Zur Seitenansicht


Zusammenfassung (Englisch)

In industrial process analytics near-infrared spectroscopy is commonly used in continuous and batch processes under application of probes. However the used NIR probes in these proceedings exhibit undesired effects, such as variance in commodities, matrix effects and fouling.

The technical definition of fouling describes depositions on surfaces, which affect the performance and durability of the used equipment. These depositions on NIR probes have a significant influence on established multivariate models.

This thesis aims to develop a concept for a NIR-probe, which compensates the fouling through subtraction, whilst rugged spectra are built for multivariate data analysis.

Preliminary tests were carried out to verify the concept of creating difference spectra. During the recording of NIR-spectra it turned out that the calculation is only valid at certain optical path lengths. Based on these successful pretests the prototype of the fouling-compensating NIR-probe has been developed.

The designed sensor has two measuring channels with different path lengths as close as possible to each other. This fact yields the preliminary condition that both channels build up the same amount of fouling, so that after a subtraction of both measured spectra the resulting difference signal yields a spectrum, which does not contain fouling information. Initially the probe was effectively tested for chemical and temperature resistance.

The prototype has been proved during a lab scaled industrial esterification process, which led to the establishment of the respective reference analytics. This procedure showed up the effects of fouling on the design of multivariate regressions. Consequentially this effect was simulated and could be eliminated by building the difference spectra.

This thesis draws the conclusion that it is possible to verify the application of fouling-compensating NIR-probe as an inline method for the monitoring of esterification processes. Even though the prototype still needs some optimisation the previous mentioned inference could be clearly outlined.