Applications of guided microwave spectroscopy in process analysis

Document Sample
Applications of guided microwave spectroscopy in process analysis Powered By Docstoc
					  Applications of guided
microwave spectroscopy in
     process analysis
            Vicki Loades* and Tony Walmsley
                      Analytical Science Group
                      Department of Chemistry
                         University of Hull

                  v.c.loades@chem.hull.ac.uk



26th April 2002      Vicki Loades, University of Hull
• Aim
    – To demonstrate that guided microwave spectroscopy
      (GMS) is an alternative method for process analysis.


• Introduction
    – Background and Benefits of GMS
    – Brief Description of some multivariate methods
    – Some Examples of Research in the area.
         • Binary solutions (low and high levels of components)
         • Analysis of a multiphase industrial samples




26th April 2002        Vicki Loades, University of Hull
   Ideal Process Spectrometer
• Wish List For a Process Spectrometer
    –   Non-invasive
    –   Non-destructive
    –   Suitable for solids/liquids, gases and suspensions
    –   Suitable for dark coloured samples
    –   Analyses the whole sample
    –   Large sample Volumes




26th April 2002      Vicki Loades, University of Hull
       Microwave Spectroscopy
• Microwave region: Approx. 200MHz and 80GHz.
• Microwave spectra are a result two properties.
  Dielectric constant (ε’) - Reduction in velocity
  – As the electromagnetic wave passes through the sample it
    causes an alternating polarization. This polarization and
    depolarization reduces the wave velocity across the chamber
    during analysis.
  Dielectric loss (ε’’) - Reduction in magnitude
  – As the molecules orientate in the electric field energy is lost to
    friction. This causes the waves magnitude to reduce across
    the sample.


 26th April 2002      Vicki Loades, University of Hull
       Epsilon Industrial Guided
       Microwave SpectrometerTM




26th April 2002   Vicki Loades, University of Hull
  Waveguide Frequency Cut - off
                  fc  vvacuum / 2a  '



                                                       a




26th April 2002     Vicki Loades, University of Hull
                    Example GM Spectra
         Response




Cut-
off




                              Frequency [MHz]

 26th April 2002        Vicki Loades, University of Hull
   Principal Component Analysis
• This is a method that decomposes the spectra
  into principal components (PC) each PC has
  score and loading.
    – By plotting the scores it is possible to visualise trends
      in spectral data that might have been difficult to see in
      the original spectra.
    – Loadings plots can show where the main regions of
      variation in the spectra occur and when the Number
      of PC’s start to represent noise in the data rather than
      useful information.

26th April 2002     Vicki Loades, University of Hull
           Partial Least Squares
• This is an extension to PCA, but this time reference
  measurements (e.g. concentration, density or pH)
  for the spectral data are also included in the
  calculations.
    – This allows calibration models to be generated which can
      then be used to predict the levels of unknown samples.
    – Prediction ability is determined by the Root Mean Squared
      Prediction Error (RMSPE).

                   RMSPE 
                               (y   act    y pred ) 2
                                           n


26th April 2002     Vicki Loades, University of Hull
          Spectral Pre-treatment
• Pre-treatment methods can be used to improve
  correlation between reference and spectral data.
    – Background subtraction removes the pure spectra of
      a component from the data.
         • Useful when components are masked by a more responsive
           component which is not of interest.
    – Mean centring subtracts the mean from the data set.
         • This removes a large chunk of the magnitude leaving the true
           variation in the data.
    – Orthogonal signal correction removes variables which
      are not orthogonal to the reference data.
         • A more involved method useful for highly co-linear and
           overlapping spectra


26th April 2002        Vicki Loades, University of Hull
                  Binary Solutions
 • Aim
      – To build calibration models which can accurately
        predict components of interest.

 • Different sample sets;
    – Aqueous samples at levels below 30%.
    – Alcohol solutions above 30%.

 • Solvents are fairly difficult to analyse by non-destructive
   methods such as spectroscopy.
      – The standard method for analysis is Gas Chromatography

26th April 2002     Vicki Loades, University of Hull
                      Sample Sets
Set Components                    Concentration Range (% v/v)
Ref
 1    Acetonitrile in water       5, 8, 10, 12, 15, 20, 23, 25, 30.

 2    Ethanol in water            1, 3, 5, 8, 10, 12, 15, 20, 23, 25, 28, 30

 3    Methanol and Ethanol        30, 35, 40, 45, 50, 55, 60, 65, 70.

 4    Ethanol and Propanol        30, 35, 40, 45, 50, 55, 60, 65, 70.

 5    Methanol and Propanol 30, 35, 40, 45, 50, 55, 60, 65, 70.


26th April 2002          Vicki Loades, University of Hull
 Acetonitrile Samples Spectra




26th April 2002   Vicki Loades, University of Hull
     Acetonitrile Samples Spectra
       Background Subtracted



RMSPE
1.07%




26th April 2002   Vicki Loades, University of Hull
      Ethanol Samples Spectra




26th April 2002   Vicki Loades, University of Hull
          Ethanol Samples Spectra
           Background Subtracted



RMSPE
0.28%




 26th April 2002   Vicki Loades, University of Hull
           Methanol and Ethanol



RMSPE
1.1%




26th April 2002   Vicki Loades, University of Hull
         Methanol and Propanol



RMSPE
0.35%




26th April 2002   Vicki Loades, University of Hull
            Ethanol and Propanol



RMSPE
0.95%




 26th April 2002   Vicki Loades, University of Hull
                  Industrial Analysis
• Aim:
    – To be able to monitor a multiphase sample of an
      industrial process as it is converted to product.
    – Investigate effect of phase variations to the GM
      Spectra of the samples.
         • Ultimately use GMS to control the conversion process
           between reactor vessels.
         • Sample is difficult to analyse by standard methods: It is
           dark in colour, has high level of solids and also has
           organic and aqueous phases.



26th April 2002       Vicki Loades, University of Hull
                  On-line Analysis
                      Potential Analysis Points




           Vessel 1               Vessel 2                 Vessel 3

26th April 2002         Vicki Loades, University of Hull
                  Industrial Samples
• Sample composition
   – 50% Aqueous
   – 30% Solid
   – 20% Organic
• Pale brown paste when mixed.
• The electromagnetic properties vary with sample phase.


   Process samples             12.5, 45.9, 50.0, 52.4, 92.0,
   Conversion levels (%)       93.0, 98.4, 99.8, 99.9, 100.

26th April 2002      Vicki Loades, University of Hull
                  Typical Sample




26th April 2002    Vicki Loades, University of Hull
GM Spectra of Industrial Samples




26th April 2002   Vicki Loades, University of Hull
 Signal Corrected GM Spectra


                                                     Increasing
                                                     % Oxidation




26th April 2002   Vicki Loades, University of Hull
PCA of Signal Corrected GM Spectra

                                                ~ 98 %
                                            Conversion
                     Increasing oxidation level




                                  ~ 50 % Conversion




                  12.5 % Conversion




26th April 2002    Vicki Loades, University of Hull
                  Sample Separation
                                  • Record GM spectra
                                    as the mixed sample
                                    separates back into
                                    the original phases
                                       – As shown on left
                                       – Takes approx. 3mins




26th April 2002      Vicki Loades, University of Hull
     GM Spectra of sample phase
            separation




26th April 2002   Vicki Loades, University of Hull
   Principal Component Analysis




                                         Time




26th April 2002   Vicki Loades, University of Hull
                     Summary
• We have shown that Guided Microwave Spectroscopy
  (GMS) when combined with multivariate methods can
  be used as an alternative for process analysis.
   – The main advantage is the suitability of this method for
     samples which are of multiple phases or that are dark in
     colour.
   – In particular the applications of the analysis of some
     binary mixtures and more importantly samples taken
     from an industrial process have been demonstrated.




26th April 2002    Vicki Loades, University of Hull
                  On Going Work
• Currently monitoring beer fermentation as a
  batch process taking place inside the guided
  microwave spectrometer’s sample chamber.
    – This is a living system containing dissolved gases,
      particulates.
    – Initial results are promising with visible trends in the
      spectra and scores plots.




26th April 2002     Vicki Loades, University of Hull
              Acknowledgements
• Chris Walker and Steward MacKenzie, ThermoONIX.
• Sylvia Ewans, Ewan Polwart and Ian Wells, Avecia.

• This research is funded by:
   – EPSRC, Engineering and Physical Sciences Research
     Council.
   – CPACT, Centre for Process Analytics & Control
     Technology.




26th April 2002   Vicki Loades, University of Hull

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:6
posted:8/17/2012
language:
pages:32