Abstract An array of piezoelectric quartz crystals was used to detect volatile organic compounds such as hydrocarbons, chlorinated compounds and alcohols. Steady-state frequency shifts have been used as the input parameters for multicomponent analysis. The coating materials chosen were side-chain-modified polysiloxanes. The results show clearly that these polymers provide excellent reproducibility over months. In addition, the performance of the array in the presence of humidity up to 70% r.h. does not decrease compared with dry air. In the multicomponent analysis, we compared commercially available partial least-squares regression (PLS) and artificial neural network (ANN) software. The neutral network designed for this application was small in order to avoid overfitting. For low-dimensional problems there is no difference between the two evaluation methods, but for complex ternary mixtures and long-term measurements the ANN offers advantages in predictability. Efforts were made to use a reduced set of calibration points, and here PLS presents the possibility of reducing the calibration time by 90% (use of Factorial and Box-Behnken designs) without loss of resolution, whereas the ANN suffers if a small number of training vectors is chosen.
[1]
Wolfgang Rosenstiel,et al.
Application of neural-network systems to the dynamic response of polymer-based sensor arrays
,
1995
.
[2]
B. Koppenhoefer,et al.
Chiral recognition in the resolution of enantiomers by GLC
,
1984
.
[3]
Kenneth R. Beebe,et al.
Selection of adsorbates for chemical sensor arrays by pattern recognition
,
1986
.
[4]
B. Kowalski,et al.
Chemical piezoelectric sensor and sensor array characterization
,
1986
.
[5]
J. Grate,et al.
Correlation of surface acoustic wave device coating responses with solubility properties and chemical structure using pattern recognition
,
1986
.