonitoring of non-isothermal reactions by vibrationalspectroscopy. It is shown that the TT method can be used toestimate reaction activation energy and rate constant usingvibrational spectroscopic data from a single non-isothermalexperiment. The bootstrap, which is a statistical resamplingtechnique, can be used to calculate confidence intervals forthe parameters.Keywords: in situ spectroscopy, reaction monitoring,non-isothermal reaction, organic synthesis, solubility, batchcooling crystallisation, multivariate calibration,self-modelling curve resolution, kinetic model, reaction rateconstant, activation energy, target testing, non-linear leastsquares