odour; principal component analysis; discriminant function analysis; partial least squares
Abstract :
[en] An array of tin-oxide sensors is used to continuously monitor different odour emissions in the environment. The paper presents some issues aiming at improving the portability and the user-friendliness of the instrument as well as testing what kind of signal may be used to monitor the odour "intensity".
Main results are the following.
The test of various pre-processing data algorithms pointed out that the use of pure reference air could be avoided, as long as the sensors are allowed to periodically regenerate in the presence of ambient air.
Sensor array in static contact with ambient air could be sufficient for the on-line monitoring, but the use of a controlled gas flow system to transfer the odour from the source is better to avoid the influence of air movement on the heated sensors.
The control of the temperature and the humidity of the gas and the thermo-regulation of the sensor chamber don't seem essential, even for outdoor operation.
When trying to build a regression model linking the odour intensity to the sensor signals, Partial Least Square (PLS) model gives better results with respect to Multi Linear Regression or Principal Component regression models. For an application around a landfill area, where sensor signals are compared to the personal feeling of the operator in the field, PLS gives a percentage of 71% correct intensity prediction.