Large language models, commonly known as LLMs, are showing promise in tacking some of the most complex tasks in AI. In this perspective, we review the wider field of foundation models—of which LLMs are a component—and their application to the field of materials discovery. In addition to the current state of the art—including applications to property prediction, synthesis planning and molecular generation—we also take a look to the future, and posit how new methods of data capture, and indeed modalities of data, will influence the direction of this emerging field.

Combining chemical sensor arrays with machine learning enables designing intelligent systems to perform complex sensing tasks and unveil properties that are not directly accessible through conventional analytical chemistry. However, personalized and portable sensor systems are typically unsuitable for the generation of extensive data sets, thereby limiting the ability to train large models in the chemical sensing realm. Foundation models have demonstrated unprecedented zero-shot learning capabilities on various data structures and modalities, in particular for language and vision. Transfer learning from such models is explored by providing a framework to create effective data representations for chemical sensors and ultimately describe a novel, generalizable approach for AI-assisted chemical sensing. The translation of signals produced by remarkably simple and portable multi-sensor systems into visual fingerprints of liquid samples under test is demonstrated, and it is illustrated that how a pipeline incorporating pretrained vision models yields > 95 % $>95\%$ average classification accuracy in four unrelated chemical sensing tasks with limited domain-specific training measurements. This approach matches or outperforms expert-curated sensor signal features, thereby providing a generalization of data processing for ultimate ease-of-use and broad applicability to enable interpretation of multi-signal outputs for generic sensing applications.

The cross-sensitivity of materials in low-selective sensor arrays, namely e-noses and e-tongues, results in a convoluted sensor array response, which renders traditional analytical methods for data processing ineffective. Machine learning approaches can help discover the latent information in such data, and various data processing methods, including unsupervised and supervised techniques, have been proposed to calibrate those devices. In this study, we demonstrate HyperTaste Lab—a notebook with a machine learning pipeline for potentiometric sensor arrays. The ability of the notebook to process raw data produced by model sensor arrays comprising cross-sensitive and/or ion-selective electrodes is demonstrated for the characterization of drinking water and consumer beverages. We describe the modular data processing and machine learning framework that can be applied by sensor researchers to accommodate different signal modalities and perform various downstream tasks, such as the verification of a product's originality, the estimation of ion concentrations, and the quantitative prediction of sensory descriptors.

Electrochemical sensor arrays have shown potential for fast and untargeted chemical analysis of multi-component media, enabling simultaneous quantification of multiple analytes and estimation of liquid properties that could correlate with human sensory perception. In particular, potentiometric electronic tongues (e-tongues) have been explored as promising alternative tools for chemical analysis in various applications, ranging from traceability of goods to characterization of food products for quality control and innovation processes. The reduced complexity of non-selective sensor fabrication and the ease of potentiometric transduction make these devices suitable for portable and decentralized chemical analysis. However, the interpretation of the sensor array response has always been a challenge due to the inherent cross-sensitivity of the sensing materials and the corresponding combinatorial signals arising upon sensor interaction with a liquid. Machine learning can help recognizing and mapping signal patterns and various data processing, unsupervised and supervised techniques have been proposed to calibrate e-tongue devices. These methods usually require exposing the sensor array to a set of liquids with known properties that serve as training base to build a calibration model. Therefore, the quality and quantity of tests is crucial to boost sensor performances. Nevertheless, performing an extensive number of measurements could be extremely time-consuming and results to be tedious to achieve in practice for certain use-cases. In this context, advances in deep learning and machine learning models have shown potential to accelerate chemical and materials discovery when combined with high-throughput experimentation, highlighting the benefits of AI-assisted research practices. Moreover, the recent advent of multi-domain and multi-task models trained by self-supervision, so-called foundation models, bears also promises for extending learnt representations across multiple fields, thus counteracting the reduced data availability in certain applications, and benefiting from information exchange across domains. Thus, in the present contribution we propose extending this approach to data-driven chemical sensors. More specifically, we leverage transfer learning based on fingerprints pretrained in other domains to model new instrument/sensor data representations. Herein, we demonstrate how the output of a model system comprising an integrated electrochemical sensor array for analysis of multi-component liquids can be encoded as image representations to leverage existing deep learning computer vision models pretrained on large collections of image data. The models effectively extract features from these representations and feed specific model heads to perform downstream tasks. Firstly, an integrated sensor array comprising 16 polymeric sensors was fabricated through electrodeposition on conventional electroless nickel immersion gold (ENIG) electrodes. The conductive polymers (PEDOT, PPy, PANI and PAPBA) were synthesized by chronoamperometry or cyclic voltammetry in a three-electrode configuration and were enriched with doping agents for enhanced sensitivity. 15 linearly independent differential voltages between these polymeric sensors were measured during the transition of the sensor array from a reference solution to a test solution, thereby obviating the need for a conventional reference electrode. Indeed, the use of low-selective polymeric sensors for potentiometric measurements does not necessarily require integration of reference electrodes, which are known to be unpractical for remote sensing applications. Training data were obtained by alternatively immersing the sensor array in reference (120 s) and test (60 s) solutions continuously using an automated test rig. The 15 raw time-series data from the sensor array were processed and concatenated to yield a spectral response, which was smoothed by means of the moving average technique and standardized using Standard Normal Variate (SNV). The obtained spectra were encoded into image representations using the Gramian Angular Summation Field transformation. Off-the-shelf features were generated leveraging pretrained neural networks developed to classify natural images and applied to the “sensor images”. Dimensionality reduction through Principal Component Analysis (PCA) yielded a set of features that could then be used to train machine learning classifiers and regressors. The pipeline was applied to generate visual fingerprints of multiple beverages, proving full discrimination of liquid types (mineral waters, coffees, fruit juices, soft beverages and wines). On a model dataset comprising 11 Italian red wines, it was demonstrated that image fingerprints of samples enabled class identification with a mean accuracy ~95%. The results demonstrate the successful creation of a new representation of the chemical sensing space which achieves comparable performance as domain-specific hand-crafted feature selection. The present contribution represents an example of integration of data processing techniques and publicly available libraries/models to support transfer of methodologies across domains. We believe this approach could be disruptive in the field of electrochemical sensor arrays, especially for processing e-tongues and e-noses response and enhancing capabilities of data-driven chemical sensors. Figure 1

A proof-of-concept system comprising a miniaturized sensor array, feature extraction and machine learning pipeline was evaluated for the direct quantification of the concentrations of three major cations, Ca 2+ , Mg 2+ , and Na + , in drinking water. Feature importance methods were applied to discover dependencies between the transient potentiometric responses of sensing materials and the cation concentrations. The proposed framework supports design of cross-sensitive sensor arrays to accelerate water testing, providing a complementary approach to traditional chemical analysis for monitoring water quality.

Methods are shown to quickly select and test best sorbent materials and structuring processes for rapid thermal swing adsorption (RTSA) gas separation with 2-4 minute cycles while making use of abundant low-grade <120 °C, waste heat co-emitted at point sources with CO2 and minimizing exergy losses. A temperature jump setup determines capacity and kinetics of adsorbent layers on finned heat sinks designed to optimize capacity per unit time, active to dead mass ratio and pressure drop. Dynamic vapor sorption is used to determine CO2 equilibrium sorption also in presence of H2O as well as sorption isotherms and adsorption selectivity. RTSA could accelerate the fight against climate change by energy efficiency being combined with reuse of abundant heat waste for carbon capture. Conclusions are drawn how to speed up material and system development for low-cost carbon capture with accelerated materials discovery concepts.

Potentiometric electronic tongues (ETs) leveraging trends in miniaturization and internet of things (IoT) bear promise for facile mobile chemical analysis of complex multi-component liquids, such as beverages. In this work, hand-crafted feature extraction from the transient potentiometric response of an array of low-selective miniaturized polymeric sensors is combined with a data pipeline for deployment of trained machine learning models on a cloud back-end or edge device. The sensor array demonstrated sensitivity to different organic acids and exhibited interesting performance for the fingerprinting of fruit juices and wines, including differentiation of samples through supervised learning based on sensory descriptors and prediction of consumer acceptability of aged juice samples. Product authentication, quality control and support of sensory evaluation are some of the applications that are expected to benefit from integrated electronic tongues that facilitate the characterization of complex properties of multi-component liquids.

The design of cost-effective and rapid screening sensing systems is key to deliver alternative tools for chemical analysis. Electronic tongues can distinguish complex liquids by combining cross-sensitive sensor arrays with machine learning and have been demonstrated in various chemical sensing applications. In this live demonstration, feature extraction from the transient potentiometric response of an array of low-selective miniaturized polymeric sensors is combined with a data processing pipeline for deployment of trained machine learning models on an edge device. A mobile app allows visualization of raw data and measurement results in addition to reconfiguration of the device based on user selection.

The use of a miniaturized potentiometric electronic tongue based on low-selective polymeric sensors was demonstrated for the discrimination and sensory characterization of coffee samples. The sensor array was able to discriminate 21 varieties of coffee with an average accuracy of 91.3% by combining hand-crafted features, predictor importance methods and trained classification models. Moreover, the e-tongue supported by a single regressor could be successfully trained to predict simultaneously the intensity of 13 coffee descriptors by leveraging dimensionality reduction to learn their interdependence. Sensory profiles of 33 samples were reconstructed with a 0.78 RV coefficient of agreement with sensory data using a rigorous leave-one-coffee-out validation. This study emphasizes the advantages of data-driven sensing approaches based on training by examples to help increase sample throughput when exploring new formulations and accelerate product design cycles.

Potentiometric electronic tongues have been proposed as versatile sensor arrays for chemical analysis of a wide variety of analytes without the need for high specificity of individual sensors [i] . The prospect of portable electronic tongues offers new approaches for distributed, decentralized chemical analysis for various applications such as traceability of goods in supply chains [ii] , [iii] . A challenge for electronic tongues, however, has often been the quantitative determination of concentrations of target analytes in the presence of multiple interferents due to the inherent cross-sensitivity of the sensors employed. In the present contribution, we extend our previously reported work [iv] on an all-solid-state electronic tongue based on an array of electrodeposited polymeric sensors to study the feasibility and accuracy of employing this class of sensors to the quantification of multiple analytes. The sensor array studied herein is entirely integrated on a conventional printed circuit board (PCB). We demonstrate how a remarkably simple sensor configuration can be calibrated by means of a prescriptive training scheme and tree-based machine learning algorithms to provide a quantitative assessment of the concentrations of multiple metal cations concomitantly in a single measurement lasting less than 2 minutes. In contrast to conventional electrochemical sensors, no reference electrode was employed. Interestingly, a relative accuracy better than 7% was found for the determination of concentrations of various ions (Na, Mn, Fe, Al, Cu, Pb) in pure solutions at concentrations ranging between 2 mg/L and 100 mg/L. The potential applicability of this chemical testing framework to decentralized analysis was showcased by deploying a trained machine learning regression model as a cloud service that was invoked from a mobile phone serving as gateway for the portable potentiometric sensor array. Thus, the direct quantification of multiple cations in aqueous mixtures was realized by applying an automated machine learning pipeline including data pre-processing and feature extraction to the transient potentiometric data obtained from the miniaturized array of polymeric sensors integrated on a PCB. The proof-of-concept system employed 16 polymeric sensors that were electrodeposited on conventional electroless nickel immersion gold (ENIG) metallization. The conductive polymers were electrodeposited by chronoamperometry or cyclic voltammetry and tailored for enhanced cationic sensitivity. Characteristic features were extracted from the transient differential voltages between these polymeric sensors during the transition of the sensor array from a reference solution to a test solution, thereby obviating the need for a conventional reference electrode. The analysis of the differential voltages revealed deviations from a linear sensitivity, showing super-Nernstian responses and non-linear voltage trends when varying the concentration of the cations mentioned above. Extraction of cross-sensitivity parameters, such as average sensitivity slope and non-selectivity factor, demonstrate how the potentiometric response of the electronic tongue based entirely on non-selective sensors differs from that of an array of exclusively selective sensors (e.g. ion-selective electrodes, ISEs) presented in previous works. Training data for the tree-based machine learning algorithm was obtained by alternatively immersing the sensor array in reference and test solutions continuously using an automated test rig. In addition to the conventional approach of measuring the equilibrium potential after a certain settling time, we find that the analysis of the complete evolution of the potentiometric signal in time provides additional information that is useful for training of the regression models. Dimension reduction of time series data was achieved through various techniques, such as down-sampling, feature transformation (e.g. PCA) and feature extraction. Here, five characteristics from each potentiometric transient were used to describe the complete voltage perturbation of the sensor during transition between reference and test solution. Discrimination of multiple ions and sensitivity to variation in their concentrations was demonstrated by testing 25 nitrate-based mixtures of six ions (Al, Cu, Na, Mn, Fe, Pb) in the range 0.5-10 mg/L following an Orthogonal Experimental Design (OED). Extra Trees-based regression models showed significantly higher quantification accuracy compared to widely used multivariate regression models such as Multiple Linear Regression (MLR). Concentrations of the metal cations in pure solutions were determined in less than 2 minutes at an average mean relative error of 1-7% in the concentration range 2-100 mg/L. In a model mixture comprising Al, Cu, Na and Fe, the mean relative error was found to depend on the type of ion, and varied between 1% for Fe and 44% for Na in the concentration range 1-10 mg/L. Regression models trained using the combination of the five extracted features showed higher predictive capabilities compared to single-features based algorithms, thereby underscoring the importance of feature selection for data-driven sensors. Finally, we demonstrate the functionality of the proposed portable device by quantifying concentrations of multi-ion mixtures using an integrated device comprising the sensor array, microcontroller-based data acquisition, wireless data transmission, a mobile app and cloud service providing the trained regression model. [i] Vlasov, Y.; Legin, A.; Rudnitskaya, A.; Di Natale, C.; D'Amico, A. Nonspecific sensor arrays ("electronic tongue") for chemical analysis of liquids: (IUPAC technical report), Pure Appl. Chem ., 2005 , 77, 1965-1983 [ii] Legin A.; Rudnitskaya, A.; Vlasov, Y. Electronic tongues: New analytical perspective for chemical sensors, Comp. Anal. Chem ., 2003 , 39 , 437-486 [iii] Kirsanov, D.; Correa, D.S.; Gaal, G.; Riul, A. Electronic Tongues for Inedible Media., Sensors 2019 , 19 , 5113 [iv] P. W. Ruch et al ., "A portable potentiometric electronic tongue leveraging smartphone and cloud platforms," 2019 IEEE International Symposium on Olfaction and Electronic Nose (ISOEN) , Fukuoka, Japan, 2019 . doi: 10.1109/ISOEN.2019.8823244 Figure 1

The direct quantification of multiple ions in aqueous mixtures is achieved by combining an automated machine learning pipeline with transient potentiometric data obtained from a single miniaturized array of polymeric sensors electrodeposited on a conventional printed circuit board (PCB) substrate. A proof-of-concept system was demonstrated by employing 16 polymeric sensors in combination with features extracted from the transient differential voltages produced by these sensors when transitioning from a reference solution to a test solution, thereby obviating the need for a conventional reference electrode. A tree-based regression model enabled concentrations of various metal cations in pure solutions to be determined in less than 2 min. In a model mixture comprising Al3+, Cu2+, Na+, and Fe3+, the mean relative error was found to depend on the type of ion and varied between 1% for Fe3+ and 44% for Na+ in the concentration range 1-10 mg/L. Overall, a mean relative error of 16% was obtained for quantification of these four ions across a total of 124 tests in different solutions spanning concentrations between 2 and 360 mg/L. These results demonstrate how the analytical capability of a multiselective sensor array can leverage data-driven approaches through training by examples for accelerated testing and can be proposed to complement traditional analytical tools to meet industrial demands, including traceability of chemicals.

Wearables that acquire relevant vital and contextual parameters improve work safety as well as quality of life of elderly citizens or patients with chronic diseases. A scalable architecture connects wearables via a hub to the cloud and combines edge with cloud computing to provide optimal user interaction and allow analytics on multi-stream data. The functionality was expanded to enable demonstrations of physiological and psychological stress classification in firemen and mobile health interventions in patients with lung diseases. Following an initial table-top edge demonstrator a hemi-spherical display improves emotional contact to users. A first use case tested an integrated acquisition and inference system that was trained to differentiate physical and emotional stress. The system measured stress in firemen during training in a cage maze and in hot training locations and provided functions to acquire expert labels. A second use case focused on mobile-health intervention for patients suffering from Chronic-Obstructive-Pulmonary-Disease (COPD), to improve their quality-of-life. Patient-physician conversations are extended through a communication channel and a virtual assistant provides disease related information, reminders, and alerts.

Monolithic, nitrogen-doped carbon sorbents were prepared from resorcinol-urea-formaldehyde resins and physically activated with CO2 for different activation times. The effect of the activation time on the water sorption behavior and the physicochemical properties were investigated. Longer activation times lead to a steeper slope of the water sorption isotherm and, due to a higher specific surface area and micropore volume, an increased water sorption capacity. It was found that, after physical activation for 3 h at 800 degrees C, the physically activated, nitrogen-doped carbon has a high surface area (>1000 m(2)/g) and a high water sorption capacity (50 wt%). In a miniaturized adsorption heat pump test stand, the best candidate material was assessed alongside commercial silica gel for reference. At a temperature swing from 90 degrees C -> 50 degrees C, the CO2-activated carbon exhibits a maximal specific cooling power which is a factor of 1.7 higher in comparison with the reference silica gel (429 W/kg versus 255 W/kg). At a more applicable temperature swing, 60 degrees C -> 30 degrees C, the CO2-activated carbon yields a specific cooling power 3.8 times higher than that of the silica gel reference (932 W/kg versus 240 W/kg).

Adsorption heat pumps offer a clean, zero-emission technology for universally applicable cooling or heating utilizing water as a refrigerant and waste or renewable heat as driving energy instead of electricity. Despite their attractive environmentally friendly prospects, the broader application of such classes of heat pumps has not yet been possible, mainly because of the low power density of adsorption heat exchangers and the corresponding large size and high cost of the adsorption heat pumps. We report an inexpensive route for the fabrication of zeolite coatings with high adsorption power density based on the bottom-up assembly of colloids directed by magnetic and capillary forces. Such an assembly process relies on the chaining of oil droplets under an external magnetic field during deposition of the coating, followed by the formation of a percolating network of bridged adsorbent particles upon drying. This results in vertically open channels and thermal bridges that facilitate directed mass and heat transport across the structured zeolite coating during sorption cycles. By reaching up to 3.3-fold higher performance than their unstructured counterparts using readily available zeolite as an adsorbent material, the architectured coatings produced through this facile, upscalable approach hold great potential for next-generation adsorption heat pumps.

Electronic tongues based on potentiometry offer the prospect of rapid and continuous chemical fingerprinting for portable and remote systems. The present contribution presents a technology platform including a miniaturized electronic tongue based on electropolymerized ion-sensitive films, microcontroller-based data acquisition, a smartphone interface and cloud computing back-end for data storage and deployment of machine learning models. The sensor array records a series of differential voltages without use of a true reference electrode and the resulting time-series potentiometry data is used to train supervised machine learning algorithms. For trained systems, inferencing tasks such as the classification of liquids are realized within less than 1 minute including data acquisition at the edge and inference using the cloud-deployed machine learning model. Preliminary demonstration of the complete electronic tongue technology stack is reported for the classification of beverages and mineral water.

Pastes based on copper (Cu) nanoparticles (NPs) are promising electronic-packaging materials for the attachment of high-power devices. However, the rapid oxidation of nanostructured Cu requires the use of reducing agents during processing, which makes it less suitable for attaching large-area dies (> 4 mm2). Recently, the functionalization of Cu-NP surfaces with a mixture of amines prevented oxidation, allowing for sintering without the need for reducing agents. Here we investigate the sintering mechanisms involved during die attachment using pastes of passivated Cu NPs, with particular focus on the critical role of the carrier solvents. Using 1-nonanol or 1-decanol as solvents, we first demonstrate the absence of Cu-oxide phases in the pastes after fabrication and the stability of the resulting nanostructured copper for as much as 30 min in air. By measuring the evolution of the electrical characteristics of the paste during drying and sintering, we show that electrically conductive agglomerates form among the NPs between 141°C and 144°C, independent of the carrier solvent used. The carrier solvent was found to affect mainly the densification temperature of the copper agglomerates. Because they lead to uniform sintering of the material, Cu pastes based on solvents with a low boiling point and high vapor pressure are preferable for attaching dies with area greater than 25 mm2. We show that dies with an area as large as 100 mm2 can be attached using a Cu paste based on 1-nonanol. These pastes enables the formation of temperature-resistant bonding for high-power devices using a simple and cost-effective approach.

A 3D printing method (the Direct Ink writing, DIW, method) is applied to produce SAPO-34 zeolite based structured adsorbents with the shape of a honeycomb-like monolith. The use of the 3D printing technique gives this structure a well-defined and easily adaptable geometry. As binder material, methyl cellulose was used. The SAPO-34 monolith was characterized by SEM as well as Ar and Hg porosimetry. The CO2 adsorption affinity, capacity and heat of adsorption were determined by recording high pressure adsorption isotherms at different temperatures, using the gravimetric technique. The separation potential was investigated by means of breakthrough experiments with mixtures of CO2 and N2. The experimental selectivity of CO2/N2 separation was compared to the selectivity as predicted by the Ideal Adsorbed Solution Theory. A drop in capacity was noticed during the experiments and N2 capacities were close to zero or slightly negative due to the very low adsorption, meaning absolute selectivity values could not be determined. However, due to the low N2 capacity, experimental selectivity is estimated to be excellent as was predicted with IAST. While the 3D printing is found to be a practical, fast and flexible route to generate monolithic adsorbent structures, improvements in formulation are required in terms of sample robustness for handling purposes and heat transfer characteristics of the obtained monoliths during gas separation.

Developing strategies to reduce mass and heat transport limitations is one of the most important challenges in adsorption heat exchanger technology. Due to the strong coupling of mass and heat transport in these systems, it is difficult to determine the individual transport limitations quantitatively. In order to find an optimal design where heat and mass transport are balanced, a quantitative method that enables a direct comparison of these two transport phenomena is needed. In the present work, a novel experimental approach to discriminate between mass and heat transport is proposed based on the measurement of adsorbent temperature, heat exchanger surface temperature and vapor pressure. The methodology is applied to micro/mesoporous silica spheres arranged in a monolayer or bilayer loose grain configuration or in a monolayer configuration adhesively bonded to the substrate. While the monolayer configuration exhibits balanced heat and mass transport, we find that the bilayer and the thermally enhanced configurations are limited by heat and mass transport, respectively. The application of the proposed methodology to compare heat and mass transport limitations in other industrially-relevant adsorbent materials should greatly aid the design of more efficient adsorption heat exchangers for a wide range of applications. (C) 2018 Elsevier Ltd. All rights reserved.