Electrochemical fingerprinting of liquids using cross-sensitive polymers arrays and machine learning

Electronic tongues have been involved in the revolution of the traditional chemical sensing of liquids. The concepts of sensitivity and selectivity have started to assume different meanings, adapting to a new approach to the fingerprinting of complex liquid solutions. Electrochemical transducers are the most widely exploited read-out mechanisms for arrays of sensors that transform a chemical information into an electrical one, more manageable and easier to interpret. With respect to the voltammetric and amperometric techniques, that require dealing with moderate voltage and current values, potentiometry demonstartes to be the most suitable sensing method, compatible with the miniaturization and portability objectives of those types of devices. Open circuit voltage measurements allow to record potential signals coming from different electrodes of the array, which are functionalized with specific active materials. The present work provides an insight into the development of a portable potentiometric sensor for the analysis of complex multi-components liquid media. In particular, conductive polymers such as Polyaniline, PAPBA and Polypyrrole are presented, studied and characterized. Properties of polymeric coatings can be tuned by properly modifying the electropolymerization conditions in terms of reagents' nature and concentration, deposition technique and control of the doping process. It is shown the possible integration on the same array of the films expressing the most reliable responses in order to build a cross-sensitive system. Each polymer, being characterized by a certain sensitivity towards multiple ionic species, would respond in a different way. The analysis of the voltage traces is performed taking into account the simultaneous variation expressed by the integrated sensors, providing qualitative responses for the identification and classification of different solutions. Machine learning algorithms produce results related to the capability of the system to interpret correctly the acquired data for the recognition process. Such analytical procedures require training and calibration of the electronic tongue prior to the actual implementation. A reasonable and balanced data set consisting of enough observations can be created by automating the tests operations through a microfluidic system that comprises multiple components to allow the sensor to be in contact with the correct test solution for a specific amount of time. Multiple additional approaches have been proposed to enhance the miniaturization of the device, which has been employed for the classification of different mineral water brands. After the collection of the potentiometric responses, a certain pre-processing allows to extract meaningful features from the signals, which are related to the dynamic interaction of the polymers with the minerals, in order to achieve a dimensional reduction. Classification results validate the electronic tongue capabilities and close the feedback loop that connect the multidisciplinary aspects that lay behind the present work.