Method and electronic device for estimating a set of component(s) of a product from a spectrometric device, computer program and associated measurement system

Abstract

Method and electronic device for estimating a set of component(s) of a product from a spectrometry device, associated computer program and measurement system. The invention relates to a method for estimating a set of component(s) of a product from a spectrometry device comprising an ionization source of an element, a module for separating the ions of the ionized element, and a detector connected to the separation module and capable of measuring an ionic flux of the element.This process is implemented by an electronic device capable of being connected to the spectrometry device and comprises the following steps: - obtaining (100) a group of signatures for potential components of the product, each signature being representative of an ionic flux of a respective potential component and measured via the spectrometry device; - acquiring (110) a quantity of an ionic flux of the product, measured via the spectrometry device; and - estimating (130) the set of product components from the group of signatures and the quantity, using an estimation algorithm. Figure 3 for the abstract.

Description

Title of the invention: Method and electronic device for estimating a set of component(s) of a product from a spectrometry device, computer program and associated measuring system

[0001] The present invention relates to a method for estimating a set of component(s) of a product from a spectrometry device, implemented by an electronic estimation device.

[0002] The invention also relates to a computer program comprising software instructions which, when executed by a computer, implement such an estimation method.

[0003] The invention also relates to an electronic device for estimating a set of component(s) of a product from a spectrometry device; and a measuring system comprising the spectrometry device and such an estimation device.

[0004] The invention then relates to the field of product analysis by spectrometry. Spectrometry is used in many technical fields, such as petrochemistry, pharmacy, gas phase chemistry, organic chemistry, physics, astrophysics, biology.

[0005] The development of the formula of a product, such as a lubricant, requires perfect knowledge of the components, such as base oils and additives, in order to have precise information on the evolution of the technological environment. Mass spectrometry and / or ion mobility spectrometry then make it possible to obtain a detailed structural characterization of the additives present in the formulated products.

[0006] Currently, in the context of identifying the components of a product, such as a lubricant, the methodology applied is a manual analysis of the spectral responses of the products. The detection of the components of each product is then carried out by verifying the presence of peaks or sets of peaks characteristic of the chemical species sought in the measured spectrum of each product. Theoretical and empirical expertise of the chemical properties of the components sought is required in order to identify the part of the data to be observed to infer the presence or absence of a respective component sought.

[0007] However, such manual analysis of the spectral responses of the products is tedious and often requires a fairly long analysis time. In addition, a relatively large portion of points of the spectral responses is not considered during the analysis, and the chemical components detectable among these points are then omitted.

[0008] The aim of the invention is therefore to propose a method, and an associated electronic device, for estimating a set of component(s) of a product from a spectrometry device, making it possible to improve the analysis of the spectral response of each product and then to estimate more effectively the composition of each product, i.e. the sets of component(s) included in each product.

[0009] For this purpose, the invention relates to a method for estimating a set of component(s) of a product from a spectrometry device, the spectrometry device comprising an ionization source inside which an element is capable of being ionized, a separation module connected to the output of the ionization source and capable of separating ions from the ionized element, and a detector connected to the output of the separation module and capable of measuring an ionic flux of the element,

[0010] the method being implemented by an electronic estimation device capable of being connected to the spectrometry device and comprising the following steps:

[0011] - obtaining a group of signatures for potential components of the product, each signature being representative of an ionic flux of a respective potential component, and measured via the spectrometry device;

[0012] - the acquisition of a magnitude of an ionic flux of the product, measured via the device of spectrometry;

[0013] - the estimation of the set of component(s) of the product from the group of si gnatures and magnitude and via an estimation algorithm.

[0014] Thus, the estimation method according to the invention makes it possible to automatically estimate the set of component(s) of the product, via the implementation of the estimation algorithm from the measured quantity and the group of signatures. The estimation method according to the invention is then more efficient and less tedious than the method of the state of the art based on a manual analysis of the spectral response of each product.

[0015] According to other advantageous aspects of the invention, the estimation method comprises one or more of the following characteristics, taken in isolation or in all technically possible combinations:

[0016] - the magnitude and each signature correspond to the same measurement modality chosen from the group consisting of:

[0017] + a mass spectrum if the separation module includes a spec cell mass trophicity,

[0018] + an ion mobility spectrum if the separation module includes a cell of ion mobility spectrometry; and

[0019] + a two-dimensional mapping if the separation module comprises a cell mass spectrometry and an ion mobility spectrometry cell, the two-dimensional mapping representing, according to their mass to charge ratio, an ion drift time in the ion mobility cell coupled to the mass spectrometry cell.

[0020] - the obtaining and acquiring steps are respectively implemented for several different measurement modalities, and during the estimation stage, the estimated set of product component(s) then depends on the signature groups and quantities for the different measurement modalities;

[0021] - the product is an additive linear mixture of the components where each component is weighted by a mixing coefficient, each mixing coefficient having a positive value; and in the estimation step, the quantity is modeled as a sum of the signatures of the group, each signature being weighted by the respective mixing coefficient, then the value of each mixing coefficient is calculated via the estimation algorithm;

[0022] - the estimation algorithm includes a minimization of a criterion dependent on the size and group of signatures;

[0023] the criterion preferably being a quadratic criterion, such as a least squares criterion, or a logarithmic criterion;

[0024] - the estimation algorithm further comprises a determination of the presence or no of each potential component in the product by comparison of each mixing coefficient with a respective detection threshold;

[0025] the detection thresholds being preferably predefined using a ROC curve from mixtures, real or simulated, of components;

[0026] - the estimation algorithm comprises an implementation of a decomposition algorithm parsimonious position from size and signature group;

[0027] the sparse decomposition algorithm preferably being chosen from the group of greedy algorithms consisting of: MP algorithm, OMP algorithm and OLS algorithm;

[0028] - an initial residue is equal to the measured quantity, and each subsequent residue is equal to the difference between the magnitude and a linear combination of signature(s) of component(s) already estimated to be present and respective mixing coefficient(s); and

[0029] - the method further comprises, prior to the estimation step, a step of factorization of the group of signatures and the quantity with the same common factor, each signature being represented in the form of a product of the common factor and a factored signature, the quantity being represented in the form of a product of the common factor and a factored quantity; and the estimation step then being carried out from the factored quantity and the group of factored signatures.

[0030] The invention also relates to a computer program comprising software instructions which, when executed by a computer, implement an estimation method, as defined above.

[0031] The invention also relates to an electronic device for estimating a set of component(s) of a product from a spectrometry device, the spectrometry device comprising an ionization source inside which an element is capable of being ionized, a separation module connected to the output of the ionization source and capable of separating ions from the ionized element, and a detector connected to the output of the separation module and capable of measuring an ionic flux of the element,

[0032] the electronic estimation device being capable of being connected to the spectrometry device and comprising:

[0033] - an obtaining module configured to obtain a group of signatures for potential components of the product, each signature being representative of an ionic flux of a respective potential component, and measured via the spectrometry device;

[0034] - an acquisition module configured to acquire a magnitude of an ionic flux of the product, measured via the spectrometry device; and

[0035] - an estimation module configured to estimate the set of component(s) of the produced from the signature group and the magnitude and via an estimation algorithm.

[0036] The invention also relates to a measuring system comprising a spectrometry device and an electronic device for estimating a set of component(s) of a product from the spectrometry device, the electronic estimation device being as defined above and connected to the spectrometry device.

[0037] These characteristics and advantages of the invention will appear more clearly on reading the description which follows, given solely by way of non-limiting example, and made with reference to the appended drawings, in which:

[0038] [Fig-1] [Fig.l] is a schematic representation of a measuring system according to the invention, comprising a spectrometry device and an electronic device for determining, via the spectrometry device, a set of component(s) of a sample;

[0039] [Fig.2] [Fig.2] illustrates different measurement methods possible with the system of measurement of [Fig.l], namely a mass spectrum if a separation module included in the spectrometry device of [Fig.l] comprises a mass spectrometry cell, an ion mobility spectrum if the separation module comprises an ion mobility spectrometry cell, and a two-dimensional map if the separation module comprises both a mass spectrometry cell and an ion mobility spectrometry cell, the two-dimensional map representing, as a function of their mass / charge ratio, a drift time of ions in the ion mobility cell coupled to the mass spectrometry cell;

[0040] [Fig.3] [Fig.3] is a flowchart of a method, according to the invention, for estimating a set of component(s) of a product using the spectrometry device of [Fig.l], the method being implemented by the electronic estimation device of [Fig.l], and the estimation of said set being carried out via an estimation algorithm;

[0041] [Fig.4] [Fig.4] is a ROC curve used to determine a detection threshold then used to determine the presence or absence of a potential component in the product, according to a first embodiment of the estimation algorithm;

[0042] [Fig.5] [Fig.5] is a diagram illustrating a list of components present in the product as estimated by the estimation device of [Fig.1] according to the estimation method of [Fig.3], in comparison with the list of components actually contained in the product, for a single measurement method and according to a first embodiment of the estimation algorithm;

[0043] [Fig.6] [Fig.6] is a view similar to that of [Fig.5], for a plurality of measurement modalities and according to the first embodiment of the estimation algorithm;

[0044] [Fig.7] [Fig.7] is a diagram representing for about twenty components a true positive rate, noted TPR; a false positive rate, noted FPR; a true negative rate, noted TNR; and respectively a false negative rate, noted FNR; according to the first embodiment of the estimation algorithm; and

[0045] [Fig.8] [Fig.8] is a view similar to that of [Fig.7], according to a second embodiment of the estimation algorithm.

[0046] In the remainder of the description, the expression “substantially equal to” designates a relationship of equality to plus or minus 10%, preferably to plus or minus 5%.

[0047] In [Fig.l], a measurement system 10 comprises a spectrometry device 12. The spectrometry device 12 comprises an ionization source 14 inside which an element is capable of being ionized; a separation module 16 connected to the output of the ionization source 14 and capable of separating ions from the ionized element; and a detector 18 connected to the output of the separation module 16 and capable of measuring an ionic flux for said element.

[0048] The measuring system 10 also comprises an electronic device 20 for estimating a set of component(s) of a product from the spectrometry device 12, the electronic estimation device 20 being connected to the spectrometry device 12.

[0049] The spectrometry device 12 is known per se, and makes it possible to detect and characterize molecules of interest of the element, by measuring their ionic mobility and / or their mass, and to characterize their chemical structure. The spectrometry is then based on the separation in the gas phase of charged molecules, namely ions, in function of their ionic mobility and / or their mass / charge ratio, also noted m / z.

[0050] As an optional addition, the spectrometry device 12 further comprises a chromatography module, not shown, connected upstream of the ionization source 14. The chromatography module comprises, for example, a liquid chromatography column, also referred to as LC (Liquid Chromatography), or a gas chromatography column, also referred to as GC (Gas Chromatography), such as an atmospheric pressure gas chromatography column, also referred to as APGC (Atmospheric Pressure Gas Chromatography).

[0051] The ionization source 14 is capable of vaporizing the molecules of the element and ionizing them. The ionization source 14 is capable of being used either in positive mode to study positive ions, or in negative mode to study negative ions.

[0052] The ionization source 14 is for example of the type chosen from the group consisting of: ionization with atmospheric solids analysis probe or AS AP (from the English Atmospheric Solids Analysis Probe), electronic ionization or El (from the English Electron lonization), chemical ionization or CI (from the English Chemical lonization), chemical desorption-ionization or DCI (from the English DesorptionChemical lonization); bombardment by fast atoms or FAB (from the English FastAtom Bombardment), bombardment by metastable atoms or MAB (from the English Metastable Atom Bombardment), bombardment by ions, such as SIMS (from the English Secondary-ion Mass Spectrometry), LSIMS (from the English LiquidSecondary-Ion Mass Spectrometry); Inductively Coupled Plasma (ICP); Ambient ionization, such as APCI (Atmospheric Pressure Chemical Ionization), DESI (Desorption ElectroSpraylonization), SESI (Secondary ElectroSpraylonization), LAESI (Laser Ablation ElectroSpraylonization); Atmospheric Pressure Photoionization (APPI); Electrospray ionization (ESI); laser desorption / ionization (LDI), matrix-assisted laser desorption / ionization (MALDI), surface-enhanced laser desorption / ionization (SELDI), desorption / ionization on silicon (DIOS);desorption ionization by interaction with metastable species or DART (Direct Analysis in Real Time); and thermal ionization or TIMS (Thermal Ionization Mass Spectrometry), such as ID-TIMS (Isotropic Dilution - Thermal Ionization Mass Spectrometry) or CA-TIMS (Chemical Abrasion - Thermal Ionization Mass Spectrometry).

[0053] The separation module 16 comprises, for example, two cells connected in cascade, namely an ion mobility spectrometry cell and a mass spectrometry cell connected to the output of the ion mobility spectrometry cell. The ion mobility spectrometry cell, also referred to as IMS (from the English lon-Mobility Spectrometry), is also called an ion mobility cell, and is capable of separating the ions according to their mobility. The mass spectrometry cell, also referred to as MS (from the English Mass Spectrometry), is capable of separating the ions according to their mass / charge ratio, also called mass to charge ratio or mass to charge ratio. When the separation module 16 comprises the mass spectrometry cell coupled to the ion mobility spectrometry cell, the spectrometry device 12 is then also called a coupled spectrometry device, also referred to as IMS-MS.Such a coupled spectrometry device is typically capable of producing two-dimensional maps. These two-dimensional maps are optionally formatted as mass or ion mobility spectra.

[0054] Alternatively, the separation module 16 comprises only the ion mobility spectrometry cell, and the spectrometry device 12 is then also called an ion mobility spectrometer. Such an ion mobility spectrometer is typically capable of producing spectra.

[0055] Those skilled in the art will also observe that spectra are likely to be obtained with the IMS-MS coupled spectrometry apparatus, or with an MS mass spectrometry cell connected to the input of an IMS ion mobility spectrometry cell, then of another MS mass spectrometry cell. In the latter case, the first MS mass spectrometry cell is not active to obtain spectra, and the spectrometry device 12 then operates in IMS-MS mode.

[0056] As a further variant, the separation module 16 comprises only the mass spectrometry cell, and the spectrometry device 12 is then also called a mass spectrometer.

[0057] As a further variant, the separation module 16 comprises two mass spectrometry cells, namely a first mass spectrometry cell and a second mass spectrometry cell, connected in cascade, i.e. coupled together. The spectrometry device 12 is then also called a tandem mass spectrometer, also referred to as tandem MS / MS. According to this addition, the first mass spectrometry cell is capable of separating the ions, a collision cell making it possible to fragment the ions, and the second mass spectrometry cell is capable of separating the fragment ions.

[0058] As a further variant, the separation module 16 comprises several spec cells ion mobility spectrometry connected in cascade, i.e. coupled together. The spectrometry device 12 is then also called a tandem ion mobility spectrometer, also noted tandem IMS / IMS, in the case where the number of ion mobility spectrometry cells is equal to two.

[0059] As a further variant, the separation module 16 comprises three cells connected in cascade, namely an ion mobility spectrometry cell connected to the input of two tandem mass spectrometry cells. The spectrometry device 12 is then also called a coupled spectrometry device with IMS coupling with tandem MS / MS.

[0060] The invention then relates in particular to the following configurations of the spectrometry device 12: IMS alone; MS alone; tandem MS / MS; tandem IMS / IMS; IMS-MS coupling; IMS-tandem MS / MS coupling.

[0061] The mass spectrometry cell is for example a low-resolution analyzer, such as a quadrupole, a triple quadrupole, or a 3D ion trap (IT - from the English Ion Trop) or linear ion trap (LIT - from the English Linearlon Trop). Alternatively, the mass spectrometry cell is a high-resolution analyzer, allowing the exact mass of the analytes to be measured, such as a magnetic sector analyzer coupled to an electric sector, a time-of-flight (TOF - from the English Time Of Flight) analyzer, a Fourier-transform ion cyclotron resonance (FTICR - from the English Fourier-Transform Ion Cyclotron Resonance) analyzer and an Orbitrap.

[0062] The ion mobility spectrometry cell is according to any one of the types of ion mobility spectrometry cell presented below. A first type of ion mobility spectrometry cell is for example a drift time ion mobility spectrometry cell, also denoted DTIMS (from the English Drift Time Ion Mobility Spectrometer). A second type of ion mobility spectrometry cell is a traveling wave ion mobility spectrometry cell, also denoted TWIMS (from the English Traveling Wave Ion Mobility Spectrometer). A third type of ion mobility spectrometry cell is a high field asymmetric waveform ion mobility spectrometry cell, also denoted FAIMS (from the English high Field Asymmetric waveform Ion Mobility Spectrometer).A fourth type of ion mobility spectrometry cell is a trapped ion mobility spectrometry cell, also known as TIMS (Trapped Ion Mobility Spectrometer). A fifth type of ion mobility spectrometry cell is an open loop ion mobility spectrometry cell, also known as OLIMS (Open Loop Ion Mobility Spectrometer); also known as an aspiration ion mobility spectrometry cell and then known as AIMS (Aspiration Ion Mobility Spectrometer). A sixth type of ion mobility spectrometry cell is a differential analyzer. mobility, also known as DMA (Differential Mobility Analyzer). A seventh type of ion mobility spectrometry cell is a transverse modulation ion mobility spectrometry cell, also known as TMIMS (Transversal Modulation Ion Mobility Spectrometer). An eighth type of ion mobility spectrometry cell is a harmonic mobility spectrometry cell, also known as OMS (Overtone Mobility Spectrometer).

[0063] The detector 18 is capable of transforming the ions into an electrical signal. The more ions there are, the greater the current. In addition, the detector 18 is capable of amplifying the signal obtained, in particular so that it can be processed more easily by the electronic determination device 20.

[0064] The electronic estimation device 20 is configured to estimate the set of component(s) of the product via the spectrometry device 12.

[0065] The electronic estimation device 20 comprises a module 22 for obtaining a group of signatures for potential components Ci of the product, and a module 24 for acquiring a magnitude of an ionic flux of the product, measured via the spectrometry device 12.

[0066] As an optional addition, the estimation device 20 comprises a module 26 for factoring the group of signatures and the magnitude, with the same common factor.

[0067] The electronic estimation device 20 also comprises a module 28 for estimating the set of component(s) of the product from the group of signatures and the quantity and via an estimation algorithm. When, as an optional addition, the estimation device 20 comprises the factorization module 26, the estimation module 28 is then configured to perform the estimation of the set of component(s) of the product from the factorized quantity and the group of factorized signatures.

[0068] In the example of [Fig.l], the electronic estimation device 20 comprises an information processing unit 30 formed for example of a memory 32 and a processor 34 associated with the memory 32.

[0069] The product is for example a petroleum matrix. By petroleum matrix is ​​meant a composition comprising a mixture of hydrocarbons. Mention may in particular be made of crude oil, base oils (in particular those used in lubricating compositions), lubricating compositions, or even bitumens.

[0070] Each component likely to be contained in the product is for example chosen from the group comprising:

[0071] - a detergent, such as a calcium phenate or calcium sulfonate type detergent calcium;

[0072] - a polymeric dispersant, such as polyisobutylene succinimide;

[0073] - an antioxidant, such as a zinc dithiophosphate type antioxidant; and

[0074] - a molybdenum dithiocarbamate type compound.

[0075] In the example of [Fig.l], the obtaining module 22, the acquisition module 24 and the estimation module 28, as well as optionally the factorization module 26, are each produced in the form of software, or a software brick, executable by the processor 34. The memory 32 of the electronic estimation device 20 is then capable of storing software for obtaining the group of signatures for potential components Ci of the product, software for acquiring the magnitude of the ionic flux of the product, measured via the spectrometry device 12, and software for estimating the set of component(s) of the product from the group of signatures and the magnitude, and via the estimation algorithm.As an optional addition, the memory 32 of the electronic estimation device 20 is capable of storing software for factoring the group of signatures and the quantity with the same common factor, and the estimation software is then configured to estimate the set of component(s) of the product from the factored quantity and the group of factored signatures. The processor 34 is then capable of executing each of the software programs among the obtaining software, the acquisition software, and the estimation software, as well as, as an optional addition, the factorization software.

[0076] In a variant not shown, the obtaining module 22, the acquisition module 24 and the estimation module 28, as well as the optional addition the factorization module 26, are each produced in the form of a programmable logic component, such as an FPGA (Field Programmable Gate Away), or in the form of a dedicated integrated circuit, such as an ASIC (Application Specific Integrated Circuit).

[0077] When the electronic estimation device 20 is produced in the form of one or more software programs, that is to say in the form of a computer program, it is also capable of being recorded on a medium, not shown, readable by a computer. The computer-readable medium is for example a medium capable of storing electronic instructions and of being coupled to a bus of a computer system. By way of example, the readable medium is an optical disk, a magneto-optical disk, a ROM memory, a RAM memory, any type of non-volatile memory (for example EPROM, EEPROM, FLASH, NVRAM), a magnetic card or an optical card. A computer program comprising software instructions is then stored on the readable medium.

[0078] The obtaining module 22 is configured to obtain the group of signatures for potential components Ci of the product, each signature being representative of an ionic flux of a respective potential component Ci, each signature being measured via the spectrometry device 12.

[0079] The obtaining module 22 is for example configured to group the different signatures obtained in the form of a matrix S verifying the following equation:

[0080] [Math.l] S = [sJ ... |sP] e KNxP ; S] e [Rn, s2 g [RN, , sP e IRN

[0081] where S represents the signature matrix, i.e. the group of signatures obtained, and

[0082] If, ..., sP represent the different signatures obtained, and

[0083] P is the number of potential components Ci likely to be contained in the analyzed product, i.e. the number of components Ci in the considered group of components.

[0084] The acquisition module 24 is configured to acquire the magnitude of the ionic flux of the product, said magnitude being measured via the spectrometry device 12, and also called fingerprint.

[0085] The quantity acquired by the acquisition module 24 is for example noted x and verifies the following property:

[0086] [Math.2] xe [Rn

[0087] where x represents the magnitude of the ionic flux, or imprint, of the product to be analyzed.

[0088] Each signature obtained and the quantity acquired typically correspond to the same measurement modality of the spectrometry device 12, such as a modality chosen from the group consisting of: a mass spectrum, an ion mobility spectrum and a two-dimensional mapping.

[0089] Those skilled in the art will then understand that the measurement method used, i.e. the type of measurement used, for each signature obtained and the quantity acquired is typically a mass spectrum 40, visible in [Fig. 2], if the separation module 16 comprises a mass spectrometry cell; or an ion mobility spectrum 50 also visible in [Fig. 2], if the separation module 16 comprises an ion mobility spectrometry cell; or a two-dimensional map 60, also visible in [Fig. 2], if the separation module 16 comprises both a mass spectrometry cell and an ion mobility spectrometry cell, the two-dimensional map representing, as a function of their mass-to-charge ratio, noted m / z in [Fig. 2], an ion drift time, noted dt and expressed in milliseconds, during an analysis with the aforementioned separation module 16.

[0090] As an optional addition, the factorization module 26 is configured to factorize the group of signatures obtained and the acquired quantity with the same common factor D, each signature then being represented in the form of a product of the common factor D and a factorized signature, and the quantity then being represented as form of a product of said common factor D and a factored quantity.

[0091] According to this optional addition, the factored quantity and the group of factored signatures are then taken into account by the estimation module 28 to estimate the set of component(s) of the product, and the factorization carried out by the factorization module 26 then allows a reduction in dimension, which offers several advantages, in particular greater speed of calculation to carry out the estimation of the set of component(s) of the product, and lower consumption of computing resources, in particular of memory space.

[0092] The factorization module 26 is for example configured to factorize the group of signatures according to the following equation:

[0093] [Math.3] S=DS;D e [RNxH S e ^HxP; H < N

[0094] where S represents the matrix of the group of signatures;

[0095] D represents the common factor; and

[0096] S represents the matrix of the group of factored signatures.

[0097] Similarly, the factorization module 26 is for example configured to factorize the quantity according to the following equation:

[0098] [Math.4] x - Dx; From ÎRN *H, xe IRH; H < N

[0099] where x represents the magnitude;

[0100] D represents the common factor; and

[0101] î represents the factored quantity.

[0102] According to this optional addition, the common factor is for example obtained using a non-negative matrix factorization type algorithm, and the common factor, also called a dictionary, is for example a dictionary of parametric form defined by two-dimensional Gaussians.

[0103] The factorization module 26 is then, for example, configured to format each signature or quantity, in particular when the signatures and quantities are two-dimensional maps, such as the sampling of a distribution in the form of a set of two-dimensional Gaussian Gs, verifying, for example, the following equations:

[0104] [Math.5] y($) = E ^i)

[0105] [Math.6] ei

[0106] [Math.7]

[0107] The factorization module 26 is for example configured to represent each signature or quantity via a set of three parameters, namely an intensity, a mean and a variance-covariance matrix.

[0108] The factorization module 26 is then typically configured to represent the two-dimensional Gaussian Gs, in the form:

[0109] of an intensity vector verifying the following equation:

[0110] [Math.8] II = [7Tb ..., nG]r G (üV)G

[0111] of a mean vector verifying the following equation:

[0112] [Math.9] -JL a T 1 tTTl 2 XG M = ... , gj G [R

[0113] and / or a set of variance-covariance matrices according to the following equation:

[0114] [Math. 10] E=[zb ... ,Eg] G

[0115] For the factorization of the group of signatures and the magnitude, the factorization module 26 is then for example configured to define a set of Gaussians G dict whose means and variance-covariances are respectively noted Mdict and S dict. These values ​​of means and variance-covariances then define the common factor D and are adjusted via an expectation-maximization algorithm on the group of signatures.

[0116] The factorization module 26 is then configured to estimate the relative intensities of the Gaussians of the common factor D for each signature and quantity, typically for each two-dimensional mapping, and a vector of intensities denoted irSpeC is then obtained for each signature and quantity, typically for each two-dimensional mapping, by executing the expectation-maximization algorithm initialized at the positions Mdict, Sdict without subsequently updating them. The irspec representation is then used as the factorized quantity and respective factorized signature for the estimation of the set of component(s) of the product.

[0117] The estimation module 28 is configured to estimate the set of component(s) of the product from the group of signatures obtained by the obtaining module 22 and the quantity acquired by the acquisition module 22, said estimation being carried out via the estimation algorithm. In other words, the estimation module 28 is configured to determining the signature(s) which, among the group of signatures, are present in the acquired quantity of the ionic flux of the product, in order to estimate the component(s) which, among the group of components corresponding to the group of signatures, are present in the analyzed product, these signatures and quantities being measured via the spectrometry device 12.

[0118] The product is considered to be an additive linear mixture of the components where each component is weighted by a mixing coefficient ai5 each mixing coefficient a; having a positive value.

[0119] The estimation module 28 is then configured to model the magnitude of the product, i.e. the footprint of the product, in the form of a sum of the signatures of the group, where each signature is weighted by the respective mixing coefficient a;, then to estimate the value of each mixing coefficient a; via the estimation algorithm.

[0120] The estimation module 28 is for example configured to model the magnitude of the ionic flux of the analyzed product, according to the following equation:

[0121] [Math. 11] pri + X — / , . _ i&pSj J 3 — , ... * Ûpj G JRI

[0122] where x represents the size, i.e. the footprint of the product;

[0123] a; represents the mixing coefficient for the component of index i, i being an integer index associated with the component and of value between 1 and P, P being the number of components in the group of components, corresponding to the group of signatures,

[0124] a represents a vector of the P mixing coefficients for the group of components,

[0125] S; represents the signature of the component of index i.

[0126] When, as an optional complement, the factorization module 26 is configured to factorize the group of signatures obtained and the acquired quantity with the same common factor D, the factored quantity and the group of factored signatures are then, as indicated previously, taken into account by the estimation module 28 to estimate the set of component(s) of the product. Those skilled in the art will then understand that the preceding equation (11), as well as all the equations below involving the quantity x and / or one or more signatures s; are then, according to this optional complement, typically implemented by replacing the quantity x by the factored quantity î and / or each signature s; by the corresponding factored signature.

[0127] According to a first embodiment of the estimation algorithm, the estimation module 28 is configured to estimate the mixing coefficients, i.e. the vector a of the mixing coefficients of the components in the product, by minimizing a criterion dependent on the magnitude and the group of signatures.

[0128] The criterion to be minimized for the estimation of the mixing coefficients a; is for example a quadratic criterion, such as a least squares criterion.

[0129]

[0130]

[0131]

[0132]

[0133]

[0134]

[0135]

[0136]

[0137]

[0138]

[0139]

[0140]

[0141]

[0142] The estimation module 28 is for example configured to estimate the vector a of the mixing coefficients by minimizing the quadratic criterion verifying the following equation: [Math. 12] a = argmin ( || x - Sa || under the constraint a > 0 a ^7 The estimation module 28 is then configured to solve the previous equation 12 and estimate the mixing coefficients a; of the vector a for example using a so-called interior point method, such as that described in the document “Solving Least Squares Problems” - Lawson et al, 1995. Alternatively, the criterion to be minimized for the estimation of the mixing coefficients a; is a logarithmic criterion. According to this variant, the estimation module 28 is for example configured to model the quantity x via a Poisson distribution. For example, when the quantity x is a mass spectrum, the number of ionic impacts measured for the ith value of the mass / charge ratio m / z is noted xi5 and this number of ionic impacts x; is considered as a sample of the parameter Poisson distribution, verifying for example the following equation: [Math. 13] VP / a, SA = E . ,a;.S.. \ j / j = t ii where Sj represents the jth row of the signature matrix S, a represents the vector of P mixing coefficients a;. A likelihood function associated with this Poisson distribution is then expressed by example as follows: [Math. 14] According to this variant, the estimation module 28 is then, for example, configured to minimize the logarithmic criterion verifying the following equation: [Math. 15] L(x|a) = E f=1Xjlog (a, SJ -(a, SJ The estimation module 28 is then, for example, configured to resolve this minimization of the logarithmic criterion, via an Expectation-Maximization algorithm, also noted EM, such as that described in the document “A Modified Expectation Maximization Algorithm for Penalized Likelihood Estimation in Emission Tomography” - De Pierro, 1995. Following the estimation of the mixing coefficients ai5 the estimation module 28 is configured to determine the presence or absence of each potential component in the product by comparing each mixing coefficient a; with a respective detection threshold 0i.

[0143] In other words, the estimation module 28 is for example configured to implement a binary classifier h, with respect to the threshold 0; and on the mixing coefficient a i according to the following equation:

[0144] [Math. 16] h„ (a, ) = 1 if as > 0, 0 otherwise

[0145] where a; represents a respective mixing coefficient,

[0146] 0; represents a respective detection threshold for the component of index i, i being between 1 and P.

[0147] Each detection threshold 0; associated with a respective component is preferably predefined. Each detection threshold 0; is typically determined during a preliminary training phase from a set of K products, each consisting of various mixtures of components, where the components are chosen from the predefined group of components. This set of K products is also called a training set, and each product of this training set consists of a mixture, real or simulated, of said components.

[0148] The detection thresholds 0; for the components of the predefined group are then for example determined using a ROC curve (from the English Receiver Operating Characteristic), such as curve 70 visible in [Fig.4].

[0149] A ROC curve is typically determined for each component. Each ROC curve, such as curve 70, comprises a plurality of points 72, where each point 72 corresponds to a respective possible value 0 of the detection threshold 0; to be determined for the component of index i.

[0150] Each point 72 then has as its abscissa an FPR value (from the English False Positive Rate) corresponding to a false positive rate for this value 0, and verifying for example the following equation:

[0151] [Math. 17] FPR / ^ _ fp(Ma)) - Total Negative

[0152] where 0 represents the value considered as the detection threshold,

[0153] hu(ai) represents the binary classifier associated with this value 0, such as the classifier defined in equation (16) above,

[0154] FP(hu(ai)) represents an estimated number of false positives for this classifier hu(ai) and this value 0, and

[0155] Total Negative represents the number of mixtures among the K mixtures training where the component with index i is actually absent.

[0156] Each point 72 has as its ordinate a TPR value (from the English True Positive Rate) corresponding to a true-positive rate for this value 0, and verifying for example the following equation:

[0157] [Math. 18] - Total Positive

[0158] where 0 represents the value considered as the detection threshold,

[0159] hu(ai) represents the binary classifier associated with this value 0, such as the classifier defined in equation (16) above,

[0160] VP(hu(ai)) represents an estimated number of true-positive(s) for this classifier hu(ai) and this value 0, and

[0161] Total Positive represents the number of mixtures among the K training mixtures where the component of index i is actually present.

[0162] In the present description, and as known per se, a true-positive corresponds to an estimate of the presence of an element which is actually present, a false-positive corresponds to an estimate of the presence of an element which is actually absent, a false-negative corresponds to an estimate of the absence of an element which is actually present, and a true-negative corresponds to an estimate of the absence of an element which is actually absent, as summarized in table (1) below.

[0163] [Tables 1] Reality Present Absent Estimation Present True Positive (TP) False Positive (FP) Absent False Negative (FN) True Negative (VN)

[0164] The optimal value among the values ​​0 considered for each component of index i, that is to say the optimal detection threshold 0;, then corresponds to the value 0 for which the point 72 obtained is closest to an ideal point A, visible in [Fig.2], and corresponding to a TPR value equal to 1 for a zero FPR value.

[0165] In the example of [Fig.4], the optimal value among the 0 values ​​considered corresponds to point 72 represented by a cross and noted OPT, the other points 72 each being represented by a circle.

[0166] The optimal detection threshold 0; for the component of index i then typically verifies the following equation:

[0167] [Math. 19] O^argmin d((TPR(0), FPR(0) ), (1,0)) o

[0168] where O; represents the optimal detection threshold for the component of index i,

[0169] FPR(0), TPR(0) are defined by the previous equations (17) and (18) respectively,

[0170] d(.,.) represents the usual Euclidean distance.

[0171] The determination of each detection threshold 0; is for example carried out using real mixtures of components, that is to say real mixtures of signatures of said components. Each real mixture then typically corresponds to the quantity, also called fingerprint, acquired by the acquisition module 24.

[0172] Alternatively, the determination of each detection threshold 0; is carried out from simulated mixtures. According to this variant, the simulation of the mixtures is carried out using the signatures of the components and a physical mixture model, verifying for example the following equation:

[0173] [Math.l9bis] xs = f ( E aj.Sj)

[0174] where xs represents the magnitude, i.e. the footprint, of the simulated mixture,

[0175] a; represents the mixing coefficient for the component of index i,

[0176] f(. ) is a function representing a joint effect of measurement noise and physical phenomena inherent to the measurement, such as nonlinearity, interaction and peak shift.

[0177] The estimation module 28 is configured to determine the presence or absence of each component in the analyzed product, by comparing the mixing coefficient a; estimated for this component of index i with the detection threshold 0; predefined for said component, that is to say by implementing a binary classifier, such as that defined in the preceding equation (16), then to provide the result of said estimation of the presence or absence of each component in the product, in the form of a binary list [3est, typically defined according to the following equation:

[0178] [Math.20] h» («i) P = ... ; e {0,lf ' is ' is 1

[0179] Those skilled in the art will then understand that for each component of index i, it is determined as being present in the analyzed product if the value hoi(ai) of the binary list pest of equation (20) is equal to 1, and conversely as being absent if said value hoi(ai) is equal to 0.

[0180] An example of such a binary list thus obtained is represented in [Fig.5] in the form of a histogram labeled “Est” where, for each component Ci with the index i between 1 and 20, a gray box corresponds to an estimate of the presence of said component in the product and a white box corresponds to the estimate of an absence of said component in the product.

[0181] Figure 5 also shows a histogram labeled “Ref” showing the components actually present in the analyzed product, and those skilled in the art will then observe that, in this example, the binary list Pest estimated by the estimation device 20 according to the invention is correct for all the components C1 to C20, with the exception of the component CIO for which its presence was estimated in the analyzed product, whereas it is not in reality present in this product.

[0182] As an optional addition, the obtaining module 22 is configured to obtain the group of signatures for potential components of the product for a plurality of distinct measurement modalities, each measurement modality typically being chosen from a mass spectrum, an ion mobility spectrum and a two-dimensional mapping.

[0183] According to this optional addition, the acquisition module 24 is configured to acquire the magnitude of the product analyzed for the plurality of distinct measurement modalities, i.e. for the same measurement modalities as those used for the group of signatures obtained by the obtaining module 22.

[0184] Those skilled in the art will then observe that the plurality of measurement modalities used is likely to contain several modalities of the same nature, for example several mass spectra and / or several ion mobility spectra and / or several two-dimensional maps. By way of example, the different mass spectra are an ASAP+ mass spectrum, and respectively an ASAP- mass spectrum, with the ionization source 14 of type ASAP+, and respectively ASAP-. Similarly, the different ion mobility spectra are an ASAP+ ion mobility spectrum, and respectively an ASAP- ion mobility spectrum, with the ionization source of type ASAP+, and respectively ASAP-.

[0185] According to this optional addition, the estimation module 28 is then configured to estimate the set of component(s) of the analyzed product, as a function of the signature groups and the quantities for said plurality of measurement modalities.

[0186] According to this optional addition, the estimation module is then configured to represent the product ion flux quantities acquired for the different measurement methods according to the following equation:

[0187] [Math.21] 4 A". le[l,t]

[0188] where 1 is an integer index designating a measurement modality, 1 having a value between 1 and L, L defining the number of distinct measurement modalities,

[0189] x® represents the quantity for the measurement modality of index 1,

[0190] a;® represents the mixing coefficient associated with the component of index i for the index measurement modality 1,

[0191] S;(1) represents the signature obtained for the component of index i and the modality of index measure 1.

[0192] The estimation module 28 is then, for example, configured to estimate each mixing coefficient a;(1) independently from one measurement modality to another. The estimation module 28 is then typically configured to estimate the mixing coefficients a;(1) separately for each of the measurement modalities of index 1 and in a manner identical to what was previously described in the case of a single measurement modality.

[0193] The estimation module 28 is for example configured to estimate the mixing coefficients a;(1) for the measurement modality of index 1 by minimizing a quadratic criterion, such as the quadratic criterion verifying the following equation:

[0194] [Math.22] a^^argmin ( || x!1)-S(1).a || ), under the constraint a > 0, the 1, Lj has 2 / LJ

[0195] where a(1) represents the vector of mixing coefficients a;(1) for the measurement modality of index 1,

[0196] x(1) represents the quantity acquired for the measurement modality of index 1, and

[0197] S(1) represents the matrix of signatures obtained for measurement modality 1.

[0198] To determine the presence or absence of each potential component in the product, a binary classifier is then, for example, defined for each component of index I and each measurement modality of index 1, for example according to the following equation:

[0199] [Math.23] W 1 • (H 1 if a- > Qj 0 otherwise

[0200] where hui(1) represents the binary classifier for the component of index i and the modality of index measure 1,

[0201] a;(1) represents the mixing coefficient for the component of index i and the modality of index measurement 1, and

[0202] 0i(1) represents the detection threshold for the component of index i and the modality of index measure 1.

[0203] Each detection threshold 0;(1) is predefined, and for example obtained separately for each of the measurement modalities of index 1 and in an identical manner to what was described previously for the detection threshold 0;in the case of a single measurement modality, typically using a ROC curve for each detection threshold 0;(1).

[0204] According to this optional addition in the case of this plurality of measurement modalities, in order to evaluate the relative importance of each measurement modality in determining mination of the presence or absence of each potential component in the product, the estimation module 28 is for example configured to calculate a weighting score for each binary classifier, for example according to the following equation:

[0205] [Math.24] ,1) n Total Positive + Total Negative

[0206] where VP represents the number of true-positives for the classifier h;(1) associated with the component of index i and the measurement modality 1,

[0207] VN represents the number of true-negative for said classifier h;(1),

[0208] Total Positive represents the number of mixtures among the K training mixtures for the measurement modality of index 1, where the component of index i is actually present, and

[0209] Total Negative represents the number of mixtures among the K training mixtures for the measurement modality of index 1, where the component of index i is actually absent.

[0210] The estimation module 28 is then, for example, configured to use, for each component of index i, a multi-classifier system, corresponding to a weighted sum of the binary classifiers h;®, typically according to the following equation:

[0211] [Math.25]

[0212] where A; represents the multi-classifier system for the component of index i,

[0213] Yi(1) represents a weighting coefficient for the component of index i and the measurement modality 1, defined for example in table (2) below in combination where appropriate with equation (24) above, and

[0214] hi(1) represents the binary classifier for the component of index i and the modality of measure of index 1, verifying for example the previous equation (23).

[0215] Table (2) below presents, in the “Equation” column, different examples of weighting coefficient y;®, in particular as a function of the variable p®' defined in equation (24), and this for different types of weighting corresponding to the denominations indicated in the “Denomination” column.

[0216] [Tables2] Denomination Equation Majority Vote (MV) v <n_ 1 Vi “L Vote Pondéré (PV) UW Y® = --- k = ri Vote Pondéré du Meilleur au Pire (BWWV) vm_, - P® V; - 1 " U - U . 1 1 max ‘ mm Vote Pondéré du Meilleur au Pire Quadratique (QBWWV) _ 1 *max y Yi “ 1 -1 p - p J 1 \rmax r mm Vote Pondéré Rééchelonné (RSWV) (b | 1 si uP = max Yi = 1 k e [I, L] “ 10 sinon Vote Majoritaire Pondéré (WMV) u0) Yf=l°g 1 (1) 1 i - a- Sélection Dynamique de Classifieur (DCS) <n P-( 1 ’ M®) 1 Y® = maxpX ^.(p 1)’} avec par exemple p = 2

[0217] According to this optional addition, in the case of the plurality of measurement modalities, the estimation module 28 is then configured to determine the presence or absence of each potential component in the analyzed product, in the form of a merged binary list obtained from the multi-classifier systems for the different components, the merged binary list verifying for example the following equation:

[0218] [Math.26]

[0219] where A; represents the multi-classifier system for the component of index i,

[0220] f represents a comparison function, such as the comparison function defined according to the following equation:

[0221] [Math.27] 1, if A i S: 4 0, otherwise

[0222] Those skilled in the art will then understand that for each component of index i, the latter is determined as being present in the product if the value f(A;) of the aforementioned merged binary list is equal to 1, and conversely absent if said value is equal to 0.

[0223] Examples of binary lists obtained for several measurement modalities and according to the first embodiment of the estimation algorithm are represented in [Fig.6] in the form of several histograms. Three first histograms labeled “MS AS AP-”, “IMS ASAP+” and “IMS AS AP-” show the binary lists obtained for respectively the modalities MS AS AP- corresponding to a mass spectrum for the ionization source 14 of type ASAP-, IMS ASAP+ corresponding to an ion mobility spectrum for the ionization source 14 of type ASAP+, and IMS ASAP- corresponding to an ion mobility spectrum for the ionization source 14 of type ASAP-. A fourth histogram labeled “Fusion” shows a merged binary list estimated by the estimation module 28, for example using the previous equations (26) and (27), a fifth histogram labeled “Ref” showing the components actually present in the analyzed product.

[0224] In [Fig.6], for each histogram represented, a gray box corresponding to a presence of said component in the product, this being estimated in the case of the first four histograms “MS ASAP-”, “IMS ASAP+”, “IMS ASAP-”, “Fusion” and real in the case of the last histogram “Ref”, and a white box corresponding to an absence of said component in the product, this absence also being estimated in the case of the first four histograms “MS ASAP-”, “IMS ASAP+”, “IMS ASAP-”, “Fusion” and real in the case of the last histogram “Ref”.

[0225] Those skilled in the art will then note that, in this example, the estimation carried out by the estimation device 20 according to the invention in the case of the fusion of the plurality of measurement modalities is perfect, the merged binary list |3fus as provided by the estimation module 28, being exact for all the components C1 to C20, the histograms “Fusion” on the one hand and “Ref” on the other hand being identical for each of the components C1 to C20, as visible in [Fig.6].

[0226] Those skilled in the art will further note that the estimations made by the estimation device 20 for each of the measurement modalities taken separately are less effective, the respective binary lists provided by the estimation module 28 being appear for each measurement modality one or more estimation errors for some of the components. For example, for the measurement modality MS ASAP- represented on the first histogram “MS ASAP-”, the determination of the presence or absence of a component in the product is erroneous for the component C5 which is estimated to be absent while it is in reality present, and also for the component C14 estimated to be present while it is in reality absent. For the measurement modality IMS ASAP+ corresponding to the second histogram “IMS ASAP+” of [Fig.6], the determination of the presence or absence of a respective component in the product is erroneous for the component C6 estimated to be present while it is in reality absent from the product, and in the same way for the components CIO, C13, C14, C17 and C20 all estimated to be present while they are in reality absent from said analyzed product, the component Cl8 being estimated to be absent while it is in reality present in the analyzed product.For the IMS ASAP- measurement modality, corresponding to the third histogram “IMS ASAP-” in [Fig.6], the determination of the presence or absence of each respective component in the product is erroneous for the components C8, CIO, C13, C15 and C20 all estimated to be present while they are in reality absent from the analyzed product, as well as for the component C18 estimated to be absent while it is in reality present in the analyzed product.

[0227] Those skilled in the art will then understand that the merging of several measurement methods makes it possible to significantly improve the precision of the estimation of the set of component(s) of the product analyzed, in particular the determination of the presence or absence of each respective component in the product analyzed.

[0228] A second embodiment of the estimation algorithm will now be described, it being noted that the differences for this second embodiment compared to what was previously described for the first embodiment, concern only the estimation algorithm implemented by the estimation module 28. Those skilled in the art will understand in particular that the obtaining module 22, the acquisition module 24, as well as, as an optional addition, the factorization module 26, are then identical from one embodiment to another.

[0229] According to the second embodiment, the estimation algorithm comprises an implementation of a sparse decomposition algorithm based on the quantity and the group of signatures.

[0230] The sparse decomposition algorithm is typically chosen from the group of greedy algorithms consisting of: MP algorithm (from the English Matching Pursuit), OMP algorithm (from the English Orthogonal Matching Pursuit), and OLS algorithm (from the English Orthogonal Least Squares).

[0231] According to the second embodiment of the estimation algorithm, the estimation module 28 is typically configured to define an initial residue r0 equal to the quantity, measured via the spectrometry device 12 and acquired by the module acquisition 24, then to calculate each subsequent residue rk as being equal to the difference between the quantity and a product of signature of component(s) already estimated present and respective mixing coefficient(s), and this until a convergence criterion is reached. By residue, we then also mean the remainder of a difference between two terms, in particular the remainder of the difference between the quantity and a linear combination of signature(s) of component(s) already estimated present and respective mixing coefficient(s). The convergence criterion is typically the achievement of a minimum value e for a given subsequent residue rk, and the list of component(s) estimated present via the estimation algorithm according to the second embodiment is then that associated with the subsequent residue rk when the convergence criterion of the algorithm is reached.

[0232] The sparse decomposition algorithm is for example the greedy algorithm of the OMP type. When initializing the OMP algorithm, the initial residue r0 is equal to the quantity x, an initial set of present esthné component(s) Io is empty and an initial set of potential component(s) Lo is equal to the interval of integer values ​​between 1 and P, this interval corresponding to the set of possible values ​​for an index j designating a respective component of the group of components considered, P being the number of components contained in said group of components.

[0233] The initial residue r0, the initial set of present esthné component(s) Io and the initial set of potential component(s) Lo then typically verify the following equations:

[0234] [Math.28] r0 = x ; xe R1N

[0235] [Math.29] Io={0}

[0236] where 0 represents the empty set,

[0237] [Math.30] k = [tP]

[0238] In the remainder of the description of this example of an OMP algorithm, the component index is denoted j to avoid any confusion with the component index i used in the equations relating to the estimation algorithm according to the first embodiment.

[0239] At each iteration of index k and until convergence of the sparse decomposition algorithm, the estimation module 28 is configured to calculate a correlation score |3(Sj,rk) between each potential component signature, i.e. each of the signatures Sj where j belongs to Lk and the current residue at this iteration k, noted rk.

[0240] The correlation score |3(Sj,rk) calculated by the estimation module 28 is for example a canonical scalar product, and typically verifies the following equation:

[0241] [Math.31] P(srrk)= (srrk)

[0242] Those skilled in the art will observe that the correlation score in the form of a canonical scalar product is particularly suitable when each signature is in the form of a Gaussian distribution.

[0243] After having calculated the correlation score [3(Sj,rk) for each index j belonging to the set Lk, the estimation module 28 is configured to add, to the set Ik_i of the estimated components present at the previous iteration of index k-1, the index jk corresponding to the maximum correlation score among those calculated at the iteration of index k, and then obtain the set Ik of the estimated components present at the current iteration of index k, and by complementarity to subtract, i.e. remove, from the current set Lk of potential component(s), the index of component jk added to the estimated current set Ik, to then obtain the potential subsequent set Lk+[.

[0244] The sparse decomposition algorithm implemented by the estimation module 28 at the iteration of index k then typically verifies the following equations:

[0245] [Math.32] j, = max s., rkA j eLk U j M

[0246] [Math.33] Ik=Ik-i u (jk)

[0247] [Math.34] Lk+1 =Lk\ {jk]

[0248] where Ik_i and Ik represent the set of estimated components present at the previous iteration of index k-1, and respectively at the current iteration of index k;

[0249] Lket Lk+[ represent the set of potential components at the current iteration of index k, and respectively at the following iteration of index k+1, also called subsequent iteration, and

[0250] jk represents the index of the newly estimated component present at the current iteration of index k, and then added to the set Ik.i to form the new set Ik, while being subtracted from the set Lk to form the new set Lk+[.

[0251] Following this estimation of the newly present component of index jk, the estimation module is configured to update parameters for the next iteration of index k+1, in particular a matrix Fk of the estimated components present, mixing coefficients via the updating of a parameter Ok, as well as the residue rk at the current iteration of index k, for example according to the following equations:

[0252]

[0253]

[0254]

[0255]

[0256]

[0257]

[0258]

[0259]

[0260]

[0261]

[0262]

[0263]

[0264]

[0265]

[0266] [Math.35] where Fk represents the matrix of estimated components present, [Math.36] 0k-min || x-Fk. <j>|| [Math.37] 1 k - A " rk.^k where rk represents the residue at the iteration of index k, x represents the magnitude, Fk represents the updated matrix of estimated components present, <pk représente les coefficients de mélange mis à jour. The estimation module 28 is then configured to calculate a convergence variable between the residue rk at the current iteration k and the residue rk.i at the previous iteration k-1, this convergence variable being noted ck(rk.b rk), and verifying for example the following equation: [Math.38] ck(rk-i, rk) = where ck(rk.b rk) represents the convergence variable between the current residue rk at the end of the current iteration with index k and the previous residue rk_i at the end of the previous iteration with index k-1. The estimation module 28 is then configured to interrupt the sparse decomposition algorithm if the current residue rk is zero, or if the iteration index k is equal to P, meaning that the entire set of potential components has been covered by the algorithm, or if the convergence variable ck is considered sufficiently small, and for example less than the predefined minimum value e, according to the following equation: [Math.39] ck(rk-3, rk) e If at least one of the three aforementioned conditions is verified, leading to the stopping of the parsimonious decomposition algorithm, the estimation module 28 is then configured to determine the list of components present in the analyzed product, as being equal to the list of components associated with the indices contained in the current set Ik of estimated components present. Otherwise, that is to say if none of the three aforementioned conditions is verified, the estimation module 28 is configured to increment by one unit the iteration index k, which then becomes equal to k+1, and to reiterate the calculation of the correlation scores, the determination mination of the maximum correlation score for adding to the set Ik the index jk+i to form the new estimated set Ik+b while removing the index jk from the current set Lk to obtain the next set Lk+i of potential components, then updating the parameters, as well as the convergence variable ck+i.

[0267] The estimation module 28 is then typically configured to implement the sparse decomposition algorithm at the iteration of index k+1 via the preceding equations (31) to (39) and by iterating the index k by one unit, this then being equal to k+1.

[0268] At the end of this new iteration, the estimation module 28 is configured to check again whether at least one of the three aforementioned criteria is met, that is to say whether the residue rk+[ is zero, or whether the index k+1 is equal to P, or whether the convergence variable ck+i(rk,rk+i) is less than or equal to the minimum value e and to stop the parsimonious decomposition algorithm if necessary; or otherwise to reiterate the algorithm by incrementing the iteration index again by one unit, this then becoming equal to k+2, and so on.

[0269] Following the implementation of the sparse decomposition algorithm by the estimation module 28, the set I of indices of the estimated components present by this algorithm then typically verifies the following equation:

[0270] [Math.40] I gf / max = (jg [ 1, P], aj >0 j

[0271] The binary list of detected components then typically verifies the following equation:

[0272] [Math.41] P, r P = ... with p. = H s* i G ? P l 0 smon

[0273] where I is the set as defined in equation (40) above.

[0274] As an optional addition to this second embodiment, the estimation module 28 is configured to estimate the set of components of the product from a plurality of distinct measurement modalities, when several groups of signatures have been obtained by the obtaining module 22 for said plurality of measurement modalities and several quantities have been acquired by the acquisition module 24 for this plurality of measurement modalities.

[0275] According to this optional addition to the second embodiment of the estimation algorithm, the estimation module 28 is further configured to calculate at each iteration of index k a joint correlation score between the different modalities of said plurality, for example according to the following equation:

[0276] [Math.42]

[0277] where pco represents the joint correlation score,

[0278] Sj represents the signature for the component of index j belonging to the set Lk,

[0279] rk represents the residue at iteration k,

[0280] am represents a mixing coefficient associated with said component of index j and for the measurement modality of index m, where m is an integer varying between 1 and L, L representing the number of distinct measurement modalities, and

[0281] ^(Sjm, rkm) represents a unit correlation score for each measurement modality of index m,

[0282] such that the canonical scalar product according to the following equation:

[0283] [Math.43] ., / t,m «m \ ... / £,m «m \ M ( sj ? rk ) - ( , rk

[0284] where Sjm represents the signature for the component of index j and the measurement modality of index m,

[0285] rkm represents the residue at the iteration of index k for the measurement modality of index m.

[0286] The estimation module 28 is then configured to update parameters similar to those described previously in the example of the single measurement modality, and this for each measurement modality of index m, for example according to the following equations:

[0287] [Math.44] min || - Fkm).$ || , me [1, l] <l>e œ

[0288] [Math.45] me[l,L]

[0289] The analysis of the convergence of the sparse decomposition algorithm is then carried out jointly for all the measurement modalities, and the estimation module 28 is configured to calculate a joint convergence variable for the different measurement modalities, for example according to the following equation:

[0290] [Math.46] .. a L y L II rk ->k+i II k || ||

[0291] where ck represents the joint variable of convergence at the iteration of index k,

[0292] L represents the number of measurement modalities,

[0293] rk(m) represents the residue at the iteration of index k for the modality of index m, and

[0294] rk+i(m) represents the residue at the iteration of index k+1 for said modality of index m.

[0295] In a manner analogous to what was previously described in the case of a single measurement modality for the second embodiment of the estimation algorithm, the estimation module 28 is then configured to check, at the end of the iteration of index k, whether the residue rk(m) is zero for each of the measurement modalities of index m; and / or whether the iteration index k is equal to P; and / or whether the convergence criterion is verified, and whether the convergence variable ck is for example less than the predefined minimum value e, according to the following equation:

[0296] [Math.47] Ck < E

[0297] The sparse decomposition algorithm has just been described in detail above in the example of the greedy algorithm of the OMP type, and the person skilled in the art will then be able to deduce the implementation of a sparse decomposition algorithm of the MP algorithm type or even the OLS algorithm, from the example above for the OMP algorithm.

[0298] In particular, those skilled in the art will then observe that each iteration of index k of the sparse decomposition algorithm, such as a greedy algorithm, in particular the OMP algorithm, the MP algorithm, or the OLS algorithm, comprises three successive phases, namely a first phase E1 corresponding to the selection of a new component to be added from among the estimated components, a second phase E2 consisting of calculating the contribution of this added component in the mixture, and a third phase E3 consisting of verifying whether or not the convergence criterion, i.e. a stopping criterion for the algorithm, is satisfied.

[0299] When the greedy algorithm is the MP algorithm, the first phase E1 comprises, for example, the calculation of a scalar product Vj between the residue rk at iteration k, and the signatures Sj for each component index j belonging to the set Lk of potential components at iteration k. The second phase E2 comprises the selection of the signature Sj most correlated to the residue rk; and the third phase E3 comprises the subtraction of the projection, on the current residue rk, of the most correlated signature

[0300] When the greedy algorithm is the OMP algorithm described previously, those skilled in the art will observe that the first and second phases E1, E2 are similar to those of the MP algorithm, and that the third phase E3 comprises the calculation of an orthogonal projection of the residue rk onto a subspace generated by the previously detected components and maintaining the one selected in the second phase E2.

[0301] When the greedy algorithm is the OLS algorithm, this algorithm essentially comprises two phases, namely the first phase E1 comprising the calculation of the orthogonal projection of the residue rk at iteration k on subspaces including the signatures of the components already selected and the candidate signatures, with a single signature each time; and the second phase E2 comprising the conservation of the signature Sj for the component of index j for which the residue is minimal after projection.

[0302] The operation of the electronic estimation device 20 according to the invention will now be described with reference to [Fig. 3] representing a flowchart of the method for estimating the set of component(s) of the product to be analyzed, said estimation method being implemented by the electronic estimation device 20.

[0303] During an initial step 100, the estimation device 20 obtains, via its obtaining module 22, the group of signatures for potential components of the product, where each signature is representative of an ionic flux of a respective potential component, and measured via the spectrometry device 12.

[0304] The estimation method then moves on to the next acquisition step 110 during which the estimation device 20 acquires, via its acquisition module 24, the magnitude of the ionic flux of the product to be analyzed, this magnitude also being measured via the spectrometry device 12.

[0305] At the end of the obtaining 100 and acquisition 110 steps, the method moves on to an optional step 120 of factoring the group of signatures and the quantity, or else moves directly to an estimation step 130.

[0306] During the factorization step 120, the estimation device 20 factors, via its factorization module 26, the group of signatures obtained during the obtaining step 100 via the obtaining module 22, and the ionic flux quantity of the product acquired during the acquisition step 110 via the acquisition module 24, with the same common factor, such as the common factor D.

[0307] The factorization of the group of signatures is for example carried out according to the preceding equation (3) and that of the quantity for example according to the preceding equation (4), the common factor D being for example determined in the manner described previously with regard to the preceding equations (5) to (10).

[0308] Those skilled in the art will observe that, as an optional addition, the obtaining 100 and acquisition 110 steps are respectively implemented for several distinct measurement modalities, as well as the optional factorization 120 step.

[0309] At the end of the obtaining 100 and acquisition 110 steps, or even the optional factorization step 120, the estimation method moves on to the estimation step 130 during which it estimates, via its estimation module 28 and the associated estimation algorithm, the set of component(s) of the analyzed product, this from the group of signatures and the quantity.

[0310] This estimation of the set of component(s) of the product is for example carried out with the estimation algorithm according to the first embodiment, that is to say the estimation algorithm based on the minimization of a criterion dependent on the magnitude and the group of signatures.

[0311] Alternatively, the estimation step 130 is performed via the estimation algorithm according to the second embodiment, that is to say the estimation algorithm based on a sparse decomposition algorithm, such as a greedy algorithm, in particular the OMP algorithm, the MP algorithm, or the OLS algorithm.

[0312] When, as an optional addition, the obtaining 100 and acquisition 110 steps have been implemented for several distinct measurement modalities, the estimated set of component(s) of the product then depends, during the estimation step 130, on the groups of signatures and the quantities for the different measurement modalities.

[0313] The estimation algorithm implemented is then the estimation algorithm according to the first embodiment for the variant with several measurement modalities, or the estimation algorithm according to the second embodiment again for the variant with several measurement modalities.

[0314] Thus, the estimation device 20 according to the invention makes it possible to provide satisfactory estimations of the set of component(s) contained in the product to be analyzed, both with the estimation algorithm according to the first embodiment and with the estimation algorithm according to the second embodiment, as will now be explained with reference to FIGS. 7 and 8.

[0315] [Fig.7] is a histogram diagram showing, for each component C2 to C20, four estimation rate histograms, namely a true-positive rate TVP histogram, also noted TPR (from the English True Positive Rate); a false-positive rate TFP histogram, also noted FPR (from the English False Positive Rate); a true-negative rate TVN histogram, also noted TNR (from the English True Negative Rate); and respectively a false-negative rate TFN histogram, also noted FNR (from the English False Negative Rate); these histograms having been obtained with the estimation algorithm according to the first embodiment.

[0316] Similarly, [Fig.8] is a histogram diagram showing the same histograms for the same components C2 to C20, but obtained here with the estimation algorithm according to the second embodiment.

[0317] The true-positive TVP, false-positive TFP, true-negative TVN and false-negative TFN rates are known per se, and verify, for example, the following equations:

[0318] [Math.48] ry p _ VP___ 1 vr ~ Total Positive

[0319] where TVP represents the true-positive rate, also noted TPR,

[0320] VP represents the number of true positives,

[0321] Total Positive represents the number of products analyzed where the corresponding component is actually present;

[0322] [Math.49]

[0323] where TFP represents the false positive rate, also noted FPR,

[0324] FP represents the number of false positives,

[0325] Total Negative represents the number of products analyzed where the component cor

[0326] respondent is actually absent; [Math.50] tvv _ ____VIN___ 1 V iN - Total Negative

[0327] where TVN represents the true-negative rate, also noted TNR,

[0328] VN represents the number of true-negative,

[0329] Total Negative represents the number of products analyzed where the corresponding component is actually absent;

[0330] [Math.51]

[0331] where TFN represents the false negative rate, also noted FNR,

[0332] FN represents the number of false negatives, and

[0333] Total Positive represents the number of products analyzed where the corresponding component is actually present.

[0334] Those skilled in the art will then understand that the true-positive rate TVP represents the capacity of the estimation device 20 to correctly detect a component when it is actually present, the ideal value of the true-positive rate TVP being equal to 1.

[0335] The false positive rate TFP represents the tendency of the estimation device 20 to erroneously detect the presence of a component when it is not actually present, the ideal value of the false positive rate TFP then being equal to 0.

[0336] The true-negative rate TVN represents the capacity of the estimation device 20 to correctly estimate the absence of a component when it is actually absent, the ideal value of the true-negative rate TVN then being equal to 1.

[0337] The false negative rate TFN represents, for its part, the tendency of the estimation device 20 to wrongly estimate an absent component when it is in reality present, the ideal value of the false negative rate TFN then being equal to 0.

[0338] Those skilled in the art will further note that in Figures 7 and 8, the histograms of true-negative TVN rates, also noted TNR, as well as those of false-negative TFN rates, also noted FNR, have been deliberately multiplied by the factor -1, in order to distinguish them more easily from the true-positive TVP rates, also noted TPR, and false-positive TFP rates, also noted FPR, and thus improve the readability of these figures. These different rates are all positive and between 0 and 1.

[0339] In order to evaluate the overall performance of the estimations obtained with the estimation device 20 according to the invention, according to the first and second embodiments of the estimation algorithm, two complementary indicators are also calculated. A first ACC (Accuracy) indicator aims to represent the proportion of good detections made by the estimation device 20, the ideal value of this first ACC indicator being equal to 100%. A second MCC (Matthews Correlation Coefficient) indicator has a value between -1 and 1, the ideal value of the second MCC indicator being 1, and conversely a value equal to -1 characterizing an estimation device systematically indicating an estimation contrary to the real situation.

[0340] The first ACC indicator then verifies, for example, the following equation:

[0341] [Math.52] a rp _______VP + VN______ J Total Positive + Total Negative

[0342] where VP, VN, Total Positive and Total Negative are as defined previously.

[0343] The second MCC indicator then verifies, for example, the following equation:

[0344] [Math.53] _ VP.VN - FP.FN <vp + fp)(vp fn)(vn fp)( vn fn)

[0345] where VP, VN, FP and respectively FN are also as defined previously.

[0346] Figures 7 and 8 then show that the estimation device 20 according to the invention generates satisfactory results for the two embodiments of the estimation algorithm, while observing that the estimation results are slightly better for the first embodiment than for the second embodiment, in particular for the components C6, C8, CIO, Cil, C13 and C18 in these examples of Figures 7 and 8.

[0347] More precisely, whatever the embodiment of the estimation algorithm, the estimation results are excellent for the components C2, C3, C4, C5, C7, C15, C16, C17, C19 and C20, with TPR and TNR rates close to 1 in absolute value, and respectively FPR and FNR rates close to 0, both in Figures 7 and 8.

[0348] For the estimates obtained with the first embodiment of the estimation algorithm, the results shown in [Fig.7] are furthermore excellent for components C9 to C11 and for component C18, these results also being satisfactory for components C8, C12 and C13 and more average for components C6 and C14.

[0349] For the estimates obtained with the second embodiment of the estimation algorithm, the results as represented in [Fig.8] are also satisfactory. for components C9, Cl 1 and C12, these results being more average for components C8, C13 and C14, or even unsatisfactory for components C6 and C18.

[0350] As a further illustration of the quality of the estimation results obtained with the estimation device 20 according to the invention, the first MCC indicator is equal to 0.764 and the second ACC indicator is equal to 91.13% for the first embodiment of the estimation algorithm. For the second embodiment of the estimation algorithm, the first MCC indicator is equal to 0.700, and the second ACC indicator is equal to 87.74%.

[0351] Those skilled in the art will then observe that whatever the embodiment of the estimation algorithm, the values ​​of the first and second indicators MCC, ACC are satisfactory, with a value of the first indicator MCC each time greater than or equal to 0.7 and a value of the second indicator ACC close to 90%, this for both embodiments.

[0352] These first and second MCC, ACC indicators also make it possible to confirm the slightly better estimation results with the first embodiment of the estimation algorithm, in comparison with those of the second embodiment of the estimation algorithm.

[0353] The electronic estimation device 20 and the estimation method according to the invention then make it possible to provide satisfactory estimations of the composition of a product to be analyzed, that is to say satisfactory estimations of the set of components constituting said product, this from spectrometry measurements carried out via the spectrometry device 12 and for products in various technical fields, such as petrochemistry, pharmacy, gas phase chemistry, organic chemistry, physics, astrophysics or even biology.

[0354] Those skilled in the art will then understand that the estimation device 20 according to the invention makes it possible to effectively estimate the set of chemical or biological component(s) forming the product to be analyzed in technical fields such as those mentioned above.

[0355] In the examples of Figures 5 to 8, the product analyzed is a petrochemical oil, or base oil, and the components C1 to C20 likely to be contained in this analyzed product are for example detergents of the calcium phenates or calcium sulfonates type, polymeric dispersants (polyisobutylene succinimide), antioxidants of the zinc dithiophosphate type, and / or compounds of the molybdenum dithiocarbamate type.

[0356] It is thus understood that the electronic estimation device 20 according to the invention and the associated estimation method make it possible to improve the analysis of a spectral frequency measured by the spectrometry device 12 of each product to be analyzed, and then to more effectively draw the composition of each product, that is to say the sets of component(s) included in each product.< / vp> < / l> < / j>

Claims

Claims

1. Method for estimating a set of component(s) of a product from a spectrometry device (12), the spectrometry device (12) comprising an ionization source (14) inside which an element is capable of being ionized, a separation module (16) connected to the output of the ionization source (14) and capable of separating ions from the ionized element, and a detector (18) connected to the output of the separation module (16) and capable of measuring an ionic flux of the element, the method being implemented by an electronic estimation device (20) capable of being connected to the spectrometry device (12) and comprising the following steps: - obtaining (100) a group of signatures for potential components of the product, each signature being representative of an ionic flux of a respective potential component, and measured via the spectrometry device (12);- the acquisition (110) of a quantity of an ionic flux of the product, measured via the spectrometry device (12); characterized in that it further comprises a step (130) of estimating the set of component(s) of the product from the group of signatures and the quantity and via an estimation algorithm, in that the product is an additive linear mixture of the components where each component is weighted by a mixing coefficient, each mixing coefficient having a positive value, and in that, during the estimation step (130), the quantity is modeled in the form of a sum of the signatures of the group, each signature being weighted by the respective mixing coefficient, then the value of each mixing coefficient is calculated via the estimation algorithm.;

2. Method according to claim 1, in which the quantity and each signature correspond to the same measurement modality chosen from the group consisting of: - a mass spectrum if the separation module (16) comprises a mass spectrometry cell, - an ion mobility spectrum if the separation module (16) comprises an ion mobility spectrometry cell; and - a two-dimensional map if the separation module (16) comprises a mass spectrometry cell and a mass spectrometry cell. ion mobility trometry, the two-dimensional mapping representing, as a function of their mass to charge ratio, a drift time of ions in the ion mobility cell coupled to the mass spectrometry cell.

3. Method according to claim 2, in which the steps of obtaining (100) and acquiring (110) are respectively implemented for several different measurement modalities, and during the estimation step (130), the estimated set of component(s) of the product then depends on the groups of signatures and the quantities for the different measurement modalities.

4. A method according to any preceding claim, wherein the estimation algorithm comprises a minimization of a criterion dependent on the magnitude and the group of signatures; the criterion preferably being a quadratic criterion, such as a least squares criterion, or a logarithmic criterion.

5. Method according to claim 4, in which the estimation algorithm further comprises a determination of the presence or absence of each potential component in the product by comparing each mixing coefficient with a respective detection threshold; the detection thresholds preferably being predefined using a ROC curve from mixtures, real or simulated, of components.

6. Method according to any one of claims 1 to 3, in which the estimation algorithm comprises an implementation of a sparse decomposition algorithm from the quantity and the group of signatures; the sparse decomposition algorithm preferably being chosen from the group of greedy algorithms consisting of: MP algorithm, OMP algorithm and OLS algorithm.

7. A method according to claim 6, wherein an initial residual is equal to the measured quantity, and each subsequent residual is equal to the difference between the quantity and a linear combination of already present component signature(s) and respective mixing coefficient(s).

8. Method according to any one of the preceding claims, in which the method further comprises, prior to the estimation step (130), a step (120) of factorizing the group of signatures and the quantity with the same common factor, each signature being represented in the form of a product of the common factor and a factored signature, the quantity being represented in the form of a product of the common factor and a factored quantity; and the estimation step (130) then being carried out from the factored quantity and the group of factored signatures.

9. A computer program comprising software instructions which, when executed by a computer, implement a method according to any one of the preceding claims.

10. Electronic device (20) for estimating a set of component(s) of a product from a spectrometry device (12), the spectrometry device (12) comprising an ionization source (14) inside which an element is capable of being ionized, a separation module (16) connected to the output of the ionization source (14) and capable of separating ions from the ionized element, and a detector (18) connected to the output of the separation module (16) and capable of measuring an ionic flux of the element, the electronic estimation device (20) being capable of being connected to the spectrometry device (12) and comprising: - an obtaining module (22) configured to obtain a group of signatures for potential components of the product, each signature being representative of an ionic flux of a respective potential component, and measured via the spectrometry device (12);- an acquisition module (24) configured to acquire a quantity of an ionic flux of the product, measured via the spectrometry device (12); characterized in that it further comprises an estimation module (28) configured to estimate the set of component(s) of the product from the group of signatures and the quantity and via an estimation algorithm, in that the product is an additive linear mixture of the components where each component is weighted by a mixing coefficient, each mixing coefficient having a positive value, and in that the estimation module (28) is configured to model the quantity in the form of a sum of the signatures of the group, each signature being weighted by the respective mixing coefficient, then to calculate the value of each mixing coefficient via the estimation algorithm.;

11. Measuring system (10) comprising a spectrometry device (12) and an electronic device (20) for estimating a set of component(s) of a product from the spectrometry device (12), the electronic estimation device (20) being connected to the spectrometry device (12), characterized in that the electronic estimation device (20) is according to the preceding claim.

Images