Multivariate sensing of ions using machine learning and composite 2D-3D graphene oxide-hexacyanoferrate electrodes

The conventional method for sensing relies on the development of highly selective materials capable of detecting specific target molecules or ions without interference from other species commonly found in real solutions. However, creating practical sensors that can effectively discriminate between analytes sharing similar chemistry presents significant challenges. To address this issue, we describe a novel approach utilizing an ensemble of four diverse amperometric sensors obtained for deposition of 2-dimensional graphene oxide nanosheets (GO) and 3-dimensional metal-organic frameworks (MOFs) based on redox active metal hexacyanoferrates. The multivariate signals obtained by the sensor array is used to train an artificial neural network (ANN) capable of analysing such complex inputs to accurately determine the concentrations of Na + and K + ions in solutions with varying ionic strengths. The sensing strategy is based on the differential intercalation and diffusion behaviour of Na + and K + ions within both GO and MOFs, resulting in distinct voltammetric signals. The neural network is trained using massive datasets comprising 327 variables as columns and over 4 million samples as rows. Following training, the sensor array demonstrates remarkable proficiency in accurately measuring the concentration of both ions present in solution, while a single sensor cannot discern between the signals generated by each ion. This ongoing work underscores the potential of integrating artificial intelligence with tunable materials to develop a new class of chemical sensors with enhanced discrimination capabilities, paving the way for more robust and versatile sensor technologies. • We describe sensing of Na+ and K+ ions in solutions mimicking artificial sweat. • Sensors based on 2-dimensional and 3-dimensional nanomaterials are used. • Machine learning allows to use non-selective Na+ and K+ sensors. • Over 4 million sample data are used for the training. • We also discriminate the contribution of each sensor to the final result.