Cascade of Artificial Neural Network committees for the calibration of small gas commercial sensors for NO2, NH3 and CO

We propose a novel structure of a cascade of Artificial Neural Network (ANN) committees for the quantification of mixtures. In this structure the committees first analyze the gases that have a better regression and then pass the predicted concentration to the other committees, thus improving the information available for the most difficult gases without increasing the complexity of the ANNs. To test the structure we did setup and experiment with three different gases: CO, NO2 and NH3. The gas flows were controlled by an automated system that also controlled the environmental conditions and mixed the gases delivering them onto the measurement cell where three small commercial sensors were placed. The sensor data were later analyzed and different calibration methods, such as Partial Least Square regression, committee of Artificial Neural Networks and the cascade of committees of ANNs were evaluated with their measurement uncertainty and compared among them.