- Author
- Blomqvist, P. | Simonson, M.
- Title
- Rokgasmatningar med FTIR matteknik: Utvardering av mjukvara utvecklad som del av SAFIR. [Measurement of Smoke Gases Using FT-IR Techniques: Methods for Evaluation of the Spectral Information Have Been Developed in SAFIR.]
- Coporate
- Swedish National Testing and Research Institute, Boras, Sweden
- Report
- SP AR 1999:45, 1999, 39 p.
- Keywords
- storage | FT-IR | smoke | toxic gases | evaluation
- Identifiers
- evaluation methods
- Abstract
- ABSTRACT IN ENGLISH Measurements of smoke gases using FI-IR technique places great restraints on the evaluation of the FT-IR spectra. During a fire, water and carbon dioxide are both produced in high concentrations. These gases both exhibit absorption spectra that interfere with many of the gases to be quantified. Methods for evaluation of the spectral information have been developed in a EU research project, SAFIR. In this report these methods will be assessed, they will also be compared to a simple univariate method presently used as an on-line method at SP. The different evaluation methods that have been assessed are: "classical" multivariate methods (CLS, PLS and INLR), one newly developed multivariate method (QTFA) and one univariate method (peak height). The build-up steps of a complete prediction method are studied in detail for both the classical and the new multivariate methods and both the multivariate and the univariate methods are compared regarding prediction of gas concentrations from smoke gas spectra. The gases specifically studied are CO, HCl, HCN and NO. Predictions of the concentrations of these gases were made based on smoke gas spectra from real scale fire tests using particleboard and polyurethane foam. The univariat method studied in this work gave good accuracy in the prediction of gas concentrations. There are, however, several drawbacks with this method. In cases of a gas concentration exceeding the calibrated concentration span, or in cases of interference from an unknown component the results from this method are severely distorted. More seriously, no tool for warning for such distortions exists for this method. Similarly, the "classical" multivariate methods gave, in most cases, good accuracy in the prediction. An advantage of PLS/INLR is the possibility of calculating residual values, which serves as a warning tool for errors in the prediction. A disadvantage of these multivariate methods, however, is their greater sensitivity to the quality of the spectra in the calibration of the model compared to the univariat method studied. QTFA showed superior qualities in the case of spectral interference from an unknown gas. This is shown in one case where acetylene interferes in the evaluation of HCN.