Dual oxygen and temperature sensing with single indicator using multi-task-learning neural networks

Francesca Venturini, Umberto Michelucci, Michael Baumgartner


The optical determination of oxygen partial pressure is of great interest in numerous areas, like medicine, biotechnology, and chemistry. A well-known optical measuring approach is based on the quenching of luminescence by the oxygen molecules. The conventional approach consists in measuring the intensity decay time and relate it to the oxygen concentration through a multi-parametric model (Stern–Volmer equation). The parameters of this equation are, however, all temperature-dependent. Therefore the temperature needs to be known to determine the oxygen concentration and is measured separately, either optically or with a completely different sensor. This work proposes a new approach based on a multi-task learning (MTL) neural network. Using the luminescence data of one single indicator, which is sensitive to both oxygen and temperature, the neural network achieves predictions of both parameters which are comparable to the accuracy of commercial senors. The impact of the new proposed approach is however not limited to dual oxygen and temperature sensing, but can be applied to all those cases in which the sensor response is too complex, to be comfortably described by a mathematical model.