SYSTEM AND METHOD FOR DETERMINING CHEMICAL COMPOSITION OF A LIQUID METALLURGICAL PRODUCT - Patent application

Abstract

A device for determining the chemical composition of a liquid metallurgical product emitting electromagnetic radiation, the device comprising a collection probe configured to acquire electromagnetic radiation emitted by the metallurgical product in a predetermined wavelength range Δλ, spectroscopic means connected to the collection probe and configured to generate a spectral signal of the acquired electromagnetic radiation, and processing means comprising a database of reference radiances. The invention also relates to a method of using said device.

Description

[Technical field]

[0001] The present invention is in the field of metallurgy, and more particularly in the field of liquid metallurgical products. [Background technology]

[0002] The present invention more particularly relates to systems and methods for determining the chemical composition of liquid metallurgical products using electromagnetic radiation emitted by such metallurgical products.

[0003] A known problem in metallurgy is to adequately characterize metallurgical products when they are subjected to very high temperatures, typically above 1000° C., and therefore in liquid form due to their melting.

[0004] From publication WO 2016 / 181185 a sensor is known that allows obtaining the chemical composition of a solid slag portion using an installation comprising a light source adapted to illuminate the slag portion, an optical system adapted to collect reflected light from the slag portion, and processing means adapted to obtain a data set from the collected light, said data set defining a matrix comprising values ​​representative of the intensity of the reflected light collected from a number of points, such that a regression algorithm is implemented that estimates the chemical composition of the slag portion based on the intensity of the reflected light and its wavelength.

[0005] However, the system and method of publication WO2016 / 181185 cannot be applied to unknown metallurgical products. Furthermore, the slag part must be excited by a light source, since only reflected light is collected by the equipment. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] International Publication No. 2016 / 181185 Summary of the Invention [Problem to be solved by the invention]

[0007] It is therefore an object of the present invention to remedy the shortcomings of the prior art by providing a system and method for determining the chemical composition of any liquid metallurgical product that emits electromagnetic radiation. [Means for solving the problem]

[0008] For this purpose, the invention relates to a device for determining the chemical composition of a liquid metallurgical product emitting electromagnetic radiation, the device comprising at least: - a collection probe configured to acquire electromagnetic radiation emitted by the metallurgical product in a predetermined wavelength range Δλ; - spectroscopic means connected to the collection probe and configured to generate a spectral signal of the acquired electromagnetic radiation; - the observed radiance L from the generated spectral signal obs (λ,T est ) and from the calculated observed radiance, calculate the temperature T of the metallurgical product in a predetermined wavelength range Δλ. est and spectral emissivity ε est (λ,T est ), and the processing means calculates a reference radiance L of X samples i (where i ranges from 1 to X) in a predetermined wavelength range Δλ. ref,i (λ,T ref,n ), and Z reference temperatures T ref,n (n varies from 1 to Z), for each reference radiance: Reference temperature T in a given wavelength range Δλ ref,n The reference spectral emissivity ε related to ref,i (λ,T ref,n ), and the chemical composition of said sample i, including a database, Estimated spectral emissivity ε est (λ,T est ) and the estimated temperature T est, given wavelength range Δλ and estimated temperature T est At the reference temperature T ref,n and the reference spectral emissivity ε ref,i (λ,T ref,n ) compared to the observed radiance L obs (λ,T est ) and the best matched reference radiance L bf (λ,T ref ) and determine the best-fit reference radiance L bf (λ,T ref ) to determine the chemical composition of the liquid metallurgical product (2).

[0009] The device may also include the following optional features, taken individually or according to any possible technical combination: the spectroscopic means comprises a plurality of spectrometers, each spectrometer being connected to a collection probe and configured to generate a portion of the spectral signal in a determined portion of the predetermined wavelength range Δλ. the spectroscopic means are adapted to generate a spectral signal in a spectral range between 200 and 12000 nanometers, which corresponds to a predetermined wavelength range Δλ; the processing means are also adapted to correct each part of the spectral signal with at least a calibration constant calculated by said processing means and associated with the spectrometer taken into account;

[0010] The present invention also provides a method for determining the chemical composition of a liquid metallurgical product emitting electromagnetic radiation, using a device according to the invention, said method comprising the following steps: i. obtaining (E1) electromagnetic radiation emitted by a metallurgical product (2) in a predetermined wavelength range Δλ; ii. generating a spectral signal (E2) by separating and measuring the spectral components of said electromagnetic radiation emitted by the metallurgical product (2); iii. From said spectral signal, the observed radiance L of the metallurgical product (2) obs (λ,T est) (E3); iv. The observed radiance L obs (λ,T est ) from the temperature T est and spectral emissivity ε est (λ,T est ) and a step (E4) of estimating v. The estimated spectral emissivity ε est (λ,T est ) and temperature T est For each reference radiance L ref,i (λ,T ref ) reference spectral emissivity ε ref,i (λ,T ref ) and criterion T ref a step (E5) of comparing with the temperature; vi. Observed radiance L obs (λ,T est ) and the best matched reference radiance L bf (λ,T ref ) (E6), vii. A step (E7) of determining the chemical composition of the discharged metallurgical product (2); Includes.

[0011] The method may also include the following optional features, taken individually or according to any possible technical combination: In step iv, the atmospheric extinction coefficient is also estimated from the observed radiances and said observed radiances are correlated with said extinction coefficients. - Step iii: Spectral signals are expressed as ideal blackbody radiance L BB A sub-step of transforming into (λ,T); This ideal radiance L BB correcting (λ,T) using at least a calibration constant calculated by the processing means and associated with the spectroscopic means, said calibration constant being calculated using a calibration lamp of known radiation temperature and known emissivity in a range of wavelengths. - Step iv is carried out according to: determining a vector, the coordinates of which are randomly selected variables representing at least radiation temperature and emissivity in a wavelength range; Expected radiance L based on selected variables exp The sub-step of computing (λ,T); Fit an inference probabilistic model by comparing observed and expected radiances to estimate the observed radiance L obs (λ,T est ) the expected radiance L exp A substep of randomly varying the expected variance by randomly varying selected variables until convergence of (λ,T); Using the adapted probabilistic inference model, the radiation temperature T est and emissivity ε est (λ,T est ) and the sub-step of estimating This will be carried out in accordance with: one of the coordinates of the vector comprises the atmospheric attenuation coefficient, said attenuation coefficient also being estimated by using the adapted probabilistic inference model. Step iv is carried out using a reduced set of wavelengths determined from the wavelength range using a triangular transfer function. Steps v to vii are carried out by a multi-layer perceptron implemented by the processing means.

[0012] Other characteristics and advantages of the invention will become apparent in the course of the following description, given by way of example and in no way limiting, with reference to the accompanying drawings, in which: [Brief description of the drawings]

[0013] [Figure 1] FIG. 1 is a schematic diagram of a system of the present invention. [Diagram 2]FIG. 2 is a graph illustrating the application of a triangular function to determine a set of wavelengths considered from a determined wavelength range. [Diagram 3] FIG. 3 shows the successive steps of the method of the invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] First, it should be noted that in the figures, like reference numerals designate like elements regardless of the figure in which they are featured and the form of those elements. Similarly, if elements are not specifically referenced in one figure, their reference can be readily found by reference to another figure.

[0015] It should also be noted that while the figures primarily depict one embodiment of the subject matter of the present invention, other embodiments may exist that correspond to the definition of the present invention.

[0016] The spectroscopic system 1 and method of the present invention finds application in estimating the chemical composition of a liquid metallurgical product 2, such as slag or liquid steel.

[0017] Hot elements, such as liquid metals like liquid steel, emit electromagnetic radiation that is correlated to their respective radiative temperature and chemical composition. These radiations are typically emitted in a large radiation spectrum ranging from ultraviolet to far infrared wavelengths, including infrared and visible wavelengths that are directly related to the radiative temperature that induces the red to white appearance of liquid metallurgical products.

[0018] Physical parameters can be extracted from measurable characteristics of said electromagnetic radiation. These physical parameters include: T, designated temperature, is the temperature of the liquid metallurgical product 2 emitting electromagnetic radiation; ε(λ,T), named spectral emissivity or simply emissivity, is the radiant flux of electromagnetic radiation emitted by a surface element at a given temperature T and a given wavelength λ, compared to a perfect black body. Thus, the value of emissivity is included between 0 and 1. ·L(λ,T), named the spectral radiance, is the radiant flux emitted by a given surface per unit solid angle, per unit projected area, and per unit wavelength. In the following, we will use the term "radiance".

[0019] While acquiring the electromagnetic radiation emitted by the liquid metallurgical product 2, the spectroscopic system 1 of the present invention as will now be described is provided to estimate at least the above-mentioned physical parameters and then use these estimated physical parameters to determine the chemical composition of the liquid metallurgical product 2 under consideration.

[0020] As depicted in Fig. 1, the spectroscopic system 1 of the present invention comprises at least one broad spectral range collection probe 3, meaning that the collection probe is fully operable for wavelengths at least from 0.1 micrometer to 15000 micrometer. This collection probe 3 comprises a collection head 4 arranged to collect radiation emitted by the hot metallurgical product 2, a collimator (not shown) to generate a parallel light beam 19, and focusing means 5 arranged to emit a focused beam 6 through an output 7 of the probe 3. Said output 7 is connected to an optical transmission means, for example a set of optical fibers forming a fiber bundle 12, arranged to transmit the focused beam 6 through the optical fibers 13-17 of the bundle 12.

[0021] The system 1 also comprises spectroscopic means 8 arranged to separate and measure the spectral components of the focused beam 6 and generate a spectral signal of the electromagnetic radiation emitted by the metallurgical product. The accuracy of the generated spectral signal depends on the resolving power of the spectroscopic means 8. In other words, the spectroscopic means 8 separates the focused beam 6 into M components m, each component m having a specific wavelength λ m M is a parameter related to the resolution, e.g., intensity. The higher the resolution, the larger M. Typically, M is 3000 or more.

[0022] 1, the spectroscopic means 8 comprises three spectrometers 9, 10, 11, each configured to analyze the collected beam 6 and generate a specific spectral signal in a determined wavelength range. Each spectrometer 9-11 is optically connected to the collection probe 3 via a fiber bundle 12.

[0023] The first spectrometer 9 of the spectroscopic means has a wavelength range from 0.2 to 1.1 μm, in other words, the first spectrometer 9 is configured to generate a spectral signal from the emitted radiation in the wavelength range from 0.2 μm to 1.1 μm, which corresponds to ultraviolet and visible light.

[0024] The second spectrometer 10 of the spectroscopic means has a wavelength range comprised between 0.9 and 2.6 μm, in other words, the second spectrometer 10 is configured to generate a spectral signal from the emitted radiation in the wavelength range between 0.9 μm and 2.6 μm, which corresponds to near infrared radiation.

[0025] The third spectrometer 11 of the spectroscopic means 8 has a wavelength range comprised between 2.5 μm and 12 μm. In other words, the third spectrometer 11 is configured to generate a spectral signal from the emitted radiation in the wavelength range between 2.5 μm and 12 μm, which corresponds to mid-infrared radiation.

[0026] Advantageously, the spectroscopic system 1 also comprises a laser device 18 connected to the collection probe 3 via a fibre bundle 12. This laser device 18 is typically a class B laser emitting visible light, for example green light at 532 nanometres, and is provided for pointing at an acquisition plane. This laser device 18 allows the selection of an acquisition site.

[0027] As described above, the collection probe 3, the three spectrometers 9-11, and the laser device 18 are all connected by a fiber bundle 12 with four inputs connected respectively to the laser device 18 and the three spectrometers 9-11, and one output connected to the collection probe 3.

[0028] More precisely, the laser device 18 is connected to the probe 3 via one low hydroxyl ion silica optical fiber 17, the first spectrometer 9 is connected to the probe 3 via another low hydroxyl ion silica optical fiber 13, the second spectrometer 10 is connected to the probe 3 via two low hydroxyl ion silica optical fibers 14 and the third spectrometer 11 is connected to the probe 3 via two polycrystalline infrared optical fibers 15.

[0029] Of course, all seven optical fibers mentioned above are connected to the output of the collection probe 3 16 .

[0030] Finally, the spectroscopic system 1 comprises processing means configured to generate from the spectral signal an observed radiance of the metallurgical product 2. According to a first algorithm, the processing means is configured to use a probabilistic reasoning model to determine an expected radiance from the observed radiance and then to estimate the temperature and the emissivity of the metallurgical product 2.

[0031] Furthermore, according to the second algorithm, the processing means is configured to estimate the chemical composition of the discharged metallurgical product 2, such as the content of in-silicon dioxide SiO2, aluminum oxide Al2O3, iron (II) oxide FeO, iron (III) oxide Fe2O3, calcium oxide CaO and magnesium oxide MgO, by means of a known regression algorithm, advantageously a multilayer perceptron, implemented by said processing means.

[0032] As will be described below, to implement this second algorithm, the processing means calculates the reference radiance L ref,i (λ,T ref,n ) database for each standard radiance L ref,i (λ, T ref,n ) is the defined reference temperature T ref,n and the known spectral emissivity ε at wavelength λ ref,i (λ,T ref,n ) (further referred to as reference emissivity) is associated with sample i. Thus, each reference emissivity value ε ref,i (λ,T ref,n) is the reference temperature value T in the database. ref,n and related to the wavelength value λ. i varies from 1 to X, where X is, for example, at least equal to 30; the higher X, the more different chemical compositions are included in the database. n varies from 1 to Z, where Z is, for example, at least equal to 5.

[0033] Furthermore, the chemical composition of each sample i in the database is known and the reference emissivity ε considered in the wavelength range Δλ ref,i (λ,T ref,n ) and the reference temperature T ref,n To determine the chemical composition of the liquid metallurgical product 2, the process involves: Estimated emissivity ε est (λ,T est ) and the estimated temperature T est , T ref,n =T ref For each reference radiance L ref,i (λ,T ref,n ) reference emissivity ε ref,i (λ,T ref,n ) and the reference temperature T ref,n Compared to, Observed radiance L obs (λ,T est ) and the best matched reference radiance L bf (λ,T est ) is determined, Determine the chemical composition of the discharged metallurgical product 2; It is configured as follows.

[0034] According to the invention, the temperature T est and emissivity ε est (λ,T est A method for estimating the chemical composition of the electromagnetic radiation-emitting liquid metallurgical product 2 is now described.

[0035] In a first step, the collection probe 3 is pointed at the acquisition site selected by the user. To accurately select this acquisition site, the laser 18 is powered up and the collection probe 3 is moved until the laser 18 is pointing at the acquisition site.

[0036] In a second step E1, the collection probe 3 acquires the electromagnetic radiation emitted by the liquid metallurgical product 2 at the acquisition site and concentrates this acquired radiation into a focused beam 6 which propagates from the output 7 of the probe 3 via a fiber bundle 12 to the three spectrometers 9-11 of the spectroscopic means 8. Each spectrometer 9-11 then generates a spectral signal in its specific wavelength range. The processing means of the spectroscopic system 1 then generate E2 a combined spectral signal in the determined wavelength range from 0.2 μm to 12 μm.

[0037] In a third step E3, the processing means derives from the combined spectral signal the observed radiance L of the metallurgical product 2. obs (λ,T est ) is generated from this observed radiance L obs (λ,T est ) is generated according to the sub-steps described below.

[0038] In a first sub-step, the processing means converts the combined spectral signal into an ideal blackbody radiance L through a polynomial calibration function measured by the spectroscopic system on a blackbody furnace. BB (λ,T est ) and the polynomial function follows the formula:

number

[0039] In the second substep, this ideal radiance L BB (λ,T est ) is corrected by a constant K which is directly related to the spectroscopic means 8. Because the spectrometers 9-11 of the spectroscopic means 8 do not measure the same area, and because the instrumentation of the spectroscopic means 8 (collection probe 3, fiber bundle 12, collimator 5) introduces inaccuracies into the measurements, the correction constant K smust be applied to each spectrometer S. This constant K is therefore a vector whose coordinates are the constants K1, K2, and K3 of the first spectrometer 9, the second spectrometer 10, and the third spectrometer 11, respectively.

[0040] Initial values ​​for these constants K1 to K3 are determined using the spectroscopic system 1 on a calibration lamp with known emissivity and known radiation temperature in the wavelength range.

[0041] Ideal radiance L corrected by the processing means using the calculated constant K BB (λ,T est ) is the observed radiance L obs (λ,T est ).

[0042] In a fourth step E4, the processing means implements a first algorithm to derive the temperature T est and emissivity ε est (λ,T est ) is estimated.

[0043] As mentioned above, radiance is known to be a function of several physical parameters including the emissivity and radiation temperature in the wavelength range of the metallurgical product 2. Therefore, the first algorithm involves applying these unknown parameters to the observed radiance L obs (λ,T est We implement a radiative transfer model that aims to extract these unknown parameters so that the radiative transfer model can be fitted to the

[0044] The processing means then calculates the expected radiance L according to a Markov Chain Monte Carlo approach, i.e., MCMC, using the Metropolis-Hasting algorithm, also named Metropolis-Hasting random walk. exp (λ,T est) The expected radiance is given by:

number

number

[0045] The MCMC method implements a Bayesian inference model and allows to estimate unknown parameters, which are the radiation temperature, the emissivity in the range of wavelengths λ, and the concentrations of water vapor and carbon dioxide, by comparing the calculated expected radiance with the observed radiance. Furthermore, the calibrated correction coefficients K1 to K3 are fine-tuned by the Metropolis-Hastings algorithm.

[0046] The Bayesian inference model constructs a probability distribution over the values ​​of the seek unknown parameters using the following formula:

number

number

[0047] P(θ│x), known as the posterior probability distribution, is obs (λ,T est) is the probability of each seek unknown parameter value given . This is the probability that the processing means needs to calculate in order to estimate the unknown parameters.

[0048] Known as the likelihood, P(x│θ), is the probability that the observed radiance L obs (λ,T est ) is distributed. The calculation of this probability will be explained later.

[0049] P(θ), known as the prior distribution, is the prior knowledge used to make the processing means compute the Metropolis-Hastings algorithm. In other words, a set of parameters θ is randomly selected at the beginning of the computation, as will be explained later.

[0050] P(x) is the evidence that the parameters were generated by the radiative transfer model. This quantity cannot often be calculated. However, P(θ│x) can be estimated using MCMC methods, in particular the Metropolis-Hastings algorithm, by simply defining the likelihood P(x│θ) and the prior distribution P(θ).

[0051] A detailed implementation of the fourth step E4 of the method of the invention will now be described.

[0052] As mentioned above, emissivity is wavelength dependent, meaning that each wavelength is associated with an emissivity. The resolution of the spectroscopic means 8 is m , so that M emissivity values ​​can be determined by the processing means implementing the method of the invention.

[0053] However, since M is greater than or equal to 3000, the processing means is not able to estimate M emissivity values ​​based on M wavelengths due to limited computational power. To solve this problem, a selection of a reduced set of N wavelengths must first be achieved by the processing means, for example using a membership triangle function according to the following formula:

number

[0054] The processing means associates the M wavelengths with a triangular function to calculate N sets of wavelengths, where N is much smaller than M, and the value of N is typically around 4 to 10. Since the N sets are calculated using fuzzy logic, the sets are typically named fuzzy sets.

[0055] Furthermore, the wavelength λ m Specific emissivity ε at m are fuzzy emissivity values ​​ε defined by the processing means at the centers of the considered fuzzy sets of wavelengths, respectively. m is calculated as a weighted value of

number

[0056] For example, as depicted in Figure 2 for N=8, the emissivity ε m =ε(λ m ) is a combination of 33% of the fuzzy emissivity ε4 = ε(λ4) and 67% of the fuzzy emissivity ε5 = ε(λ5). As a result, only N values ​​of emissivity are directly calculated by the processing means, which is the number of emissivities m = 1 to N for each emissivity ε m Other emissivity values ​​within the wavelength range are calculated from one or more fuzzy emissivities. Thus, λ can fairly represent a wavelength or a wavelength range when using such a triangular function, or any other similar function that allows reducing the set of wavelengths.

[0057] Once the reduced set of wavelengths has been selected, the processing means using the Metropolis-Hasting algorithm randomly generates a vector θ whose coordinates are the unknown parameters randomly selected by the processing means. The processing means then calculates a prior distribution P(θ) from the randomly selected parameters using a uniform distribution function and a normal distribution function.

[0058] The uniform distribution function is given by the following formula:

number

[0059] The uniform distribution function is given by the following formula:

number

[0060]

number

number

number

[0061]

number

number

number

number

number

number

number

number

[0062] Using the results of the calculations of each of the above distributions, the processing means determines the prior distribution P(θ0).

[0063] Furthermore, once the coordinates of the vector θ are randomly determined, the processing means uses the unknown parameter values ​​of θ to calculate the expected variance L according to the above formula. exp Calculate the first value of (θ0).

[0064] In this way, the processing means exp Normal distribution function centered on (θ0)

number

Number

[0065] The processing means is a normal distribution centered on θ0

Number

[0066] A number r1 is then calculated according to the following formula:

Number

[0067] The random number r’ also follows the standard uniform distribution

Number

[0068] The processing means compares r1 and r’, and if r1 > r’, it retains the vector θ1 and calculates the vector θ2 according to the normal distribution

Number

Number

[0069]

number

[0070] The Metropolis Hastings random walk performed by the processing means has expected variance L exp (λ,T) is the final vector

number

[0071] Observed radiance L obs (λ,T est ), the MCMC method can be used to calculate the radiation temperature T est and spectral emissivity ε est (λ,T est In addition, the internal correction parameters K1 to K3 of the spectrometers 9, 10, and 11, and the atmospheric water vapor

number

number

[0072] The processing means executes a second algorithm to implement a fifth step E5, a sixth step E6 and a seventh step E7 of the method of the invention. In the fifth step E5, the processing means calculates the observed radiance L obs (λ,Test ) estimated emissivity ε est (λ,T est ) and temperature T est , T ref,n =T ref For each reference radiance L ref,i (λ K ,T ref,n ) reference emissivity ε ref,i (λ,T ref,n ) and the reference radiation temperature T ref Using the comparison result obtained in the fifth step E5, the processing means, in a sixth step E6, compares the observed radiance L obs (λ,T est ) and the best matched reference radiance L bf (λ,T ref Finally, the processing means determines the best-fit reference radiance L bf (λ,T ref ) to the excreted metallurgical product 2. In this way, the processing means determines the chemical composition of the excreted metallurgical product 2.

[0073] Advantageously, this second algorithm is a known regression algorithm and preferentially is a multi-layer perceptron, an artificial neural network comprising a number J of interconnected neural layers, said artificial neural network being realized by processing means.

[0074] Each digital neuron of a layer of a neural network comprises a number of inputs provided to receive data from the outputs of neurons of the preceding layer and an output provided to transmit output data to a neuron of the next layer. More precisely, the input data are typically processed by the digital neuron using an operator such as a combination function that sums all the input data, each weighted by a synaptic weight, to generate output data. Furthermore, each neuron may include a nonlinear transfer function, a threshold function, provided to generate output data when the weighted sum of the input data exceeds a threshold defined by the threshold function. Typically, said threshold function is a sigmoid function, the formula of which is

number

[0075] Each input of a neuron j in a neural layer is influenced by a particular synaptic weight. The synaptic weights of the inputs are initially chosen randomly when the multi-layer perceptron is programmed. As we will see later, these synaptic weights are adjusted during the training phase of the neural network.

[0076] The artificial neural network is provided to predict and classify data by performing a regression analysis of the data fed into it through the input of the neural network, i.e. through the input of the neurons of the first layer. The fed data is a vector X of j coordinates. 0 and is processed successively by the first neural layer and all intermediate neural layers.

[0077] The output data means the data obtained at the output of the neural network, and the input data is represented by a vector X 0 A vector Y of j coordinates related to J It is.

[0078] Vector X 0 The neural network has j coordinates, and the first layer and all intermediate layers also contain j neurons with j inputs each, where each input is influenced by a particular synaptic weight. The final layer produces an output vector Y J Since has j coordinates, it has j neurons.

[0079] To summarize the data regression performed by a neural network, the input data received by each neuron in layer j is the output data generated by the outputs of all neurons in layer j-1. In other words, each layer j of a neural network receives an input vector X j The output vector Y processed by the transfer function of the neuron in layer j from J Generate Xj The coordinates of are the output vector Y J-1 and the input vector X j+1 The coordinates of the output vector Y J These are the same coordinates.

[0080] At the output of the neural network, the output vector Y J is the input vector X by the output layer of the neural network. j =Y J-1 Thus, data regression is performed via data propagation in successive layers of the neural network.

[0081] Regarding the fifth step E5 to the seventh step E7 of the method of the invention, the input vector X 0 The coordinates of are the central wavelengths λ of N fuzzy sets. m Estimated radiation temperature T est , and the estimated emissivity ε est,m (λ m ,T est ) are different values ​​of this input vector X 0 are successively processed by layers of a neural network to produce an output vector Y whose coordinates are the proportions of different chemical compounds in the metallurgical product 2. M In other words, the processing means in the fifth to seventh steps generates the estimated spectral emissivity ε est (λ,T est ) and the estimated temperature T est from which the chemical composition of the metallurgical product 2 is determined. However, such processing is only possible if the weights are correctly adjusted.

[0082] A key issue in this method is to finely tune the synaptic weights to obtain good data classification and / or prediction.

[0083] Reference radiance L ref,i (λ,T ref,n ) database of different radiation temperatures T ref,nThe method is previously constructed by measuring a collection of samples i with known chemical composition at . For each sample, the observed radiance is measured and the first algorithm described above is applied to the observed radiance to obtain the wavelength λ k Some reference emissivity values ​​ε ref,i,k (λ k ,T ref ) is obtained.

[0084] In this way, the processing means builds up a database of samples, each sample having an input vector X' associated with a reference emissivity value at a different reference radiation temperature. 0 And, X' 0 and output vector Y' related to known chemical composition J In other words, the emissivity value at a particular radiation temperature is directly related to the chemical composition of the metallurgical product 2.

[0085] To train the neural network, a backpropagation function is implemented. The output vector Y' is known. J Input vector X' correlated with 0 is input to the neural network. X' 0 is processed by the neural network and the calculated output vector Y J Then, Y J Y' J to determine an error value. The processing means then compares this error value with a predetermined threshold value.

[0086] If the error value is greater than a threshold, this error value, or preferentially the error gradient, is back-propagated in the neural network to slightly adapt the weights of each neuron input. Then, X' 0 to the new output vector Y J is calculated and the output vector Y Jis again compared to Y'J using the loss function and a new error value is calculated. The backpropagation step is repeated as long as the error value or its gradient is greater than a threshold. When the error value becomes smaller than the threshold, the neural network is trained.

[0087] The spectroscopic system 1 and method of the present invention: Obtaining the electromagnetic radiation emitted directly by the liquid metallurgical product 2; the spectral emissivity ε in a given wavelength range λ of the system 1 is calculated from the emitted radiation by the radiative transfer model realized by the processing means of the spectroscopic system 1; est (λ,T est ) and the temperature of metallurgical product 2 T est and - estimating the chemical composition of the metallurgical product 2 by means of a regression model realized by the processing means; This becomes possible.

Claims

1. A device for determining the chemical composition of a liquid metallurgical product (2) emitting electromagnetic radiation, comprising: a collection probe (3) configured to acquire electromagnetic radiation emitted by the metallurgical product (2) in a predetermined wavelength range Δλ; Spectroscopic means (8) connected to the collection probe (3) and configured to generate a spectral signal of the acquired electromagnetic radiation; Using the MCMC method, the observed radiance L ｏｂｓ (λ, T ｅｓｔ ) and calculating the calculated observed radiance L ｏｂｓ (λ, T ｅｓｔ ) to obtain the temperature T of the metallurgical product (2) in a predetermined wavelength range Δλ. ｅｓｔ and spectral emissivity ε ｅｓｔ (λ, T ｅｓｔ ) , At least The processing means calculates the reference radiance L of X samples i (i varies from 1 to X) in a predetermined wavelength range Δλ. ｒｅｆ，ｉ (λ, T ｒｅｆ，ｎ ), and Z reference temperatures T ｒｅｆ，ｎ (n varies from 1 to Z), for each reference radiance: Reference temperature T in a given wavelength range Δλ ｒｅｆ，ｎ The reference spectral emissivity ε related to ｒｅｆ，ｉ (λ, T ｒｅｆ，ｎ ), and Chemical composition of sample i including a database, The processing means estimates the spectral emissivity ε ｅｓｔ (λ, T ｅｓｔ ) and the estimated temperature T ｅｓｔ , the predetermined wavelength range Δλ and the estimated temperature T ｅｓｔ Reference temperature T ｒｅｆ，ｎ and the reference spectral emissivity ε ｒｅｆ，ｉ (λ, T ｒｅｆ，ｎ ) compared to the observed radiance L ｏｂｓ (λ, T ｅｓｔ ) with the best fit reference radiance L ｂｆ (λ, T ｒｅｆ ) to determine and the determined best-fit reference radiance L ｂｆ (λ, T ｒｅｆ ) from the liquid metallurgical product (2),

2. 2. The device (1) according to claim 1, wherein the spectroscopic means (8) comprises a plurality of spectrometers (9, 10, 11), each spectrometer being connected to a collection probe (3) and configured to generate a portion of the spectral signal in a determined portion of the predetermined wavelength range Δλ.

3. 3. The device (1) according to claim 1 or 2, wherein the spectroscopic means (8) are configured to generate a spectral signal in a spectral range between 200 and 12000 nanometers, which corresponds to a predetermined wavelength range Δλ.

4. 4. The device (1) according to any one of claims 1 to 3, wherein the processing means is also configured to correct each part of the spectral signal using at least a calibration constant associated with the spectrometer (9, 10, 11) calculated and taken into account by said processing means.

5. A method for determining the chemical composition of a liquid metallurgical product (2) emitting electromagnetic radiation, using a device (1) according to any one of claims 1 to 4, comprising: i. acquiring (E1) electromagnetic radiation emitted by a metallurgical product (2) in a predetermined wavelength range Δλ; ii. A step (E2) of generating a spectral signal by separating and measuring the spectral components of said electromagnetic radiation emitted by the metallurgical product (2); iii. From said spectral signal, the observed radiance L of the metallurgical product (2) ｏｂｓ (λ, T ｅｓｔ ) and iv. The observed radiance L ｏｂｓ (λ, T ｅｓｔ ) from the temperature T ｅｓｔ and spectral emissivity ε ｅｓｔ (λ, T ｅｓｔ ) and v. The estimated spectral emissivity ε ｅｓｔ (λ, T ｅｓｔ ) and temperature T ｅｓｔ From the database, each reference radiance L ｒｅｆ，ｉ (λ, T ｒｅｆ ) reference spectral emissivity ε ｒｅｆ，ｉ (λ, T ｒｅｆ ) and the reference T ｒｅｆ A step (E5) of comparing with the temperature; vi. Observed radiance L ｏｂｓ (λ, T ｅｓｔ ) with the best fit reference radiance L ｂｆ (λ, T ｒｅｆ ) (E6), vii. A step (E7) of determining the chemical composition of the discharged metallurgical product (2); A method comprising:

6. 6. The method of claim 5, wherein in step iv (E4) the atmospheric attenuation coefficient is also estimated from the observed radiances, and said observed radiances are correlated with said attenuation coefficient.

7. Step iii (E3) The spectral signal is expressed as the ideal blackbody radiance L ＢＢ (λ,T) This ideal radiance L ＢＢ correcting (λ,T) using at least a calibration constant calculated by the processing means and associated with the spectroscopic means (8), said calibration constant being calculated using a calibration lamp of known radiation temperature and known emissivity in the wavelength range.

7. The method according to claim 5 or 6, which is carried out according to

8. Step iv (E4) comprises: determining a vector, the coordinates of which are randomly selected variables representing at least radiation temperature and emissivity in a wavelength range; Based on the variables selected, the expected radiance L ｅｘｐ the sub-step of computing (λ,T); A probabilistic inference model is fitted by comparing the observed radiance with the expected radiance to determine the observed radiance L ｏｂｓ (λ, T ｅｓｔ ) the expected radiance L ｅｘｐ a sub-step of randomly varying the expected variance by randomly varying selected variables until convergence of (λ,T); By using the adapted probabilistic inference model, the radiation temperature T in the wavelength range Δλ of the liquid metallurgical product (2) is calculated. ｅｓｔ and the spectral emissivity ε ｅｓｔ (λ, T ｅｓｔ ) and the sub-step of estimating The method of claim 7, carried out according to

9. 9. The method of claim 8, wherein one of the coordinates of the vector comprises an atmospheric attenuation coefficient, said attenuation coefficient also being estimated by using said adapted probabilistic inference model.

10. 10. The method according to claim 8 or 9, wherein step iv (E4) is performed using a reduced set of wavelengths determined from the wavelength range using a triangular transfer function.

11. A method according to any one of claims 5 to 10, wherein steps v (E5) to vii (E7) are performed by a multi-layer perceptron implemented by processing means.

Images