Previous Topic Back Forward Next Topic
Print Page Dr. Frank Dieterle
Ph. D. ThesisPh. D. Thesis 3. Theory – Quantification of the Refrigerants R22 and R134a: Part I3. Theory – Quantification of the Refrigerants R22 and R134a: Part I 3.1. Experimental3.1. Experimental
About Me
Ph. D. Thesis
  Table of Contents
  1. Introduction
  2. Theory – Fundamentals of the Multivariate Data Analysis
  3. Theory – Quantification of the Refrigerants R22 and R134a: Part I
    3.1. Experimental
    3.2. Single Analytes
    3.3. Sensitivities
    3.4. Calibrations of the Mixtures
    3.5. Variable Selection by Brute Force
    3.6. Conclusions
  4. Experiments, Setups and Data Sets
  5. Results – Kinetic Measurements
  6. Results – Multivariate Calibrations
  7. Results – Genetic Algorithm Framework
  8. Results – Growing Neural Network Framework
  9. Results – All Data Sets
  10. Results – Various Aspects of the Frameworks and Measurements
  11. Summary and Outlook
  12. References
  13. Acknowledgements
Research Tutorials
Site Map
Print this Page Print this Page

3.1.   Experimental

Both setups, which are based on the reflectometric interference spectroscopy, are described in detail in section 4.3 (Array set-up) and in section 4.4 (4l setup). For the sensor array setup, 6 sensitive polymer layers were prepared using the polymers Polyetherurethane (PUT), Polydimethylsiloxane (PDMS), a hyperbranched polyester (HBP), Ultrason (UE 2010) and Makrolon (M 2400). Besides of measurements of single analyte vapors for a sensitivity analysis, two data sets of binary mixtures were measured based on an equidistant 6-level full factorial design [155]. Thereby the relative saturation pressures and thus the concentrations of the analytes R22 and R134a were varied between 0 and 0.1 with synthetic air as ambient gas. The first data set was generated by measuring the experimental design 4 times with the sensor array RIfS setup and the second data set was produced by measuring the experimental design twice with the miniaturized 4l RIfS setup. The sensor signals were recorded after 10 minutes of exposure to analyte and a recovery time of 2 hours was chosen.

A 20-fold random subsampling procedure described in section 2.4 was used for splitting the data into a calibration data set (75%) and a test data set (25%) with the confinement that all repeated measurements of a concentration combination went into one subset to prevent overoptimistic predictions [156]. The neural networks implemented for this example had a topology of 1 output neuron, 4 neurons in 1 hidden layer and 6 respectively 2 input neurons with all features and parameters described in  section 2.7.3 in detail.

Page 42 © Dr. Frank Dieterle, 14.08.2006 Navigation