Method for measuring the concentration of a chemical compound contained in a fluid by means of an optical measurement system

Abstract

The present invention relates to a method for determining the concentration of a chemical compound of a fluid by means of an optical measurement system for measuring a light intensity spectrum, such that: (i) by means of a learning base comprising a plurality of learning light intensity spectra measured for a plurality of learning fluids each having a predetermined concentration, a model is constructed in order to determine the concentration of at least one chemical compound in a fluid by training a one-dimensional convolutional neural network on the learning base, the network being such that the number of convolutional filters per layer decreases with the position of the layer in the network; (ii): by means of the optical measurement system, at least one light intensity spectrum of the fluid is measured; (iii) the constructed model is applied to the measured light intensity spectrum, and the concentration of the chemical compound in the fluid is determined.

Description

[0001] METHOD FOR MEASURING THE CONCENTRATION OF A CHEMICAL COMPOUND

[0002] CONTAINED IN A FLUID, BY MEANS OF AN OPTICAL MEASURING SYSTEM

[0003] Technical field

[0004] The present invention relates to the field of measuring the concentration of at least one chemical compound contained in a fluid, by means of an optical measuring system for measuring light intensity spectra.

[0005] The present invention applies advantageously and non-limitingly to the field of depollution of engine exhaust gases or to the depollution of industrial fumes, or to the field of monitoring polluting emissions from engine exhaust gases or the monitoring of polluting emissions from industrial fumes. The present invention can also and non-limitingly apply to the field of monitoring air quality or bathing water, to the field of biogas treatment, to the monitoring of gas leaks and more generally of industrial accidents.

[0006] Emission standards for polluting substances resulting from combustion (in internal combustion engines, turbines, industrial furnaces, incineration plants, thermal power plants, etc.) now require, in most cases, the reduction of the concentrations of these substances, for example using post-combustion catalytic treatments. Polluting substances in exhaust gases are typically unburned hydrocarbons, carbon monoxide (CO), nitrogen monoxide (NO) and nitrogen dioxide (NO2) commonly referred to by the acronym NOx, nitrous oxide N2O, ammonia (NH3), sulfur compounds such as hydrogen sulfide (H2S) or sulfur oxides such as sulfur dioxide (SO2), ozone O3, etc. Some of these substances are subject to strict regulations, according to which limit concentrations must be respected when released into the atmosphere.NOx, for example, has a harmful impact directly on human health and indirectly through the secondary formation of tropospheric ozone. Regulations already require control of the concentration of the NOx group formed by these nitrogen oxides, and may be required to evolve to require control of the concentrations of each nitrogen oxide individually, for example NO2. Regarding NOx, it is known to implement a selective catalytic reduction (SCR) system on the exhaust gas line which selectively reduces NOx to nitrogen through the action of a reducing agent generally injected upstream of the SCR system, for example ammonia or a compound generating ammonia by decomposition, such as urea which can be used advantageously in aqueous solution.Thus, ammonia also represents a substance for which concentration monitoring can be useful, not only in the context of concentration control imposed by current or future standards, but also for the purpose of diagnosis and / or control of the pollution control system. In this context, it is advantageous to be able to measure the concentration of polluting substances, in particular in order to control and / or diagnose post-combustion pollution control systems. For the purpose of combustion control, it may also be relevant to measure other combustion products such as carbon dioxide (CO2), oxygen (O2) and water (H2O).

[0007] In the field of air quality monitoring (e.g. confined or semi-confined spaces such as underground car parks, road or railway tunnels), ventilation and extraction and / or air supply systems are generally installed. However, these systems are often insufficient to guarantee correct air quality. For example, for metro-type railway networks, urban outdoor air is sampled to renew the air. However, urban air can already present a significant level of pollution, particularly in terms of fine particles but also in terms of gaseous pollutants (such as NOx, VOCs, ozone (O3) or sulfur derivatives such as sulfur oxides SOx), in particular due to car traffic, district heating, or nearby industrial activities.As for confined spaces such as road tunnels or parking lots, where the air is loaded with polluting particles and gases due to underground road traffic, this polluted air is extracted and released outside, thus polluting the external atmosphere close to these confined spaces. There is therefore a real need to estimate the concentration of pollutants in the air arriving in and / or extracted from these confined spaces, to improve the quality of the air in these spaces, and / or the air in their vicinity.

[0008] Biogas is produced by the anaerobic decomposition of organic waste, such as sludge from wastewater treatment plants, agricultural waste, and landfills. Biogas is mainly composed of methane (40 to 70%), CO2, and water vapor, but it also contains impurities, such as sulfur compounds (H2S, SO2, etc.), siloxanes, halogens, and VOCs (Volatile Organic Compounds). Biogas is therefore not directly usable. To be able to use biogas, it must be purified (or even purified), in particular to eliminate carbon dioxide and hydrogen sulfide, but also other impurities. This produces biomethane that can be injected into the natural gas distribution network or used as a biofuel.In addition, the biogas produced may be supplemented with an odorous molecule, most often sulfurous (mercaptans, for example Tetraydrotiophene (THT) in France) which requires precise dosage, verified analytically. For biogas, a particular use in purified form is the fuel cell, for which the tolerance thresholds for impurities or contaminants are particularly demanding in order not to damage the system. The development of methods for measuring each of the polluting substances is therefore of major interest, to control the biogas treatment process and for the qualification of the biomethane obtained after purification, with a view to its use.

[0009] Prior art

[0010] The following documents will be cited during the description:

[0011] Kawamura, Kensuke; Nishigaki, Tomohiro; Andriamananjara, Andry; Rakotonindrina, Hobimiarantsoa; Tsujimoto, Yasuhiro; Moritsuka, Naoki et al. (2021) Using a One- Dimensional Convolutional Neural Network on Visible and Near-Infrared Spectroscopy to Improve Soil Phosphorus Prediction in Madagascar. In : Remote Sensing, vol. 13, n° 8, p. 1519. DOI: 10.3390 / rs13081519.

[0012] Malek, Salim; Melgani, Farid; Bazi, Yakoub (2018) One-dimensional convolutional neural networks for spectroscopic signal regression. In : Journal of Chemometrics, vol. 32, n° 5, e2977. DOI: 10.1002 / cem.2977.

[0013] The classical quantification technique of chemical compounds present in a fluid used in absorption spectrometry techniques is based on a chemometric approach. More precisely, absorption spectrometry techniques allow the quantification of chemical compounds by measuring the intensity of the electromagnetic radiation that they absorb at different wavelengths (absorption spectrum), under the assumption that absorption is proportional to the concentration of these compounds. The absorption spectrum (A (2))) is obtained by applying the Beer-Lambert law to a reference light intensity spectrum, or baseline ( / 0(2)), and to the light spectrum ( / S (Â)) obtained in the presence of the sample of interest according to the following equation:

[0014] In the field of depollution and / or monitoring of engine exhaust gases or industrial fumes, patent application WO 2019 / 020326 A1 is known in particular, which relates to a system and a method for measuring the concentration of different chemical species contained in exhaust gases. More precisely, from the measurement of the intensity as a function of the wavelength of U radiation having passed through exhaust gases present in a measurement zone, an absorbance is determined as a function of the wavelength of the exhaust gases contained in the measurement zone. Then, the concentration of each compound present in the gases is determined, from the absorbance of the gases thus calculated, and also from predetermined absorbance characteristics and information relating to the temperature and pressure of each of the chemical species to be quantified.Different types of algorithms can be used for concentration calculations, such as least-squares fitting algorithms applied to the absorbance signals themselves, to derivatives of the absorbance signals, or to absorbance signals in the frequency domain (typically derived from a Fourier transform). Similarly, a number of statistical methods can be used for this process, such as principal component analysis (PCA) or partial least squares (PLS) algorithms.

[0015] The theoretically simple chemometric approach suffers from three robustness problems

[0016] Nonlinear system response: The response of the real system may deviate from the linear regime, either due to a too high concentration of the species (high optical density) or due to a nonlinear response of the sensor (spectrometer). While the first aspect can be calibrated and taken into account in chemometric theory, the second aspect requires fine calibration of the spectrometer and correction for each physical system. Interference from different species: While it is true that each species is characterized by a specific absorption spectrum, several species absorb over the same spectral range. Differentiating the different spectral signatures and the relative quantification is a major challenge in this aspect. The classical approach is to search and define spectral ranges of interest for each species to reduce interference.This aspect is often resolved empirically and must be adapted according to the type of gas mixture expected and the target species, but is very time consuming and can have a great impact on the quality of the results.

[0017] Adaptation to system evolution: the physical measurement system can evolve over time due, for example, to aging, fouling or misalignment of optical elements and active elements (spectrometer, lamp). Empirical mathematical solutions can be implemented but the effectiveness of this solution is strictly linked to the type of evolution expected.

[0018] More recently, methods for detecting or classifying molecules in liquids and sometimes gases, based on 1D convolutional neural networks and applied to spectroscopy, have been developed. In particular, we know the document (Malek et al., 2018) which concerns a method for measuring concentrations in food liquids (wine, orange juice) from near infrared (NIR) spectra to which 1D convolutional networks are applied. More precisely, the method described in this document uses 1D convolutional neural networks having at most 3 layers, a variable number of features less than or equal to 5 and a free convolution size limited to 10% of the size of the input data. The optimal structure is determined by an exhaustive calculation of all structure combinations on the training data.This phase is therefore very long and the optimal structure may present risks of overfitting. The final structure of the network is a priori unknown.

[0019] Furthermore, patent application CN114707598 A is known, which relates to a method for identifying and quantifying species in a mixed gas, from spectra and using a 1D convolutional neural network. This method, however, is silent on the structure of the neural networks used. In addition, it would seem that it is necessary to add advanced parameters ("meaningful high level features parameters") for the method to work.

[0020] We also know the document (Kawamura et al. 2021), which concerns a method for the quantification of phosphorus in soils, and which is based on an "increasing" structure of a convolutional neural network, that is to say on an increasing number of filters as one goes deeper into the network (more precisely 32 filters, then 64, 128, and finally 256), and on a data size divided by two between each layer. However, this method has the disadvantage of having a very large number of parameters (from 1 to several million) which makes it not very robust and subject to overfitting on small training and validation data sets.

[0021] The present invention aims to overcome these drawbacks. More specifically, the present invention relates to a method for determining the concentration of at least one chemical compound of a fluid, by means of at least one optical measuring system capable of measuring a light intensity spectrum and a particular one-dimensional convolutional neural network. More specifically, the one-dimensional convolutional neural network according to the invention is characterized in that the number of convolutional filters per convolutional layer decreases with the position of the convolutional layer in said convolutional neural network.Such a structure allows for faster implementation of a method for quantifying a chemical compound in a fluid based on one-dimensional convolutional neural networks, requiring less training data, less storage memory and allowing for greater robustness of the network, in the sense that it is less sensitive to overfitting phenomena. Furthermore, compared to chemometric approaches, the method according to the invention does not require a reference spectrum (baseline) to constitute absorption spectra, since the inputs of the method according to the invention are directly light intensity spectra.

[0022] Summary of the invention

[0023] The invention relates to a method for determining the concentration of at least one chemical compound of a fluid, by means of at least one optical measuring system capable of measuring a light intensity spectrum.

[0024] Said method comprises at least the following steps:

[0025] A) by means of a learning base comprising a plurality of learning light intensity spectra measured for a plurality of learning fluids whose concentration of at least said chemical compound is predetermined, a model is constructed for determining the concentration of at least one chemical compound of a fluid by training a one-dimensional convolutional neural network on said learning base, said one-dimensional convolutional neural network being such that a number of convolutional filters per convolutional layer of said one-dimensional convolutional neural network decreases with a position of said convolutional layer in said one-dimensional convolutional neural network;

[0026] B) by means of said optical measuring system, at least one light intensity spectrum of said fluid is measured;

[0027] C) applying said model constructed to determine the concentration of at least one chemical compound of a fluid in step A) to said at least one light intensity spectrum measured in step B), and determining said concentration of at least said chemical compound of said fluid.

[0028] According to one implementation of the invention, said optical measuring system may comprise at least one light source emitting radiation and at least one sensor, preferably a spectrometer, for measuring the change in light intensity as a function of the wavelength of radiation transmitted through said fluid, and from which, by means of said light source, radiation can be emitted through said fluid, then, by means of said sensor, at least the change in light intensity can be measured as a function of the wavelength of said radiation transmitted through said fluid. According to one implementation of the invention, said optical measuring system may be chosen from means capable of measuring a light intensity spectrum of said fluid by means of Ultra-iolet spectroscopy, visible spectroscopy, Raman spectroscopy, Infrared spectroscopy or near Infrared spectroscopy.

[0029] According to an implementation of the invention, said number of convolutional filters per convolutional layer of said one-dimensional convolutional neural network can decrease by a multiple of two from one convolutional layer to another.

[0030] According to an implementation of the invention, said one-dimensional convolutional neural network may be of the ResNet type.

[0031] According to an implementation of the invention, said one-dimensional convolutional neural network may further comprise at least one intermediate maxpooling layer arranged between two convolutional layers of said one-dimensional convolutional neural network.

[0032] According to one implementation of the invention, a data flattening step may be performed between a last convolutional layer of said one-dimensional convolutional neural network and a fully connected layer of said one-dimensional convolutional neural network.

[0033] According to an implementation of the invention, in step B), a plurality of light intensity spectra of said fluid can be measured by means of said optical measuring system, and step C) can be applied to an average of said plurality of measured light intensity spectra.

[0034] According to one implementation of the invention, said learning base can be diversified by means of at least one additional light intensity spectrum I corr W resulting from a correction made to one of said learning light intensity spectra ImesW according to a formula of the type: 1 rcorr e / nJ = l Imes / l I cAxW where A x ( ) is an absorption spectrum measured for one or more chemical compounds distinct from said at least one chemical compound and present in predetermined concentrations, and λ is the wavelength.

[0035] The invention further relates to a computer program product downloadable from a communications network and / or recorded on a computer-readable medium and / or executable by a processor, comprising program code instructions for implementing at least steps A) and / or C) of the method as described above, when said program is executed on a computer. List of figures

[0036] Figure 1 schematically shows an embodiment of the optical measuring system that can be implemented in the method according to the invention.

[0037] Figure 2 shows an example of a light intensity spectrum measured for a fluid whose concentration of NO, SO2, NH3, O2 and H2O is to be determined.

[0038] Figure 3A shows the comparison between the values ​​of the H2O and O2 concentrations determined by the method according to the invention and by a reference method.

[0039] Figure 3B shows the comparison between the values ​​of the concentration of NH3, NO, and SO2 determined by the method according to the invention and by a reference method.

[0040] Description of the embodiments

[0041] The method according to the invention relates to a method for determining the concentration of at least one chemical compound of a fluid, by means of at least one optical measuring system capable of measuring a light intensity spectrum of the fluid.

[0042] The fluid according to the invention may be a gas such as an engine exhaust gas, a biogas, or industrial fumes (for example from an incineration site, a biomass processing plant, etc.). The fluid according to the invention may be a liquid, such as bathing water, water suitable for consumption, or industrial waste water (for example from a dye works, a water treatment plant, oil production infrastructure, etc.).

[0043] By optical measuring system capable of measuring a light intensity spectrum, we mean an optical measuring system capable of producing, by measurement, a representative curve of revolution of the light intensity as a function of the wavelength. It is clear that the optical measuring system is at least capable of emitting radiation and measuring the intensity of radiation in the wavelength range characteristic of the chemical compound whose concentration is sought. By wavelength range characteristic of the chemical compound, we mean a wavelength range in which the absorption spectrum of the chemical compound (obtained by applying the Beer-Lambert law to the light intensity spectra) has non-zero (or even measurable) absorbance values.

[0044] According to one implementation of the invention, the optical measuring system may comprise at least one source emitting radiation and a sensor for measuring the change in light intensity as a function of the wavelength of radiation. From such an optical measuring system, by means of the light source, radiation can be emitted through the fluid of interest, for example within a measurement zone, then, by means of the sensor, at least the change in light intensity is measured as a function of the wavelength of the radiation having passed through the fluid in the measurement zone.

[0045] According to one implementation of the invention, the optical measurement system may in particular be chosen from means suitable for Ultraviolet (UV) spectroscopy, visible spectroscopy, Raman spectroscopy, Infrared (IR) spectroscopy or near Infrared spectroscopy. Thus, according to this implementation of the invention, the optical measurement system may comprise a spectrometer, such as a UV spectrometer, a visible spectrometer, a Raman spectrometer, or an IR spectrometer. The spectrometric measurement means are advantageous because they are compact, easy to implement and can be easily used online.More generally, according to this implementation of the invention, the optical measurement system may comprise: a light source for emitting radiation at least in a wavelength range characteristic of the chemical compound; a spectrometer for measuring a light intensity as a function of the wavelength in at least the wavelength range characteristic of the chemical compound.

[0046] Advantageously, in the case where the chemical compound corresponds to ammonia (NH3), sulfur dioxide (SO2), nitrogen oxide (NO), nitrogen dioxide (NO2), oxygen (O2) or water (H2O), the optical measurement system may comprise a light source for emitting at least one radiation in the UV, and a UV spectrometer. Indeed, the absorption spectrum of these chemical compounds has non-zero absorbance values ​​in the UV.

[0047] Advantageously, in the case where the chemical compound corresponds to carbon monoxide (CO), nitrous oxide (N2O) or carbon dioxide (CO2), the optical measurement system may comprise a light source for emitting at least one radiation in the 1'1 R, and an IR spectrometer or an IR detector. Indeed, the absorption spectrum of these chemical compounds has non-zero absorbance values ​​in the IR.

[0048] Figure 1 schematically shows an embodiment of the optical measuring system SMO that can be implemented in the method according to the invention, comprising a light source SL for emitting radiation RE, in a measuring zone ZM comprising a fluid FL, and a sensor SP for measuring the intensity as a function of the wavelength of the radiation RT having passed through the fluid FL in the measuring zone ZM along an optical path of length d.

[0049] According to an implementation according to which the method aims to determine the concentration of a plurality of chemical compounds of a fluid, the optical measurement system implemented may comprise a plurality of light sources and / or a plurality of sensors for measuring the evolution of the light intensity as a function of the wavelength of radiation, this set of sources and sensors being capable of emitting radiation and of measuring the intensity of radiation in the wavelength range characteristic of each of the chemical compounds whose concentration is sought.

[0050] According to an exemplary implementation of the invention, the light source may be a halogen-deuterium source, such as the AvaLight-DH-S-BAL model from Avantes (Netherlands) and / or the spectrometer may be a high-sensitivity fiber spectrometer, such as the AvaSpec-HS2048XL-EVO model from Avantes (Netherlands).

[0051] According to one implementation of the invention, the optical measurement system may comprise processing means for measuring a change in light intensity as a function of the wavelength of radiation, which may comprise a computer on which the Avasoft software from Avantes (Netherlands) is installed. Advantageously, the optical measurement system may further comprise means for transmitting (for example by electrical wire, by optical fiber or by a wireless communication system) the measurements made by the sensor of the optical measurement system to the processing means of the optical measurement system.

[0052] According to one implementation of the invention, the method may comprise a step, possibly prior, of calibrating the optical measurement system. Such a calibration step may comprise a measurement of a light intensity as a function of the wavelength transmitted through a reference fluid ideally transparent in the spectral range considered, such as nitrogen (N2) or dry air. According to one embodiment of this implementation of the invention, the calibration step may comprise the following steps:

[0053] - remission of radiation through the reference fluid within a measurement zone

[0054] - the measurement of light intensity as a function of the wavelength of the radiation having passed through the reference fluid.

[0055] The method according to the invention comprises at least the following steps:

[0056] 1) Construction of a model to determine the concentration of at least one chemical compound of a fluid from at least one light intensity spectrum

[0057] 2) Measurement of at least one light intensity spectrum of the fluid using the optical measuring system

[0058] 3) Application of the constructed model to determine the concentration of at least one chemical compound of a fluid from at least one measured light intensity spectrum Step 1) can be carried out only once, and beforehand. In other words, once the model to determine the concentration of at least one chemical compound of a fluid from at least one light intensity spectrum is constructed, this model can be applied one or more times, to acquired light intensity spectra. For example, if one wishes to determine the concentration of a chemical compound in different samples of the fluid of interest, one can repeat steps 2 and 3 for these different samples, the model being able to be constructed only once. For example, one can thus follow the evolution over time of the concentration of at least one chemical compound of a fluid, for example circulating in an exhaust pipe, a chimney of an industrial site, etc.

[0059] At least steps 1) and / or 3) may be implemented by computer means, in particular a computer, a processor or a calculator.

[0060] The steps of the method according to the invention are detailed below.

[0061] 1) Construction of a model to determine the concentration of at least one chemical compound of a fluid from at least one light intensity spectrum

[0062] During this step, by means of a learning base comprising a plurality of learning light intensity spectra measured for a plurality of learning fluids whose concentration of at least said chemical compound is predetermined, a model is constructed for determining the concentration of at least one chemical compound of a fluid by training a one-dimensional convolutional neural network (i.e. a 1D convolutional neural network) on said learning base.

[0063] According to the invention, the one-dimensional convolutional neural network is such that the number of convolutional filters per convolutional layer decreases with the position of the convolutional layer in the one-dimensional convolutional neural network. In other words, the number of convolutional filters per convolutional layer decreases as the position of the convolutional layer in the one-dimensional convolutional neural network moves deeper into the one-dimensional convolutional neural network relative to the input layer of the one-dimensional convolutional neural network. Indeed, convolutional neural networks consist of a plurality of convolutional layers, each layer comprising convolutional filters.

[0064] Typically in the field of convolutional neural networks in the broad sense (i.e. including in other fields than determining the concentration of a chemical compound in a fluid), the number of filters in the convolutional layers of a convolutional neural network increases with the position of the layer in the network. For example, the first convolutional layer may include 4 convolutional filters, the second convolutional layer may include 8 convolutional filters, the third convolutional layer may include 16 convolutional filters, etc., with one or two fully connected neural network layers at the end of the network. This is justified by the fact that simple operations are performed at the beginning, and then an increasing number of "features" (properties determined by the network) are processed depending on the diversity of the input population.

[0065] Surprisingly, the Applicant observed that a number of convolutional filters per convolutional layer decreasing with the position of the convolutional layer in the one-dimensional convolutional neural network allows for faster implementation of a method for quantifying a chemical compound in a fluid, requiring moreover less training data, less storage memory and allowing greater robustness to extrapolation, while ensuring sufficient accuracy. This is due to the modular (pre-established stacking of layers) and sparse (much fewer parameters than networks used conventionally) structure of the convolutional neural network according to the invention.Subsequently, the structure of the one-dimensional convolutional neural network according to the invention is referred to as the “decreasing structure”, and the structure conventionally used in the state of the art in the broad sense is referred to as the “increasing structure”.

[0066] Indeed, in the case of spectra (here light intensity spectra), the input population is fixed (a few molecules) and it is a question of predicting their quantity. In fact, the number of convolutional layers required is not as large. By taking a decreasing structure, the size of the intermediate data (between two layers of the network) decreases even more quickly and the total number of parameters remains contained, in particular, that of the final, fully connected layer.

[0067] According to one implementation of the invention, the ResNet type network can be used as a one-dimensional convolutional neural network. This type of network is a robust evolution of convolutional neural networks, by reducing the problems of vanishing gradient. The blocks of such networks have parallel paths at each layer which add the undersampled input information to the output of the neural layer. This type of network is for example described in the document (He et al., 2015). Advantageously, the number of convolutional filters from one convolutional layer to another of the convolutional neural network decreases by a multiple of two from one convolutional layer to another.

[0068] According to an implementation of the invention, the one-dimensional convolutional neural network may further comprise at least one intermediate layer, called "maxpooling" ("subsampling by maximum value" in French), arranged between two convolutional layers. Such a layer makes it possible to reduce the size of the data while preserving their main characteristics. Generally, each "maxpooling" layer makes it possible to reduce the size of the intermediate data by two and the number of characteristics is also reduced at each layer, preferably also in a ratio of 2 (for example, respectively 32 / 16 / 8 / 4 characteristics). This drastically reduces the number of parameters and the amount of calculations in the neural network. This improves the efficiency of the network and avoids over-learning.

[0069] Advantageously, a data flattening step can be performed between the last convolutional layer and the final layer of the convolutional neural network, called the fully-connected layer.

[0070] During step 1), the model of the method according to the invention is constructed by training the one-dimensional convolutional neural network as described above on a learning base, the learning base being formed by a plurality of learning light intensity spectra measured for a plurality of learning fluids whose concentration, in at least the chemical compound, is predetermined. We subsequently refer to “learning base data” for all the pairs formed by a light intensity spectrum and a concentration relating to each of the learning fluids of the learning base.

[0071] Advantageously, the learning base can be constructed from at least 100 learning fluids, preferably between 100 and 300 learning fluids. Indeed, such a learning base proves sufficient to solve the problem posed.

[0072] According to an implementation of the invention, the method according to the invention may comprise a prior step of constructing the learning base.

[0073] According to one implementation of the invention, the learning base can be constructed in the following manner: from a plurality of samples of learning fluids for which the concentration of at least the chemical compound of interest is predetermined, a light intensity spectrum is measured for each sample of the plurality of learning fluids, preferably by means of the optical measurement system according to any one of its embodiments as described above. According to one implementation of the invention, each sample of the plurality of samples of learning fluids can have a chemical composition identical to that of the fluid of interest, but with varied concentrations of the chemical compound of interest.The concentration of at least the chemical compound of interest can be predetermined by measurement, for example by means of reference analytical means such as Tunable Diode Laser analyzers (known by the acronym TDL) or by Fourier Transform Infrared Spectroscopy (known by the acronym FTIR). The concentration of at least the chemical compound of interest can alternatively be predetermined during the preparation of the different samples of learning fluids. Advantageously, the learning base can be constructed under pressure and / or temperature conditions close (for example to within 10%) to those of the implementation of the method according to the invention.

[0074] Advantageously, the learning base can be divided into 2, preferably 3, subsets: a first subset, comprising for example between 70% to 85%, preferably 80% of all the data in the learning base, intended for training the network: this data will directly determine the estimation error measurement of the network and the way to adjust the values ​​of the parameters of the neural network; this is the training database. a second subset, comprising for example between 15% and 30%, preferably 20% of all the data in the learning base, intended to validate the learning process during its execution: this data is used to monitor the success of the learning operation during its execution but does not directly influence the adjustment of the parameters of the network; this is the validation database.If the results obtained are not satisfactory, this may mean that the learning base is not sufficiently representative and that it may be necessary to increase the learning base with training data covering a wider range of concentrations of the chemical compound of interest. optionally but advantageously, a third subset, comprising a number of data from the learning base in a proportion equivalent to the validation database, intended to test the learning performance, at the end of the learning; this is the test database. According to an implementation of the invention, to constitute the subsets described above, a random draw can be carried out from all the data in the learning base.

[0075] According to one implementation of the invention, the data in the learning base can be diversified in order to increase the robustness of the model determined by learning. According to a first alternative, the learning base can be diversified by means of light intensity spectra measurements carried out regularly spaced over time and / or irregularly over time. For example, in the case of an online analysis of the composition of an exhaust gas circulating in an exhaust pipe, light intensity spectra measurements can be carried out over several non-consecutive days. This can make it possible to take into account the evolution over time of the composition of the exhaust gas to be analyzed and / or of the optical measurement system (its fouling for example). Generally speaking, a person skilled in the art knows how to set the ad hoc time intervals as a function of the fluid to be analyzed and the expected composition variations over time.According to a second alternative which can however be combined with the first alternative described above, the learning base can be diversified by means of measurements of light intensity spectra carried out using separate optical measuring systems. This can allow for the inherent defects of each of the optical measuring systems to be taken into account. According to a third alternative which can however be combined with the first and / or second alternatives described above, the learning base can be diversified by means of additional light intensity spectra I. corr ( ) resulting from a correction made to the learning light intensity spectra / mes (Â) (in other words to the spectra forming part of the learning base) according to a formula of the type: where A x( ) is a measured absorption spectrum for one or more chemical compounds different from the one(s) of interest (e.g. from the literature or previous studies), these compounds being present at known concentrations. In this way, the presence of unmeasured species whose absorption spectra are available from previous studies or in the literature can be simulated mathematically to increase the learning base.

[0076] Thus, at the end of this step, we obtain a model for determining the concentration of at least one chemical compound of a fluid. 2) Measurement of at least one light intensity spectrum of the fluid using the optical measuring system

[0077] During this step, it is a question of measuring at least one light intensity spectrum of the fluid of interest by means of the optical measuring system according to the invention. Advantageously, a plurality of light intensity spectra of the fluid of interest can be measured by means of the optical measuring system according to the invention, and an average of the spectra thus measured can be produced to which step 3) described below can be applied.

[0078] For this step, any of the implementation variants of the optical system according to the invention described above can be implemented. As a reminder, the optical measurement system according to the invention is at least capable of emitting radiation and measuring the intensity of radiation in the wavelength range characteristic of the at least one chemical compound whose concentration is to be determined.

[0079] According to an implementation of the invention according to which one seeks to determine the concentration of a plurality of chemical compounds of the fluid of interest, one can use an optical measurement system comprising a plurality of sources and sensors capable of emitting radiation and measuring the intensity of radiation in the wavelength range characteristic of all the chemical compounds whose concentration one seeks to determine.

[0080] 3) Application of the constructed model to determine the concentration of at least one chemical compound of a fluid from at least one measured light intensity spectrum

[0081] During this step, the model is applied to determine the concentration of at least one chemical compound of a fluid constructed in step 1) described above to said at least one light intensity spectrum measured in step 2 described above. In other words, during this step, the one-dimensional convolutional neural network trained as described in step 1) is applied to the spectrum(s) measured in step 2) above, optionally averaged. The concentration of at least the chemical compound of the fluid in question is thus determined.

[0082] According to an implementation in which the method aims to determine the concentration of a plurality of chemical compounds, and if the model for determining the concentration of at least one chemical compound of a fluid has been constructed in step 1) so as to determine the concentration of the plurality of chemical compounds, this step can be implemented only once.

[0083] Alternatively, if the model for determining the concentration of at least one chemical compound of a fluid has been constructed in step 1) so as to determine the concentration of a single chemical compound, steps 1) to 3) as described above are repeated for each chemical compound of the plurality of chemical compounds whose concentration is to be determined.

[0084] Furthermore, the invention relates to a computer program product downloadable from a communication network and / or recorded on a medium readable by a computer (for example an on-board computer) and / or executable by a processor. This program comprises program code instructions for implementing the method as described above, in particular steps 1) and / or 3) described above, when the program is executed on a computer.

[0085] Example

[0086] The characteristics and advantages of the method according to the invention will appear more clearly on reading the application example below.

[0087] For this application example, the fluid of interest corresponds to a mixture of gases with variable and random concentrations. The chemical compounds whose concentration we are seeking to determine are the following: NO, SO2, NH3, O2 and H2O. These gases have the particularity of being absorbent in a UV range from 200 to 400 nm.

[0088] This application example was implemented using a portable optical measuring system comprising a Xenon source and a compact spectrometer (weight < 50g) sensitive in the UV range 200 - 400 nm, and with a spectral resolution of 0.3 nm. In other words, the optical measuring system used for this application example makes it possible to cover the wavelength range characteristic of each of the chemical compounds whose concentration is to be determined.

[0089] A learning base consisting of a set of 265 light intensity spectra as well as concentrations of the chemical compounds of interest, associated with each of these light intensity spectra, was constructed. The light intensity spectra were measured using the optical measurement system used for this application example on 265 learning fluids characterized by varying concentrations of the chemical compounds of interest. The concentrations were determined using a reference measuring device, in this case an FTIR (Fourier Transform Infrared Spectroscopy) analysis bay.A first subset, comprising 80% of the entire training database data set, is intended for training the network (training database), and a second subset, comprising 20% ​​of the entire training database data set, is intended to validate the training process (validation database).

[0090] The model for determining the concentration of the chemical compounds of interest was built by training a one-dimensional convolutional neural network according to the invention on the learning basis described above. More precisely, the one-dimensional convolutional neural network is a ResNet type network, consisting of a stack of convolutional layers whose number of convolutional filters per layer is divided by 2 from one convolutional layer to another. Between each layer, a reduction in the data size of the “maxpooling” type is carried out until obtaining an intermediate data size, before the final fully connected layer, of the order of 10 to 30 times the number of output data to be predicted (For example, to predict 5 concentrations, the order of magnitude of the number of intermediate data before the last layer must be approximately 50 to 150).A data flattening step is performed between the last ResNet layer and the final fully connected neural network layer. The model for determining the concentration of the chemical compounds of interest constructed as described above was applied to an average of a plurality of light intensity spectra measured by means of the optical measuring system according to the invention.

[0091] Figure 2 shows an example of a light intensity spectrum, representing the revolution of light intensity I as a function of wavelength L, measured for a fluid for which we are seeking to determine the concentration of NO, SO2, NH3, O2 and H2O.

[0092] Figure 3A shows the comparison between the values ​​of the concentration C in H2O (denoted H2O in the figure) and in O2 (denoted O2 in the figure) determined by the method according to the invention (INV histograms) and the concentration measured by a reference method (REF histograms) for these same compounds. Figure 3B shows the comparison between the values ​​of the concentration C in NH3 (denoted NH3 in the figure), in NO and in SO2 (denoted SO2 in the figure) determined by the method according to the invention (INV histograms) and the concentration measured by a reference method (REF histograms) for these same compounds. A perfect correspondence can be observed between the value of the concentration determined by the method according to the invention and the concentration measured by a reference method for the compounds NO, SO2, NH3, O2 and H2O.An absolute deviation of the order of ppm and a relative deviation of approximately 5% are observed between the concentration values ​​determined by the method according to the invention and the concentrations measured by a reference method for NO. These deviations can be explained by the uncertainty in the measured data and the uncertainty of the model itself for the lowest absolute values. These results were obtained in less than ten minutes on an Intel Core i7 processor with 16 GB of memory. Thus, the method according to the invention makes it possible to determine the concentration of chemical compounds in a fluid precisely, quickly and efficiently.

[0093] More specifically, the present invention allows, compared to conventional methods based on chemometrics, to:

[0094] - Simplify and automate the implementation of a system for measuring the concentration of gases or liquids by spectroscopy, avoiding the need for chemometrics, and therefore reducing development times without sacrificing response accuracy.

[0095] - Improve the consideration of spectral interference between different species Limit characterization tests, and potentially improve precision, for example via compensation for non-linearities.

[0096] - Add the measurement of new species by numerically increasing the learning base, potentially with a very limited number of calibration measurements.

[0097] Furthermore, the present invention makes it possible, compared to existing methods based on convolutional neural networks, to:

[0098] - simplify the implementation of the convolutional neural network with a parsimonious structure in parameters (approximately a factor of 100 compared to the state of the art) limit the number of training data, limit the risk of overfitting, require less computing time and less memory capacity, which allows embedded use of the algorithm.

Claims

Claims 1. Method for determining the concentration of at least one chemical compound of a fluid, by means of at least one optical measurement system (OMS) capable of measuring a light intensity spectrum of said fluid, said method being characterized in that it comprises at least the following steps: A) By means of a learning base comprising a plurality of learning light intensity spectra measured for a plurality of learning fluids whose concentration of at least said chemical compound is predetermined, a model is constructed for determining the concentration of at least one chemical compound of a fluid by training a one-dimensional convolutional neural network on said learning base, said one-dimensional convolutional neural network being such that a number of convolutional filters per convolutional layer of said one-dimensional convolutional neural network decreases with a position of said convolutional layer in said one-dimensional convolutional neural network; B) by means of said optical measuring system (OMS), at least one light intensity spectrum of said fluid is measured; C) applying said model constructed to determine the concentration of at least one chemical compound of a fluid in step A) to said at least one light intensity spectrum measured in step B), and determining said concentration of at least said chemical compound of said fluid.

2. Method according to claim 1, wherein said optical measuring system (SMO) comprises at least one light source (SL) emitting radiation (RE) and at least one sensor (SP), preferably a spectrometer (SP), for measuring the evolution of the light intensity as a function of the wavelength of a transmitted radiation (FL) through said fluid (FL), and from which, by means of said light source (SMO), an emitted radiation (RE) is emitted through said fluid (FL), then, by means of said sensor (SP), at least the evolution of the light intensity is measured as a function of the wavelength of said radiation (RT) transmitted through said fluid (FL).

3. Method according to one of the preceding claims, in which said optical measurement system (SMO) is chosen from means capable of measuring a light intensity spectrum of said fluid (FL) by means of Ultra-Violet spectroscopy, visible spectroscopy, Raman spectroscopy, Infrared spectroscopy or near Infrared spectroscopy.

4. Method according to one of the preceding claims, wherein said number of convolutional filters per convolutional layer of said one-dimensional convolutional neural network decreases by a multiple of two from one convolutional layer to another.

5. Method according to one of the preceding claims, in which said one-dimensional convolutional neural network is of the ResNet type.

6. Method according to one of the preceding claims, wherein said one-dimensional convolutional neural network further comprises at least one intermediate maxpooling layer arranged between two convolutional layers of said one-dimensional convolutional neural network.

7. Method according to one of the preceding claims, wherein a data flattening step is performed between a last convolutional layer of said one-dimensional convolutional neural network and a fully connected layer of said one-dimensional convolutional neural network.

8. Method according to one of the preceding claims, in which, in step B), a plurality of light intensity spectra of said fluid are measured by means of said optical measuring system, and in which step C) is applied to an average of said plurality of measured light intensity spectra.

9. Method according to one of the preceding claims, in which said learning base is diversified by means of at least one additional light intensity spectrum l corr (A) resulting from a correction made to one of said learning light intensity spectra / mes ( / l) according to a formula of the type: 1 Icorr e / nJ = l Imes Q / l I cAxW where A X (A) is an absorption spectrum measured for one or more chemical compounds distinct from said at least one chemical compound and present in predetermined concentrations, and λ is the wavelength.

10. Computer program product downloadable from a communication network and / or recorded on a computer-readable medium and / or executable by a processor, comprising program code instructions for implementing at least steps A) and / or C) of the method according to one of the preceding claims, when said program is executed on a computer.

Images