Multisensor arrays, multisensor systems and related methods

Abstract

Described is a multisensor array comprising multiple sensors for detection of one or more analyte(s) in the gas phase or in the liquid phase, wherein the multiple sensors comprise: a first sensor configured to detect a variation of an electrical parameter based on a dipole-dipole interaction with the analyte(s); a second sensor configured to detect a variation of an electrical parameter based on a hydrogen bonding interaction with the analyte(s); and a third sensor configured to detect a variation of an electrical parameter based on a dispersive interaction with the analyte(s). Furthermore, a multisensor system is disclosed, which comprises the aforementioned multisensor array and a processing unit comprising a pattern recognition and machine learning analyzer wherein the pattern recognition and machine learning analyzer is configured to receive output signals of the multisensor array and compare them to stored data in order to classify and / or identify the one or more analyte(s); wherein the stored data includes Hansen solubility parameters obtained from a library database, thus enabling classification of unknown analytes. A method of analyzing a sample present in gas or liquid phase by using the aforementioned multisensor array is also described.

Description

MULTISENSOR ARRAYS, MULTISENSOR SYSTEMS AND RELATED METHODSFIELD OF INVENTION

[0001] The present invention generally relates to the field of multiarray sensors for categorization, classification and / or identification of one or more analyte(s) in gas or liquid phase.

[0002] In certain embodiments, this invention relates to a multisensor array for detection of one or more analyte(s) in the gas phase or liquid phase and to a multisensor system comprising the same. In further embodiments, the invention relates to a method of analyzing a sample present in gas or liquid phase by using the aforementioned multisensor array.BACKGROUND OF THE INVENTION

[0003] Chemical sensors are devices which transform chemical information, ranging from the concentration of a specific sample component to total composition analysis into an analytically useful analogue or digital signal. For this purpose, such devices include a sensing material (or receptor) and a transducer as essential components. For example, WO 01 / 44796 describes a nanotube device used as a chemical or biological sensor, the nanotube device comprising at least one nanotube (e.g. a carbon nanotube), which is electrically connected with its ends to first and second conducting elements, such as electrodes. To tune the sensitivity of the device to a variety of molecular species the nanotubes may be modified by coating, or decorating with one or more sensing agents, so as to impart sensitivity to a particular species in its environment.

[0004] Multisensor arrays have found widespread use in detection and identification of chemical compounds in chemical industry, medicine, environmental monitoring, safety, military applications and public use, for example.

[0005] Conventional multisensor arrays typically employ receptors selected to specifically bind to the analyte of interest. Whereas such “lock-and-key” designs may be advantageous in terms of their selectivity towards the analyte, their application does not enable identification of similar chemical species or compound classes, and the necessary custom-tailored synthesis of receptor materials is often elaborate and accompanied with high costs.

[0006] Further examples of multisensor arrays include so-called electronic noses (e-noses) and electronic tongues for sensing analytes in gas phase and liquid phase, respectively, which often combine chemical sensing with pattern recognition systems (see, e.g., US 2017 / 199159 A1 ).

[0007] Moreover, US 2008 / 050839 A1 , and US 2019 / 317079 A1 disclose methods for identifying an analyte comprising: receiving image data from a sensor array that has been exposed to the analyte and which is configured to respond to the presence of the analyte, processing the image data to derive one or more input image features, and using a trained machine learning (ML) classification technique, detecting the analyte and classifying the analyte based on the one or more input image features.

[0008] However, in the above methods, machine learning techniques such as neural networks and statistical pattern recognition are used to process and analyze the data received from the sensor array by comparison with a database previously established by measuring defined analytes as a reference. Therefore, in order to identify and / or classify an unknown analyte, the ML algorithms have to be extensively trained on the same feature vectors to establish the required reference database beforehand, and analytes not recorded in the database tend to remain unidentified even if they are similar in chemical properties or composition.

[0009] In view of the above, it remains desirable to provide a cost-effective multisensor array and a method which enable rapid detection and / or classification of one or more analyte(s) in the gas phase or in the liquid phase without exclusively relying upon extensive data generation and training.

[0010] Simultaneously, it would be desirable to provide a multisensor device which may be fabricated in a compact and light-weight form amenable for field deployment, e.g. as a handheld device, and which may be used by untrained personal without specific method knowledge or analytical skills.SUMMARY OF THE INVENTION

[0011] The present invention solves this object with the subject matter of the claims as defined herein. The advantages of the present invention will be further explained in detail in the section below and further advantages will become apparent to the skilled artisan upon consideration of the invention disclosure.

[0012] Generally speaking, in a first aspect the present invention provides a multisensor array comprising multiple sensors for detection of one or more analyte(s) in the gas phase or in the liquid phase, wherein the multiple sensors comprise: a first sensor configured to detect a variation of an electrical parameter based on a dipole-dipole interaction with the analyte(s); asecond sensor configured to detect a variation of an electrical parameter based on a hydrogen bonding interaction with the analyte(s); and a third sensor configured to detect a variation of an electrical parameter based on a dispersive interaction with the analyte(s).

[0013] In a second aspect, the present invention provides a multisensor system for classification and / or identification of one or more analyte(s) in gas or liquid phase comprising: the multisensor array according to the first embodiment; and a processing unit comprising a pattern recognition and machine learning analyzer wherein the pattern recognition and machine learning analyzer is configured to receive output signals of the multisensor array and compare them to stored data in order to classify and / or identify the one or more analyte(s); wherein the stored data includes Hansen solubility parameters obtained from a library database.

[0014] In a third aspect, the present invention relates to a method of analyzing a sample present in gas or liquid phase, comprising the steps of: a) introducing the sample into a multisensor array according to the first embodiment described above; and b) applying a pattern recoginition and / or machine learning algorithm to data received from at least one of the first, second or third sensors to classify and / or identify the one or more analyte(s) in the sample; wherein in step b), the received data is correlated with Hansen solubility parameters obtained from a library database.BRIEF DESCRIPTION OF THE DRAWINGS

[0015] FIG. 1 is a Teas diagram projecting three dimensional Hansen solubility parameters onto a triangular space

[0016] FIG. 2 schematically depicts an exemplary detector channel as a part of a multisensor array.

[0017] FIG. 3A schematically illustrates the top view onto a channel system having the shape of a Hilbert curve.

[0018] FIG. 3B schematically illustrates the sensor configuration according to Fig. 3A with functional nanotubes randomly deposited thereon.

[0019] FIG. 4 schematically illustrates an exemplary configuration of a multisensor system according to the present invention.DETAILED DESCRIPTION OF THE INVENTION

[0020] For a more complete understanding of the present invention, reference is now made to the following description of the illustrative embodiments thereof:Multisensor Array

[0021] In a first embodiment, the present invention generally relates to a multisensor array comprising multiple detector channels for detection of one or more analyte(s) in the gas phase or in the liquid phase, wherein the multiple sensors comprise: a first sensor configured to detect a variation of an electrical parameter based on a dipole-dipole interaction with the analyte(s); a second sensor configured to detect a variation of an electrical parameter based on a hydrogen bonding interaction with the analyte(s); and a third sensor configured to detect a variation of an electrical parameter based on a dispersive interaction with the analyte(s).

[0022] Advantageously, by functionalizing the sensors so as to be reactive to non-bonding intermolecular forces, including dipole-dipole interaction (first sensor), H-bond interaction (second sensor) and London dispersion forces, also referred to instantaneous dipole-induced dipole forces (third sensors), sensor data may be correlated with Hansen solubility parameters.

[0023] The Hansen Solubility Concept (see C. Hansen, “The Three Dimensional Solubility Parameter and Solvent Diffusion Coefficient and Their Importance in Surface Coating Formulation". PhD Thesis 1967, Copenhagen: Danish Technical Press) has been originally developed as a way of predicting if one material is solved in another, based on the idea that if the parameters are similar, a good solubility is established. For this purpose, each molecule is given three Hansen parameters, each generally measured in MPa05at 298.15 K (equivalent to joules / cubic centimeter; 2.0455 x (cal / cc)1 / 2):

[0024] 62OT= 8 + 82+ 82(Eq.1 )

[0025] Herein, 6D, bp and 5H represent the dispersion, polar and hydrogen bonding interactions respectively. These three parameters can be treated as coordinates for a point in three dimensions also known as the Hansen space. The nearer the Hansen parameters of two substances are in this three-dimensional space, the more likely they are to dissolve into each other. To calculate the distance (Ra) between Hansen parameters in Hansen space the following formula is used:

[0026] (7?a)2= 4(<JD1- 6D2)2+ (5P1- SP2)2+ (5H1- 8H2)2(Eq. 2)

[0027] Herein 6DI , 6PI and 6m represent the dispersion, polar and hydrogen bonding interactions of a first substance, respectively, and 6D2, 6P2 and 5H2 represent the dispersion, polar and hydrogen bonding interactions of a second substance, respectively.

[0028] Molecules in fluid phase are generally interacting by such intermolecular forces, and every compound class has functional groups contributing to one or the other force and providing an indication of the molecule or compound class (e.g., water, hydrocarbons, alcohols, ethers, esters, chlorinated liquids, nitrogen compounds, aromatics or fluorinated liquids), which is illustrated by Fig. 1 , depicting a Teas fractional solubility diagram showing solubility parameters of common solvents and families of solvents with similar properties, wherein Fh represents the hydrogen bonding force, Fprepresents the polar force and Fd represents the dispersive force, each in [%].

[0029] An extensive database of Hansen solubility parameters for a large number of compounds has been compiled (cf. Hansen Solubility Parameters: A User’s Handbook, 20072ndEd., Boca Raton, CRC Press) and solubility parameters for an unknown solvent or solute can be calculated by using the software “Hansen Solubility Parameters in Practice (HSPiP)” (available under https: / / ww.hansen-solubility.com / buv-HSPiP-software.php).

[0030] Exemplary Hansen parameters of a number of selected analytes are shown in the following Table 1 (data retrieved from Hansen Solubility Parameters: A User’s Handbook, 20072ndEd., Boca Raton, CRC Press)

[0031] Table 1 : Hansen Solubility Parameters of selected analytes.

[0032] The present invention advantageously makes use of the fact that upon exposure to the specific analyte substance, an analyzable electrical parameter of each sensor is changed and results in a pattern that can be analyzed, while the resulting pattern can be correlated to an already available database of tabulated Hansen solubility parameters, which allows the identification of substance classes and possibly even dedicated substances without elaborate previous training of pattern recognition algorithms with similar feature vectors.

[0033] The expression “electrical parameter” as used herein, denotes an electrical signal measured upon exposure of the individual sensor to the test sample and is typically selected from the group consisting of resistance, conductance, alternating current (AC), capacitance, impedance, inductance, electrical potential, and voltage threshold.

[0034] According to the present invention, a first sensor is configured to detect a variation of an electrical parameter based on a dipole-dipole interaction with the analyte(s), the second sensor is configured to detect a variation of an electrical parameter based on a hydrogen bonding interaction with the analyte(s); and the third sensor configured to detect a variation of an electrical parameter based on a dispersive interaction with the analyte(s).

[0035] Each of the functions is achieved by different substituents or compounds chemically bound to or physically immobilized on a solid support, respectively, which may be independently selected for each sensor of the array.

[0036] The solid support of the sensor array can be based on silicon (e.g., poly- and monocrystalline silicon, silicon dioxide, silicon nitride, silica, glass, controlled pore glass, silica gel), metals, metal alloys or metal oxides (e.g., gold, platinum, silver, titanium, vanadium, chromium, iron cobalt, nickel, copper, palladium, aluminum, gallium, germanium, indium, zinc, and oxides thereof, alloys like cadmium telluride and copper-indium-selenide), synthetic or natural polymers, carbon (e.g., glassy carbon, carbon paste, graphite, carbon nanostructures).

[0037] In a particularly preferred embodiment, one or more of the first, second or third sensors and more preferably all of the three sensors comprise functionalized carbon nanostructures.

[0038] The term “carbon nanostructures”, as used herein, comprises single-wall carbon nanotubes, multi-wall carbon nanotubes, amorphous carbon, 2D-3D graphene structures, fullerites, fullerene molecules, buckyballs, carbon nano wires, carbon nano whiskers, carbon nano horns, carbon nano cones, carbon nano fibers, carbon nano ribbons, carbon nano spheres, carbon nano rods, carbon nano bowls, carbon peapod structures, carbon nano onions, carbon nanotori, carbon nano buds, or any combination thereof. From the viewpoint of ease of synthesis and functionalisation, carbon nanotubes (CNT) or graphene are preferably used as carbon nanostructures.

[0039] For purposes of the present invention, the term “carbon nanotube (CNT)”, unless specified otherwise, refers to any type of carbon nanotube. CNTs typically exist as single layers or multiple layers of cylindrical layers of graphene sheets. The individual sheets can vary in layering, and functionality. For example, CNTs can exist as single-walled CNTs (SWCNT), and multi-walled CNTs (MWCNT). The CNTs of an array may also be chiral, achiral, open headed, capped, budded, coated, uncoated, functionalized, neat, anchored, unanchored, basal plane, edge plan, step, or have any other known configuration. For the purposes of the present invention, single-walled CNTs (SWCNT) are particularly preferred. While not being limited thereto, CNTs typically have diameters ranging from about 0.6 nm to about 100 nm and lengths ranging from about 50 nm to about 10 mm. In preferred embodiments, the CNTs with a diameter in the range of ranging from about 0.7 nm to about 20 nm and lengths ranging from about 0.5 pm to about 250 pm are used.

[0040] The term “functionalized”, as used herein, refers to the presence of one or more substituents bound, complexed or otherwise associated with a solid support such as a carbon nanostructure.

[0041] For instance, CNTs may be chemically functionalized by methods known to the skilled artisan, by endohedral functionalization, non-covalent functionalization, sidewall functionalization and defect functionalization. Exemplary methods are disclosed and reviewed in J. L. Bahr et al., Chem. Commun. 2001 , 193; S. Niyogi et al., Acc. Chem. Res. 2002, 35, 1105; A. Hirsch, Angew. Chem. Int. Ed. Engl. 2002, 41 , 1853; S. Sinnott, Journ. Nanosci.Nanotechn. 2002, 2, 113; and A. Hirsch et al., Top. Curr. Chem. 2005, 245, 193, for example. Among these methods, non-covalent functionalization, which is based on the ability of the extended n-system of the carbon nanotubes sidewall to bind guest molecules via TT-n-stacking interactions, and sidewall functionalization may be preferred. Functionalization of fullerenes can be achieved either through covalent means due to the electrophilic nature of 060 (e.g. nucleophilic addition) or non-covalent means which include hydrogen bonding, TT-TT- interactions and metal ion coordination. Graphene may be functionalized by covalent functionalization based on binding of organic functionalities like free radicals (e.g. benzoyl peroxide, nitrobenzyl radicals) and dienophiles (e.g., azomethine diylide, phenyl azides, alkyl azides) on pristine graphene or attachment through the chemistry of oxygen groups of graphene oxide.

[0042] Notably, it is also possible to incorporate the required functional groups (examples of which will be further discussed below) in polymer chains, which may be grafted onto the carbon nanostructures by covalent bonding or which may be physically bonded to the carbon nanostructures by adsorption or wrapping of the polymer to the surface of carbon nanostructures, thus providing a polymer-carbon nanostructure composite.

[0043] Suitable functional substituents are selected by the skilled artisan by their ability to induce a change of a measurable electrical parameter (i.e. resistance, voltage, current flow, capacitance or impedance) in the specified sensor upon interacting with the analyte(s).

[0044] According to the present invention, the first sensor is sensitive to dipole-dipole interactions, which is preferably brought about by functionalizing the carbon nanostructure with solvatochromic or dipolar groups or agents.

[0045] The term “solvatochromic”, as used herein, refers to the phenomenon that the spectrum of a substance shifts upon interaction with the analyte(s). Different analytes induce a different effect on the electronic ground state and excited state of the solvatochromic group, so that the size of energy gap between them changes in dependence of the analyte(s). This is reflected in the absorption or emission spectrum of the solute as differences in the position, intensity, and shape of the spectroscopic bands. When the spectroscopic band occurs in the visible part of the spectrum solvatochromism is observed as a change of colour. Negative solvatochromism corresponds to a hypsochromic shift (or blue shift) with increasing polarity of the analyte(s) and positive solvatochromism corresponds to a bathochromic shift (or red shift) with increasing polarity of the analyte(s).

[0046] Examples of solvatochromic groups or agents include, but are not limited to 4- dicyanmethylene-2-methyl-6-(p-dimethylaminostyryl)-4H-pyran (DCM), 6-propionyl-2- (dimethylamino)-naphthalene (PRODAN), 9-(diethylamino)-5H-benzo[a]phenoxazin-5-one (Nile Red), 4-(dicyanovinyl)julolidine (DCVJ), phenol blue, stilbazolium dyes, coumarin dyes,ketocyanine dyes, Reichardt's dyes (e.g., Reichardt's Betaine dye (2,6-diphenyl-4-(2,4,6- triphenylpyridinio) phenolate)), merocyanine dyes (e.g., merocyanine 540), N,N-dimethyl-4- nitroaniline (NDMNA), N-methyl-2-nitroaniline (NM2NA), 4-(4'-hydroxystyryl)-N- methylpyridinium iodide, 4,4'-bis(dimethylamino)fuchsone, Nile Blue, 1-anilinonaphthalene-8- sulfonic acid (1 ,8-ANS), dapoxylbutylsulfonamide (DBS), Rhodamine B, Auramine O and derivatives thereof. In a preferred embodiment, the solvatochromic dye is selected from Reichardt's dyes. Notably, the solvatochromic group may be covalently attached to a carbon nanostructure or non-covalently bound or immobilized in the first sensor.

[0047] As an alternative or in addition to solvatochromic groups, dipolar groups may be incorporated, which increase sorption of dipolar analytes by oriented dipole-dipole interactions. Preferred examples of dipolar groups include substituents comprising cyano groups, such as cyanoalkyl or cyanoalkoxy groups.

[0048] According to the present invention, the second sensor is sensitive to H-bonding interactions with the analyte(s), which is preferably brought about by functionalization (e.g., of a carbon nanostructure) with a functional group comprising one or more and preferably a plurality of highly electronegative atoms, such as halogen atoms (e.g., F, Cl, I and Br), oxygen atoms, sulfur atoms and / or nitrogen atoms.

[0049] Preferred examples of halogen-containing functional groups are fluorinated groups, such as perfluoroalkyl, perfluoroaryl or perfluorophenylazide (PFPA) groups. In terms of covalent bonding to carbon nanostructures, PFPA and its derivatives are particularly preferred in view of their highly reactive azide group, which facilitates bond formation with graphene (L. H. Liu et al., Nano. Lett. 2009, 9, 3375-3378; L. H. Liu et aL, J. Mater. Chem. 2010, 20, 5041- 5046), CNTs (S. J. Pastine et al., J. Am. Chem. Soc. 2008, 4238-4239), and fullerenes (M. Yan et al., J. Org. Chem. 1994, 59, 5951-5954), for example.

[0050] Preferred nitrogen-containing groups are selected from primary, secondary and tertiary amines, amides, amine-N-oxides and phosphazenes and their derivatives. While not being limited thereto, examples of sulfur-containing groups include sulfoxides. As non-limiting examples of oxygen atom-containing groups, hydroxy groups (including, for example, fluoroalcohol, phenol or substituted phenol groups) and ester groups may be mentioned.

[0051] The third sensor is configured to detect a variation of an electrical parameter based on a dispersive interaction with the analyte(s). Suitable substituents for this purpose include, but are not limited to, aliphatic hydrocarbons, which may be open-chained (branched or linear) or cyclic, saturated or unsaturated. As specific non-limiting examples, propyl, isopropyl, n-butyl, t-butyl, isobutyl, hexyl, heptyl, cyclohexyl, 2-ethylhexyl, octyl, isooctyl, nonyl, isononyl, decyl, isodecyl, undecyl, dodecyl, tridecyl, tetradecyl, butadienyl, prenyl, cyclohexyl, ethynyl, propynyl, butynyl groups and combinations thereof may be mentioned.

[0052] The functional groups mentioned above in conjunction with the first to third sensors may also be incorporated in polymers, which are immobilized on the solid support by chemical bonding, physical methods or combinations thereof. Examples of polymer-CNT and polymer- graphene composites and their use for chemical sensing are reviewed in H. J. Salavagione et al., J. Mater. Chem. A 2014, 2, 14289 and may be suitably adopted for the purpose of the present invention. Suitable physical deposition techniques include vapor deposition or deposition of layers from aqueous solutions, for example.

[0053] It will be understood that a single sensor may be sensitive to more than one type of interaction with the analyte (e.g. basicity and dipolarity). In the present invention, however, it is preferred that such mixed-type sensors are provided in addition to the first to third sensors in order to minimize interference with respect to the individual Hansen solubility parameters while providing additional information about the analyte(s).

[0054] In a preferred embodiment of the invention, the multisensor array further comprises a fourth sensor comprising a pH-sensitive functionality, which provides additional information on the analyte(s). Suitable pH sensors are not particularly limited and may be suitably selected by the skilled artisan. As examples, CNT-based pH sensors (see US 2014 / 0330100 A1 and ON 114624302 A, for example), graphene-based pH sensors (see US 2019 / 0079043 A1 , for example), hydrogel-based pH sensors (see A. Richter et al., Sensors 2008, 8, 561-581 ; and N.F. Sheppard, Sens. Actuat. B 1995, 28, 95-102, for example), and ion-sensitive field effect transistor (ISFET)-based pH sensors or the like may be mentioned. Preferred examples include CNTs modified with functional groups that act as Brbnstedt acids or bases for pH estimation.

[0055] In a preferred embodiment of the invention, the multisensor array further comprises a fifth sensor configured to detect a variation of an electrical parameter based on redox reactions with the analyte(s). Such redox sensors are not particularly limited and may be suitably selected by the skilled artisan. As examples, CNT-based redox sensors (see US 2007 / 0278111 A1 , for example), carbon nanostructured-based redox sensors (see WO 2019 / 074618 A2, for example) or the like may be mentioned. Preferred examples include CNTs modified with redox-active functional groups.

[0056] By employing the first to fifth sensors simultaneously, a 5-dimensional feature space is opened by measured data allowing a more precise classification of substances based on existing literature tables as data input for the pattern recognition algorithm.

[0057] In order to minimize errors and thus improve selectivity and / or sensitivity of the multisensor array, it is preferable to employ several of each of the first to third sensors (as well as the fourth and fifth sensors, if present) to determine a mean and a standard deviation based on multiple readings.

[0058] A preferred embodiment of the present invention relates to a multisensor array, comprising independent detector channels for each sensor, wherein a first detector channel comprises a substrate comprising an insulating layer, a pair of electrodes formed on said insulating layer, and a semiconductive material coupled to each of the electrodes and comprising the first sensor; wherein a second detector channel comprises a substrate comprising an insulating layer, a pair of electrodes formed on said insulating layer, and a semiconductive material coupled to each of the electrodes and comprising the second sensor; wherein a third detector channel comprises a substrate comprising an insulating layer, a pair of electrodes formed on said insulating layer, and a semiconductive material coupled to each of the electrodes and comprising the third sensor; wherein the semiconductive material is selected from functionalized carbon nanostructures, preferably from functionalized carbon nanotubes according to the above description; wherein each pair of electrodes is configured as a conductive layer patterned so as to form a channel separating one electrode from another; and wherein the channels the between each pair of electrodes are partially or entirely filled with the respective semiconductive material to enable conduction of electric current.

[0059] Optionally, the aforementioned multisensor array may additionally comprise one or two additional independent detector channels selected from: a fourth detector channel comprising a substrate comprising an insulating layer, a pair of electrodes formed on said insulating layer, and a semiconductive material coupled to each of the electrodes and comprising the fourth sensor; and a fifth detector channel comprising a substrate comprising an insulating layer, a pair of electrodes formed on said insulating layer, and a semiconductive material coupled to each of the electrodes and comprising the fifth sensor.

[0060] Improved selectivity and / or sensitivity of the multisensor array may be achieved by incorporating a multiplicity (e.g., three or more) of each of the first to third detector channels (as well as the fourth and fifth detector channels, if present) to identify outliers, sensor defects and to improve the accuracy of the measurement by averaging multiple readings.

[0061] Notably, the substrates and their constituents may be independently selected for each of the detector channels. For example, the insulating layer may be selected from dielectric inorganic materials (e.g. SiO2) or organic materials (such as PET, PEN, paper and / or epoxy resin). The electrodes may be independently selected from any conductive material, including metals (such as copper, gold, palladium, silver, aluminum, chromium, etc.) and organic conductors (e.g. conductive polymers).

[0062] While not being limited thereto, the channel separating the electrodes from each other may be formed by laser patterning, lift off, lithographical methods, direct laser patterning, ink jet printing or screen printing, for example. These methods are preferable as they enable low- cost and high-throughput manufacturing.

[0063] An example of a single detector channel (1 ) is illustrated in Fig. 2, comprises a substrate comprising an insulating layer (2), wherein a channel (3) is formed in a conductive layer (4) so as to provide a pair of electrodes separated from one another. A semiconductive material, exemplarily in the form of functionalized carbon nanotubes (5), serves as the sensing element.

[0064] As depicted in Fig. 2, the channel (3) separating one electrode from another may have a serpentine shape. However, in a preferred embodiment, the channel has the geometric shape of a FASS curve, wherein the acronym FASS denotes a space-filling, self-avoiding, simple and self-similar curve. The feature of space-filling refers to the fact that the curve passes within a small distance from all points of a given two-dimensional figure, such as a square, which surrounds the curve. This distance can be arbitrarily reduced by carrying on the recursive construction to an appropriate level. Self-avoidance means that the segments of the curve do not touch nor intersect. Simplicity indicates that the curve can be drawn by a single stroke of a pen, without lifting the pen nor drawing any segments more than once. The feature of self-similarity indicates that the curve can be constructed by the recursive application of a set of rules for connecting components. In most cases, it is additionally assumed that the curves consist of straight line segments which run vertically or horizontally along the edges of a square grid. Examples of FASS curves and methods for their construction are described in the document “Synthesis of Space-filling Curves on the Square Grid” by P. Prusinkiewiczl, A. Lindenmayer and F. D. Fracchia (published in the textbook “Fractals in the Fundamental and Applied Sciences”, edited by H.-O. Peitgen, J.M. Henriques & L.F. Penedofor Elsevier Science Publishers B.V. in 1991 ).

[0065] Advantageously, a channel shaped as a FASS curve maximizes the probability of functionalized CNTs contacting the channel and bridging the pair of electrodes, even if the CNTs are deposited randomly on the substrate (as is the case in Fig. 2). The increased number of contacts between the nanotubes and the channels also increases the current flowing between the electrodes and optimizes the signal-to-noise ratio.

[0066] According to a further preferred embodiment of the present invention, the FASS curve comprises a Hilbert curve, which can be advantageously designed so that the ends of the Hilbert curve are on opposite sides of the electrode contact area, in such a way that it is possible to form two distinct electrodes. In addition, a predefined extension can be obtained by simply increasing the number of iterations while constructing the curve.

[0067] Figure 3A schematically illustrates the top view onto a channel system having the shape of a Hilbert curve (with a Lindenmayer number equal to 2), wherein the channel (10) is formed on the substrate layer (11 ) and separates a pair of electrodes, i.e. a first electrode (12) and a second electrode (13). The channel (10) may be partially or totally filled with a semiconductor material. For instance, as illustrated in Figure 3B, a plurality of functionalized CNTs (14) may be randomly deposited on the electrodes (12) and (13). Since the Hilbert curve is a spacefilling curve, the probability that one or more CNTs (14) overlap the channel (10) at least in one point is maximized. Depending on the length of the CNTs (14), it is also possible that one or more single CNTs (14) overlap the channel (10) in multiple points, hence, enhancing the performance of the device. In addition, the extension or width of the curve can be increased by drawing a Hilbert curve without affecting the area (A) available for the electrodes in the detector channel.

[0068] In another preferred embodiment, the FASS curve comprises a Peano curve, which can also be constructed in a way that it allows rounded edges, which may simplify and / or render the manufacturing process more efficient.

[0069] According to another preferred embodiment of the present invention, one or more of the pairs of electrodes may form side walls facing the channel dividing them, wherein the side walls are configured so as to have a tilted or curved profile (in side view). The advantage of this configuration is that the contact between the side walls of the electrodes and a semiconducting material filling the channel can be optimized.

[0070] The multisensor arrays described in the embodiments above have the advantage that they may be produced in an inexpensive, compact and light-weight manner. Accordingly, in an especially preferred embodiment, the multisensor array is provided within a low-power handheld device, which may be deployed for field measurements.Multisensor System

[0071] In order to enhance the measurement accuracy, to enable improved discrimination between multiple target compounds or classes in a single analyte sample (e.g. reduce crosstalk) and / or to account for necessary calibration shifts, the multisensor array according to the present invention may be combined with machine learning tools to analyse the data retrieved from the sensing elements.

[0072] In a second embodiment, the present invention therefore relates to a multisensor system for classification and / or identification of one or more analyte(s) in gas or liquid phase comprising: the multisensor array according to the first embodiment; a processing unit comprising a pattern recognition and machine learning analyzer wherein the pattern recognition and machine learning analyzer is configured to receive output signals of themultisensor array and compare them to stored data in order to classify and / or identify the one or more analyte(s); wherein the stored data includes Hansen solubility parameters obtained from a library database.

[0073] The term “library database”, as referred to herein, includes a set of solubility parameters retrieved from the textbook Hansen Solubility Parameters: A User's Handbook, 2007 2ndEd., Boca Raton, CRC Press or calculated by using the software “Hansen Solubility Parameters in Practice (HSPiP)”.

[0074] It will be understood that the output signal originating from the first sensor of the multisensor array is correlated with the Hansen parameter bp for polarity, the output signal originating from the second sensor of the multisensor array is correlated with the Hansen parameter bnfor hydrogen bonding, and the output signal originating from the third sensor of the multisensor array is correlated with the Hansen parameter 5D for dispersion.

[0075] The multisensor system may further comprise additional components, including, for example: a communication circuitry configured to receive data gathered based on the variation of electrical parameters from at least one of the first, second and third detector channels; a pre-processing unit configured to generate a response pattern; a memory configured to store the data and pattern recognition and machine learning algorithms; and a processing circuitry configured to apply the pattern recognition and machine learning algorithms to the data and to probabilistically classify and / or identify the one or more analyte(s). An exemplary scheme illustrating an example of the multisensor system is shown in Fig. 4.

[0076] The multisensor system may communicate data to an application executing on a user device, such as a smartphone, a tablet, or a virtual assistant enabled device, through wireless and wired communication. For example, the multisensor system may include a wireless network interface (e.g., WiFi®, Bluetooth®, or the like) to enable the system to communicate with other sensor devices (e.g. in an Internet-of-Things (loT)-setting), an application executing on a user's computing device or a cloud-based server, for example.

[0077] The multisensor system may be provided as a standalone device or as a kit of parts, wherein the constituents or combinations of the constituents are provided separately. For instance, the processing unit or components thereof may be provided independently from the multisensor array, e.g. by means of a computing device (PC or smartphone) in wired or wireless connection with the multisensor array.

[0078] Standalone devices may comprise a display configured for viewing analytical data and the measurement result, including the identified analyte(s) or analyte class. Information on the operational mode (e.g., detection mode or low-power mode), network connectivity and other information associated with the multisensor array may also be displayed, for example. In addition, standalone devices may be battery-powered.

[0079] The signal processing to determine the analyte(s) interacting with the sensor array may include a probabilistic determination of the class of the analyte(s) rather than an accurate determination of the analyte(s), enabling a rapid assessment of the presence of substance classes of interest, such as potentially dangerous or harmful compound classes.

[0080] Correlating the data received from the multisensor array with Hansen solubility parameters further enables a probabilistic determination of compound classes in the analyte unknown to the pattern recognition and machine learning analyzer (i.e. on which the pattern recognition and machine learning analyzer has previously not been trained), thus providing added value over conventional sensor systems employing chemical sensors in combination with machine learning tools.

[0081] The pattern recognition and machine learning analyzer employed in the multisensor system according to the present invention is configured to apply at least one algorithm in order to probabilistically classify and / or identify the one or more analyte(s).

[0082] In preferred embodiments, the at least one algorithm is selected from the group consisting of: artificial neural network (ANN) algorithms, principal component analysis (PCA), support vector machine (SVM), decision tree (DT), random forest (RF), voting classifier (VC), multi-layer perception (MLP), generalized regression neural network (GRNN), fuzzy inference systems (FIS), self-organizing map (SOM), radial bias function (RBF), neuro-fuzzy systems (NFS), adaptive resonance theory (ART), partial least squares (PLS), linear regression (LR), multiple linear regression (MLR), principal component regression (PCR), discriminant function analysis (DFA), linear discriminant analysis (LDA), cluster analysis, and nearest neighbor algorithms.

[0083] In further preferred embodiments, the at least one algorithm is selected from one or more of support vector machine (SVM), nearest neighbor algorithms (e.g., K-nearest neighbor (KNN)), decision tree (DT), random forest (RF), and voting classifier (VC) algorithms.

[0084] In machine learning-enabled sensor systems, the data generated by the sensors is collected and compiled into an unprocessed dataset, which is then preprocessed to extract relevant features from the dataset to create a so-called smart model. Once adequate features are extracted, they may be recompiled into a sub-dataset for training and a sub-dataset for validation and testing of the ML-enabled sensor system. Accordingly, algorithms selected to define the smart model for the specific sensor application should be ideally validated and tested with real-life samples.Method for Analyzing Chemical Samples

[0085] In a third embodiment, the present invention relates to a method of analyzing a sample present in gas or liquid phase, comprising the steps of: a) introducing the sample into amultisensor array according to the first embodiment described above; and b) applying a pattern recognition and / or machine learning algorithm to data received from at least one of the first, second or third sensors to classify and / or identify the one or more analyte(s) in the sample; wherein in step b), the received data is correlated with Hansen solubility parameters obtained from a library database.

[0086] The method of sample introduction is not particularly limited. For example, in gas phase, the multisensor array can be exposed to saturated vapours of analytes, or vapours can be into the independent detector channels using a flow cell and a pump. In liquid phase, the multisensor array may be immersed into the sample liquid to introduce the analyte(s) into the detector channels.

[0087] In a preferred embodiment, the method makes use of pattern recognition and machine learning algorithms as described above in conjunction with the second embodiment.

[0088] By correlating the output signal originating from the first sensor with the Hansen parameter bp for polarity, the output signal originating from the second sensor of the multisensor array with the Hansen parameter 6H for hydrogen bonding, and / or the output signal originating from the third sensor of the multisensor array with the Hansen parameter 6D for dispersion, the method of the present invention advantageously enables rapid detection and / or classification of one or more analyte(s) in the gas phase or in the liquid phase without exclusively relying upon data generation and training via machine learning.

[0089] It will be appreciated that the present invention may employ any of the preferred features specified above with respect to the description of the first to third embodiments, and that the preferred features may be combined in any combination, except for combinations where at least some of the features are mutually exclusive.

[0090] In general, the analyzable samples are not particularly limited. The multisensor array according to the present invention is sensitive towards a vast number of analytes and mixtures thereof. Exemplary analytes and mixtures thereof include aliphatic or aromatic hydrocarbons, compounds with groups containing halogen, oxygen, nitrogen, sulfur, phosphorus, boron, and / or silicon atoms. Typical analytes include ammonia, amines, amides, amidines, carboxamides, amidines, imides, azides, nitrates, nitriles, isonitriles, pyridine derivatives, carbamates, thiols, sulfides, solfoxides, disulfides, sulfones, sulfinic acids, sulfanate esters, thiocyanates, thioketones, phosphines, phosphonic acids, phosphates, boronic esters, alcohols, aldehydes, acyl halides, carboxylic acids, organic acid anhydrides, lactones, peroxides, hydroperoxides, ethers, amino acids, alkanes, aromatics, aliphatics, halogen compounds, inorganic acids and gases (e.g. CO2).

[0091] The multisensor array and system according to the present invention is especially effective for classification and / or identification of liquid analytes and analyte mixtures. Asexamples thereof, volatile organic compounds (VOC), which are understood as organic compounds having a vapour pressure of 0.01 kPa or more at 293.15 K, may be mentioned.

[0092] The present invention may be used to rapidly detect and identify VOCs, aroma compounds, explosives, toxic compounds, toxins, drugs, drug precursors, narcotics, exhaust gases and environmental pollutants, for example. Accordingly, the fields of use are widespread and include, but are not limited to industry, safety, public, military, construction work and environmental monitoring. Furthermore, since they do not require any specific method knowledge or analytical skills, the multisensor arrays and multisensor systems according to the present invention are well-suited for home use.

[0093] Once given the above disclosure, many other features, modifications, and improvements will become apparent to the skilled artisan.

[0094] REFERENCE NUMERALS1 detector channel2 insulating layer3 channel between electrodes4 conductive layer5 functionalized carbon nanotubes (CNTs)10 channel between electrodes11 substrate layer12 first electrode13 second electrode14 functionalized carbon nanotubes (CNTs)A area

Claims

CLAIMS1. A multisensor array comprising multiple sensors for detection of one or more analyte(s) in the gas phase or in the liquid phase, wherein the multiple sensors comprise: a first sensor configured to detect a variation of an electrical parameter based on a dipole-dipole interaction with the analyte(s), a second sensor configured to detect a variation of an electrical parameter based on a hydrogen bonding interaction with the analyte(s), and a third sensor configured to detect a variation of an electrical parameter based on a dispersive interaction with the analyte(s).

2. The multisensor array according to claim 1 , wherein the first sensor is functionalized with a solvatochromic group, the solvatochromic group being preferably selected from one or more of 4-dicyanmethylene-2- methyl-6-(p-dimethylaminostyryl)-4H-pyran (DCM), 6-propionyl-2-(dimethylamino)- naphthalene (PRODAN), 9-(diethylamino)-5H-benzo[a]phenoxazin-5-one (Nile Red), 4- (dicyanovinyl)julolidine (DCVJ), phenol blue, stilbazolium dyes, coumarin dyes, ketocyanine dyes, Reichardt's dyes, merocyanine dyes, N,N-dimethyl-4-nitroaniline (NDMNA), N-methyl-2- nitroaniline (NM2NA), 4-(4'-hydroxystyryl)-N-methylpyridinium iodide, 4,4'- bis(dimethylamino)fuchsone, Nile Blue, 1-anilinonaphthalene-8-sulfonic acid (1 ,8-ANS), dapoxylbutylsulfonamide (DBS), Rhodamine B, Auramine O and derivatives thereof, more preferably from Reichardt's dyes; and / or wherein the first sensor is functionalized with a dipolar group, the dipolar group being preferably selected from one or more of a cyano group, a cyanoalkyl group or a cyanoalkoxy groups.

3. The multisensor array according to claim 1 or claim 2, wherein the second sensor is functionalized with a group comprising one or more and preferably a plurality of electronegative atoms selected from halogen, oxygen atoms, sulfur atoms and / or nitrogen atoms; the halogen-containing functional groups being preferably selected from fluorinated groups, more preferably perfluoroalkyl, perfluoroaryl or perfluorophenylazide (PFPA) groups; the oxygen atom-containing groups being preferably selected from hydroxy groups and ester groups;the nitrogen-containing groups being preferably selected from primary, secondary and tertiary amines, amides, amine-N-oxides and phosphazenes and their derivatives, and the sulfur-containing groups being preferably selected from sulfoxides.

4. The multisensor array according to any of claims 1 to 3, wherein the third sensor is functionalized with an aliphatic hydrocarbon group.

5. The multisensor array according to any one of claims 1 to 4, further comprising a fourth sensor comprising a pH-sensitive functionality.

6. The multisensor array according to claim 5, wherein the fourth sensor is selected from any of a carbon nanostructure-based pH sensor, a hydrogel-based pH sensor or a ion-sensitive field effect transistor (ISFET)-based pH sensor.

7. The multisensor array according to any one of claims 1 to 6, further comprising a fifth sensor configured to detect a variation of an electrical parameter based on redox reactions with the analyte(s).

8. The multisensor array according to claim 7, wherein the fifth sensor comprises a carbon nanostructure modified with redox-active functional groups.

9. The multisensor array according to any one of claims 1 to 8, wherein at least one and preferably at least two, and more preferably all three of the first, second and third sensors comprise functionalized carbon nanostructures as sensing elements.

10. The multisensor array according to claim 9, wherein the functionalized carbon nanostructures comprise single-wall carbon nanotubes, multi-wall carbon nanotubes, amorphous carbon, 2D graphene, 3D graphene, fullerites, fullerenes, carbon nano wires, carbon nano fibers, carbon nano ribbons, carbon nano spheres, carbon nano rods, or any combination thereof, the functionalized carbon nanostructures preferably comprising singlewall carbon nanotubes.

11. The multisensor array according to any one of claims 1 to 10, comprising independent detector channels for each sensor,wherein a first detector channel comprises a substrate comprising an insulating layer, a pair of electrodes formed on said insulating layer, and a semiconductive material coupled to each of the electrodes and comprising the first sensor; wherein a second detector channel comprises a substrate comprising an insulating layer, a pair of electrodes formed on said insulating layer, and a semiconductive material coupled to each of the electrodes and comprising the second sensor; wherein a third detector channel comprises a substrate comprising an insulating layer, a pair of electrodes formed on said insulating layer, and a semiconductive material coupled to each of the electrodes and comprising the third sensor; wherein the semiconductive material is selected from functionalized carbon nanostructures, preferably from functionalized carbon nanotubes; wherein each pair of electrodes is configured as a conductive layer patterned so as to form a channel separating one electrode from another; and wherein the channels the between each pair of electrodes are partially or entirely filled with the respective semiconductive material to enable conduction of electric current.

12. The multisensor array according to claim 11 , wherein at least one of the channels separating one electrode from another has the shape of a FASS curve, and the FASS curve preferably comprises a Peano curve or a Hilbert curve.

13. A multisensor system for classification and / or identification of one or more analyte(s) in gas or liquid phase comprising: the multisensor array according to any one of claims 1 to 12; and a processing unit comprising a pattern recognition and machine learning analyzer wherein the pattern recognition and machine learning analyzer is configured to receive output signals of the multisensor array and compare them to stored data in order to classify and / or identify the one or more analyte(s); wherein the stored data includes Hansen solubilty parameters obtained from a library database.

14. The multisensor system according to claim 13, wherein the pattern recognition and machine learning analyzer is configured to apply at least one algorithm in order to probabilistically classify and / or identify the one or more analyte(s), wherein the at least one algorithm is selected from the group consisting of: artificial neural network (ANN) algorithms, principal component analysis (PCA), support vector machine (SVM), decision tree (DT), random forest (RF), voting classifier (VC), multi-layer perception (MLP), generalized regression neural network (GRNN), fuzzy inference systems (FIS), self-organizing map (SOM),radial bias function (RBF), neuro-fuzzy systems (NFS), adaptive resonance theory (ART), partial least squares (PLS), linear regression (LR), multiple linear regression (MLR), principal component regression (PCR), discriminant function analysis (DFA), linear discriminant analysis (LDA), cluster analysis, and nearest neighbor algorithms; and preferably at least one algorithm selected from support vector machine (SVM), nearest neighbor, decision tree (DT), random forest (RF) and voting classifier (VC) algorithms.

15. A method of analyzing a sample present in gas or liquid phase, comprising the steps of: a) introducing the sample into a multisensor array according to any one of claims 1 to 12; and b) applying a pattern recoginition and / or machine learning algorithm to data received from at least one of the first, second or third sensors to classify and / or identify the one or more analyte(s) in the sample; wherein in step b), the received data is correlated with Hansen solubilty parameters obtained from a library database.

16. The method according to claim 15, wherein the pattern recognition and / or machine learning algorithm is applied to data received from of the first, second and third sensors to classify and / or identify the one or more analyte(s) in the sample.

17. The method according to claim 16, wherein the data received from the first sensor is correlated with a Hansen polarity parameter bp, the data received from the second sensor is correlated with a Hansen hydrogen bonding parameter 5H, and the data received from the third sensor of the multisensor array with a Hansen dispersion parameter 5D.

Images