Variability of the Operating Conditions

Real life field conditions are never reproducible. For the same site of interest, diverse activities, different process management strategies or various raw material compositions induce highly variable emissions. 

As a result, for one odour source, there are many different mixtures of compounds. 

the various observations belonging to one class of signal patterns are far from being homogeneous. 

Other sources of variability are the meteorological conditions. They influence the rate and the composition of the generated odour and affect its dispersion. But they also exert an effect on the gas sensor behaviour (e.g. signal variation due to humidity change). 

For all those above reasons, trying to classify environmental odour sources in real life with pattern recognition techniques gives rise to a spread of observation points in the different clusters. For the signal processing, that implies to have a great number of samples in order to consider the various conditions and chemical compositions. But that induces a high dispersion of the data of a same class and/or of the same concentration. 

---

The data dispersion is shown in principal component analysis (PCA) scores plot (Figure 8.3). The projection in the two first components plane of the PCA highlights the effect of the variability of the operating conditions on the data. 

Fig. 8.3. PCA sores plot for measurements with a lab-made e-nose (array of 6 Figaro " sensors) in a composting plant. The high variance of the data is due to the variability of the operating conditions.

---