Use of an artificial neural network to model the quenching of a Ru.dpp:ormosil-based optrode to gaseous and dissolved oxygen

A variety of modelling schemes have been applied to the response of Ru.dpp:ormosil optrode. Artificial neural network model more accurately approximates the optrode response to oxygen than traditional two-site modelling and other standard models. Although there are a few neural networks that have been applied to optical sensor responses, comparison of the method with other modelling techniques has not previously been reported. This paper aims to re-investigate the artificial neural network concept for optical sensors, while explaining a clear methodology and development behind its application. The ANN model is simple when compered to other common modelling techniques. ANN methods present an alternative to numerical modelling systems that attempt to strictly quantify the influence of variables linked to physical parameters such as distribution of luminophores, rate of non-radiative energy transfers, analyte solubility and porosity of matrix. While networks can be designed to predict these physical parameters they are not restricted by them. The network accounts for variables not foreseen in numerical modelling which introduce error in approximating sensor response. High accuracy in predicting response versus actual experimental data is reflected in excellent R2 values. The ANN method applied was found to be superior to two-site modelling, linear regression, multi-linear regression and the oxygen solubility model. While this model applies strictly to the luminophore:matrix combination used here the methodology can be applied to other luminescent based sensor systems.