Multivariant Partial Least Square and Gradient Decent Regularization for transfer of training models in MEMS based gas sensor arrays

摘要

In this present work, three identical MEMS based multi gas sensor array were developed for measurement of major pollutants present in air pollution. The aim was to develop a methodology where in minimum sets of experiments were required to transfer the training model developed on one system to other. To perform the transform, three sensor arrays had been developed having same identical MEMS gas sensors. One out of three was chosen as main and other two as secondary sensor array. Box - Behnken (BB) design of experiments was used to develop different combination of 8 experimental sets. The mixture of gases consisted of NH3, CO, H2 S and C$_{2}H_{5}$OH. Multivariant partial least square regression and Gradient Decent Regularization was used to develop a methodology that could be used to transfer the training models from one system to another. Analysis and results of these techniques are presented in this paper.