System and method for determining the chemical composition of a liquid metallurgical product

Abstract

An apparatus for determining the chemical composition of a liquid metallurgical product that emits electromagnetic radiation. The apparatus includes: a collection probe configured to acquire electromagnetic radiation emitted by the metallurgical product within a predetermined wavelength range Δλ; a spectroscopic device connected to the collection probe and configured to generate a spectral signal of the acquired electromagnetic radiation; and a processing device including a database of reference radiance. The invention also relates to a method of using said apparatus.

Description

[0001] This invention relates to the field of metallurgy, and more specifically, to the field of liquid metallurgical products.

[0002] More specifically, the present invention relates to a system and method for determining the chemical composition of a metallurgical product using electromagnetic radiation emitted by the liquid metallurgical product.

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

[0004] A sensor capable of obtaining the chemical composition of a solid slag portion is known from publication WO 2016 / 181185. This sensor comprises a light source suitable for illuminating a slag portion, an optical system suitable for collecting reflected light from the slag portion, and a processing apparatus suitable for obtaining a dataset based on the collected light. The dataset is defined as a matrix containing values ​​representing the intensity of reflected light collected from multiple points. Therefore, a regression algorithm is implemented to estimate the chemical composition of the slag portion based on the intensity and wavelength of the reflected light.

[0005] However, the system and method in publication WO 2016 / 181185 cannot be applied to unknown metallurgical products. Furthermore, the slag portion must be excited by a light source, as only reflected light is collected by the facility.

[0006] Therefore, the object of the present invention is to overcome the disadvantages of the prior art by providing a system and method for determining the chemical composition of any liquid metallurgical product that emits electromagnetic radiation.

[0007] For this purpose, the present invention relates to an apparatus for determining the chemical composition of a liquid metallurgical product that emits electromagnetic radiation, the apparatus comprising at least:

[0008] - A collection probe, configured to acquire electromagnetic radiation emitted by metallurgical products within a predetermined wavelength range Δλ.

[0009] - A spectroscopic device, which is connected to a collection probe and configured to generate a spectral signal of the acquired electromagnetic radiation.

[0010] - A processing device configured to calculate the observed radiance L based on the generated spectral signal. 观察 (λ,T 估计 And is configured to estimate temperature T based on the observed radiance calculated therefrom. 估计 The spectral emissivity ε of metallurgical products within a predetermined wavelength range Δλ 估计 (λ,T 估计 The processing device includes reference radiance L for X samples i within a predetermined wavelength range Δλ. 参考，i(λ,T 参考，n The database and Z reference temperatures T 参考，n i varies from 1 to X, and n varies from 1 to Z. For each reference radiance, it includes:

[0011] Within the predetermined wavelength range Δλ, the reference temperature T 参考，n Associated reference spectral emissivity ε 参考，i (λ,T 参考，n ),as well as

[0012] The chemical composition of sample i

[0013] and configured to estimate the spectral emissivity ε 估计 (λ,T 估计 ) and estimated temperature T 估计 With reference temperature T included in the database 参考，n and within the predetermined wavelength range Δλ and at the estimated temperature T 估计 The reference spectral emissivity ε 参考，i (λ,T 参考，n Compare with the observed radiance L to determine its value. 观察 (λ,T 估计 The best-fit reference radiance L bf (λ,T 参考 ), and based on the determined best-fit reference radiance L bf (λ,T 参考 (2) is used to determine the chemical composition of the liquid metallurgical product (2).

[0014] The device may also include the following optional features, either individually or based on all possible combinations of technologies:

[0015] - The spectroscopic device includes multiple spectrometers, each connected to a collection probe and configured to generate a portion of the spectral signal within a predetermined wavelength range Δλ.

[0016] - The spectroscopic device is configured to generate a spectral signal in the spectral range of 200 nm to 12000 nm, corresponding to a predetermined wavelength range Δλ.

[0017] The processing device is also configured to correct each part of the spectral signal using at least one calibration constant calculated by the processing device and associated with the spectrometer under consideration.

[0018] The present invention also provides a method for determining the chemical composition of a liquid metallurgical product emitting electromagnetic radiation using the apparatus according to the present invention, the method comprising the following steps:

[0019] i. Acquire (E1) electromagnetic radiation emitted by the metallurgical product (2) within a predetermined wavelength range Δλ;

[0020] ii. A spectral signal (E2) is generated by separating and measuring the spectral components of the electromagnetic radiation emitted by the metallurgical product (2);

[0021] iii. The observed radiance L of the metallurgical product (2) is generated based on the spectral signal. 观察 (λ,T 估计 (E3);

[0022] iv. Based on the observed radiance L 观察 (λ,T 估计 Estimate (E4) temperature T 估计 The spectral emissivity ε of the metallurgical product (2) within the predetermined wavelength range 估计 (λ,T 估计 );

[0023] v. The estimated spectral emissivity ε 估计 (λ,T 估计 ) and temperature T 估计 With each reference radiance L from the database 参考，i (λ,T 参考 The reference spectral emissivity ε 参考，i (λ,T 参考 ) and reference T 参考 Temperature comparison (E5);

[0024] vi. Determine and observe the radiance L 观察 (λ,T 估计 The best-fit reference radiance L bf (λ,T 参考 (E6), and

[0025] vii. Determine the chemical composition (E7) of the metallurgical product (2) to be launched.

[0026] The method may also include the following optional features, either individually or based on all possible combinations of technologies:

[0027] - In step iv, an atmospheric attenuation coefficient is also estimated based on the observed radiance, which is related to the attenuation coefficient.

[0028] -Step iii is implemented after the following sub-steps:

[0029] • Convert the spectral signal into ideal blackbody radiance L BB (λ, T);

[0030] • The ideal radiance L is corrected using at least one calibration constant calculated by the processing device and associated with the spectral device. BB (λ, T), the calibration constant is calculated using a calibration lamp with a known emissivity within a known emission temperature and wavelength range.

[0031] Step iv is implemented after the following sub-steps:

[0032] • Determine a vector, the coordinates of which are randomly selected variables that at least represent the emissivity within the emission temperature and wavelength range;

[0033] • Calculate the expected radiance L based on the selected variables. 预期 (λ, T);

[0034] • An inference probability model is fitted by comparing observed radiance with expected radiance, and the expected variance is randomly modified by randomly changing the selected variables until the expected radiance L is reached. 预期 (λ, T) Radiance L towards the observation 观察 (λ,T 估计 )convergence;

[0035] • The emission temperature T is estimated by using this fitted inference probability model. 估计 Emissivity ε of liquid metallurgical products within the wavelength range 估计 (λ,T 估计 ).

[0036] One of the coordinates of the vector includes the atmospheric attenuation coefficient, which is also estimated using a fitted inference probability model.

[0037] Step iv is achieved using a reduced set of wavelengths, which is determined from the wavelength range using a trigonometric transfer function.

[0038] Steps v through vii are performed by the multilayer perceptron implemented by the processing device.

[0039] Other features and advantages of the invention will become apparent from the following description by way of indication rather than limitation, and with reference to the accompanying drawings, in which:

[0040] - Figure 1 A schematic diagram showing the system of the present invention.

[0041] - Figure 2 A graph representing the application of trigonometric functions used to determine a set of considered wavelengths from a defined wavelength range.

[0042] - Figure 3 A diagram illustrating the sequential steps of the method of the present invention.

[0043] First, it should be noted that in the accompanying drawings, the same reference numerals denote the same elements, regardless of the drawings in which they are described or the form of these elements. Similarly, if an element is not specifically referenced in one of the accompanying drawings, its reference can be easily found by referring to the other drawing.

[0044] It should also be noted that the accompanying drawings mainly illustrate one embodiment of the purpose of the present invention, but other embodiments corresponding to the limitations of the present invention may exist.

[0045] The spectroscopic system 1 and method of the present invention are applied to the estimation of the chemical composition of liquid metallurgical products 2, such as slag or liquid steel.

[0046] Thermal elements, such as liquid metals like liquid steel, emit electromagnetic radiation that is related to their respective emission temperatures and chemical compositions. This radiation is emitted in a broad emission spectrum, typically ranging from ultraviolet to far-infrared wavelengths, including infrared and visible wavelengths directly related to the emission temperature that results in the red to white appearance of liquid metallurgical products.

[0047] Physical parameters can be extracted from the measurable characteristics of the electromagnetic radiation. These physical parameters include:

[0048] •T, called temperature, is the temperature of the liquid metallurgical product 2 that emits electromagnetic radiation;

[0049] ε(λ, T), called spectral emissivity or simply emissivity, is the radiative flux of electromagnetic radiation emitted by a surface element at a given temperature T and a given wavelength λ, compared to the radiative flux of an ideal blackbody. Therefore, the value of emissivity is between 0 and 1.

[0050] • L(λ, T), called spectral radiance, is the radiant flux emitted per unit projected area per unit solid angle and per unit wavelength from a given surface. The term "radiance" will be used below.

[0051] When acquiring electromagnetic radiation emitted by the liquid metallurgical product 2, the spectral system 1 of the present invention, which will now be described, is provided to at least estimate the physical parameters mentioned above, and then these estimated physical parameters are used to determine the chemical composition of the liquid metallurgical product 2 under consideration.

[0052] like Figure 1As depicted, the spectral system 1 of the present invention includes at least one collection probe 3 with a broad spectral range, meaning that the collection probe is fully operable for wavelengths from 0.1 micrometers to 15,000 micrometers. The collection probe 3 includes a collection head 4 configured to collect radiation emitted by the thermometallurgical product 2, a collimator (not depicted) generating parallel rays 19, and a focusing device 5 arranged to emit a focused beam 6 through an output end 7 of the probe 3. The output end 7 is connected to an optical transmission device, such as a set of optical fibers forming an optical fiber bundle 12, and the output end 7 is configured to transmit the focused beam 6 through the optical fibers 13 to 17 of the bundle 12.

[0053] System 1 also includes a spectroscopic device 8, which is configured 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 device 8. In other words, the spectroscopic device 8 splits the focused beam 6 into M components m, each component m being related to a specific wavelength λ. m Relevant parameters, such as intensity. The higher the resolution, the larger the value of M. Typically, M exceeds 3000.

[0054] according to Figure 1 The spectroscopic device 8 includes three spectrometers 9, 10, and 11, each configured to analyze the focused beam 6 and generate a specific spectral signal within a defined wavelength range. Each spectrometer 9 to 11 is optically connected to the collection probe 3 via an optical fiber bundle 12.

[0055] The first spectrometer 9 of the spectroscopic device has a wavelength range from 0.2 μm to 1.1 μm. In other words, the first spectrometer 9 is configured to generate a spectral signal based on emitted radiation in the wavelength range from 0.2 μm to 1.1 μm corresponding to ultraviolet and visible radiation.

[0056] The second spectrometer 10 of the spectroscopic device has a wavelength range including 0.9 μm to 2.6 μm. In other words, the second spectrometer 10 is configured to generate a spectral signal based on emitted radiation within the wavelength range of 0.9 μm to 2.6 μm corresponding to near-infrared radiation.

[0057] The third spectrometer 11 of the spectroscopic device 8 has a wavelength range including 2.5 μm to 12 μm. In other words, the third spectrometer 11 is configured to generate a spectral signal based on emitted radiation within the wavelength range of 2.5 μm to 12 μm corresponding to mid-infrared radiation.

[0058] Advantageously, the spectral system 1 also includes a laser device 18 connected to the collection probe 3 via an optical fiber bundle 12. This laser device 18 is typically a Class B laser emitting visible light (e.g., 532 nm green light) and is configured to be pointed at the acquisition surface. The laser device 18 allows for selection of the acquisition site.

[0059] As described above, the collecting probe 3, the three spectrometers 9 to 11 and the laser device 18 are all connected to the fiber bundle 12, which includes four input terminals connected to the laser device 18 and the three spectrometers 9 to 11, and one output terminal connected to the collecting probe 3.

[0060] More precisely, the laser device 18 is connected to the probe 3 via a low-hydroxyl-ion silica fiber 17, the first spectrometer 9 is connected to the probe via another low-hydroxyl-ion silica fiber 13, the second spectrometer 10 is connected to the probe 3 via two low-hydroxyl-ion silica fibers 14, and the third spectrometer 11 is connected to the probe 3 via two polycrystalline infrared fibers 15.

[0061] Of course, all seven optical fibers mentioned above are connected from 16 to the output end of the collecting probe 3.

[0062] Finally, the spectral system 1 includes a processing device configured to generate the observed radiance of the metallurgical product 2 based on the spectral signal. Due to a first algorithm, the processing device is configured to use an inference probability model to determine the expected radiance based on the observed radiance, and then estimate the temperature and emissivity of the metallurgical product 2.

[0063] In addition, due to the second algorithm, the processing device is configured to use a known regression algorithm, advantageously a multilayer perceptron implemented by the processing device, to estimate the chemical composition of the metallurgical product 2 being emitted, such as the contents of silicon dioxide (SiO2), aluminum oxide (Al2O3), iron oxide (II) FeO, iron oxide (III) Fe2O3, calcium oxide (CaO), and magnesium oxide (MgO).

[0064] As will be described below, in order to implement this second algorithm, the processing device includes a reference radiance L. 参考，i (λ,T 参考，n The database contains reference radiance L for each reference radiance. 参考，i (λ,T 参考，n At the defined reference temperature T 参考，n and at wavelength λ with a known spectral emissivity ε 参考，i (λ,T 参考，n The sample i is associated with the reference emissivity (also known as the reference emissivity). Therefore, each reference emissivity value ε 参考，i (λ,T 参考，n (and the reference temperature value T in the database) 参考，nIt is related to the wavelength value λ. i varies from 1 to X, where X is at least equal to 30, and the higher X is, the more different chemical components are included in the database. n varies from 1 to Z, where Z is at least equal to 5.

[0065] In addition, the chemical composition of each sample i in the database is known and correlated with the reference temperature T. 参考，n and the reference emissivity ε within the wavelength range Δλ 参考，i (λ,T 参考，n Related to this. To determine the chemical composition of the liquid metallurgical product 2, the processing device is configured to perform the following operations:

[0066] • The estimated emissivity ε 估计 (λ,T 估计 ) and estimated temperature T 估计 With regard to T 参考，n =T 参考 Each reference radiance L from the database 参考，i (λ,T 参考，n The reference emissivity ε 参考，i (λ,T 参考，n ) and reference temperature T 参考，n Comparison

[0067] • Using the observed radiance L 观察 (λ,T 估计 Determine the optimal fit reference radiance L bf (λ,T 估计 ),as well as

[0068] • Determine the chemical composition of metallurgical product 2 to be launched.

[0069] According to the present invention, a method for estimating temperature T will now be described. 估计 and emissivity ε 估计 (λ,T 估计 ) and a method for determining the chemical composition of a liquid metallurgical product 2 that emits electromagnetic radiation.

[0070] In the first step, the collection probe 3 is pointed toward the acquisition site selected by the user. To accurately select the acquisition site, the laser 18 is energized and the collection probe 3 is moved until the laser 18 is pointed toward the acquisition site.

[0071] In the second step E1, the collecting probe 3 acquires electromagnetic radiation emitted by the liquid metallurgical product 2 at the acquisition site and focuses the acquired radiation into a focused beam 6. This focused beam 6 propagates from the output end 7 of the probe 3 through the fiber bundle 12 to the three spectrometers 9 to 11 of the spectroscopic device 8. Each spectrometer 9 to 11 then generates a spectral signal within its specific wavelength range. The processing device of the spectroscopic system 1 then generates a combined spectral signal E2 within a defined wavelength range from 0.2 μm to 12 μm.

[0072] In the third step E3, the processing device generates the observed radiance L of metallurgical product 2 based on the combined spectral signals. 观察 (λ,T 估计 The observed radiance L 观察 (λ,T 估计 This is generated after the sub-steps described below.

[0073] In the first sub-step, the processing device converts the combined spectral signal into ideal blackbody radiance L using a polynomial calibration function measured by the spectral system on the blackbody furnace. BB (λ,T 估计 The polynomial function follows the following formula:

[0074]

[0075] Where h is Planck's constant, c is the speed of light, and k is the speed of light. B It is the Boltzmann constant.

[0076] In the second sub-step, the ideal radiance L is corrected using a constant K directly related to the spectral device 8. BB (λ,T 估计 Since spectrometers 9 to 11 of spectroscopic device 8 do not measure the same region, and due to inaccuracies in the measurement results caused by the instrumentation of spectroscopic device 8 (collection probe 3, fiber bundle 12, collimator 5), a correction constant K must be applied to each spectrometer S. s Therefore, the constant K is 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.

[0077] The initial values ​​of these constants K1 to K3 are determined using a spectral system 1 on a calibration lamp with a known emission temperature and a known emissivity within a wavelength range.

[0078] The ideal radiance L is corrected by the calculated constant K from the processing device. BB (λ,T 估计 ) is the observed radiance L 观察 (λ,T 估计 ).

[0079] In the fourth step E4, the processing device implements the first algorithm to estimate the temperature T of the liquid metallurgical product 2. 估计 and the emissivity ε within the wavelength range λ 估计 (λ,T 估计 ).

[0080] As described above, radiance is known to be a function of several physical parameters, including the emission temperature of metallurgical product 2 and the emissivity within a wavelength range. Therefore, the first algorithm implements a radiative transfer model aimed at extracting these unknown parameters so that they can be correlated with the observed radiance L. 观察 (λ,T 估计 Fitting.

[0081] The processing device then uses, for example, the Metropolis-Hasting algorithm, also known as a Metropolis-Hasting random walk, following the Markov chain Monte Carlo method or the MCMC method to calculate the expected radiance L. 预期 (λ,T 估计 The expected radiance can be described by the following formula:

[0082]

[0083] in γ is the atmospheric attenuation coefficient caused by water vapor and carbon dioxide. x is related to the concentration of the gas considered, γ is related to the known attenuation coefficient of the gas considered, and d is the distance between the radiating element and spectrometers 9 and 11. Additionally, f is a transfer function modeling the reflectivity attenuation, fiber attenuation, and nonlinear sensitivity of spectrometers 9 and 11. This transfer function f is calculated using a calibrated blackbody furnace with spectral system 1.

[0084] The MCMC method allows for the estimation of unknown parameters, namely emission temperature, emissivity within the wavelength λ range, and concentrations of water vapor and carbon dioxide, by implementing a Bayesian inference model and comparing the calculated expected radiance with the observed radiance. Furthermore, the calibration correction coefficients K1 to K3 are fine-tuned due to the Metropolis-Hasting algorithm.

[0085] Bayesian inference models use the following formula to construct a probability distribution for finding the values ​​of unknown parameters:

[0086]

[0087] Where x represents the observed radiance L 观察 (λ,T 估计 ), where θ is a vector whose coordinates are unknown parameters:

[0088] The posterior probability distribution P(θ|x) is given L 观察 (λ,T 估计 The probability of finding an unknown parameter value each time. This is the probability that the processing device needs to calculate in order to estimate the unknown parameter.

[0089] The likelihood P(x|θ) is the observed radiance L given a set of parameters. 观察 (λ,T 估计 The distribution pattern. The calculation of this probability will be explained later.

[0090] The prior distribution P(θ) is the prior knowledge used to enable the processor to compute the Metropolis-Hasting algorithm. In other words, as will be explained later, the parameter set θ0 is randomly chosen at the start of the computation.

[0091] P(x) is evidence that the parameters are derived from the radiative transfer model. This quantity is uncomputable in many cases. However, P(θ|x) can be estimated using MCMC methods, especially the Metropolis-Hasting algorithm, simply by defining the likelihood P(x|θ) and the prior P(θ).

[0092] The detailed implementation of the fourth step E4 of the method of the present invention will now be described.

[0093] As specified above, emissivity depends on wavelength, meaning that each wavelength is associated with emissivity. Because the resolution of the spectroscopic device 8 allows for the generation of M discrete spectral values, each associated with a unique wavelength λ. m Therefore, the M emissivity values ​​can be determined by the processing device implementing the method of the present invention.

[0094] However, since M exceeds 3000, the processing device cannot estimate M emissivity values ​​based on M wavelengths due to computational limitations. To address this issue, the selection of the reduced N wavelength groups must first be implemented by the processing device using membership trigonometric functions, such as those expressed below:

[0095]

[0096] Where, λ Cm is the center wavelength, and D is the distance between two consecutive wavelengths.

[0097] The processing device associates M wavelengths with trigonometric functions to calculate N sets of wavelengths. N is much smaller than M, and the value of N is typically around 4 to 10. Because the N sets are calculated using fuzzy logic, these sets are often referred to as fuzzy sets.

[0098] In addition, wavelength λm Specific emissivity ε m The fuzzy emissivity value ε is calculated by the processing device. m The weighted value, for each fuzzy emissivity value ε m The ambiguity is defined at the center of the considered group of fuzzy wavelengths, thus conforming to the following equation:

[0099]

[0100] For example, and as in which N=8 Figure 2 The emissivity ε described in the text m =ε(λ) m The result is a combination of 33% of the fuzzy emissivity ε4 = ε(λ4) and 67% of the fuzzy emissivity ε5 = ε(λ5). Consequently, only N emissivity values ​​are directly calculated by the processing device, meaning that for each emissivity ε from m = 1 to N... m Other emissivity values ​​within the wavelength range are calculated based on one or more ambiguous emissivity values. Therefore, when using this trigonometric function or any other similar function that allows for the reduction of the wavelength group, λ can represent the wavelength or wavelength range indiscriminately.

[0101] Once the reduced wavelength group is selected, the processor using the Metropolis-Hasting algorithm randomly generates a vector θ0, whose coordinates are unknown parameters randomly chosen by the processor. The processor then uses a uniform distribution function and a normal distribution function to compute the prior distribution P(θ0) based on the randomly chosen parameters.

[0102] Follow Apply the emission temperature T to the uniform distribution function, where T 最小 =400℃ and T 最大 =1500℃.

[0103] Follow Applying the spectral emissivity ε(λ) to the uniform distribution function, where ε 最小 =0 and ε 最大 =1.

[0104] Follow Xiang Yi Applying the normal distribution function centered at each correction coefficient K s ,in It is the initial calibration constant calculated using calibration lamps for spectrometers S from 9 to 11, and σ 2 =0.001 2 , σ 2 It is the variance of the distribution.

[0105] respectively and The normal distribution function centered around and where and σ 2 = 50 2 , and where and σ 2 = 500 2 .

[0106] Using the calculation results of each of the above distributions, the processing device determines the prior distribution P(θ0).

[0107] In addition, once the coordinates of the vector θ0 are randomly determined, the processing device uses the unknown parameter values of θ0 and calculates the first value of the expected variance L 预期 (θ0).

[0108] Therefore, the processing device uses the normal distribution function centered around L 预期 (θ0) to calculate the likelihood P(L 观察 |θ0), where σ 2 is a half-Cauchy distribution σ ∼ HalfCauchy(β), and its formula is: β is a fixed parameter, and its value is chosen to be 10.

[0109] The processing device randomly generates a vector θ1 that follows a normal distribution centered around θ0 . Therefore, a new set of unknown parameters is randomly selected, and the processing device calculates L 预期 (θ1), the likelihood P(L 观察 |θ1), and the prior P(θ'1).

[0110] Then, the number r1 is calculated according to the following formula:

[0111]

[0112] The standard uniform distribution is also used to calculate the random number r'. The value of r' is between 0 and 1.

[0113] The processing device compares r1 and r', and if r1 > r', it keeps the vector θ1 and calculates the vector θ2 according to the normal distribution centered around θ1 . If r1 < r', the vector θ1 is rejected, and as long as r1 < r', a new vector θ'1 is randomly calculated according to the normal distribution .

[0114] The vector θ y is calculated step by step in this way using until y = y最大 For example, y 最大 =2000. y 最大 The larger the value, the more accurate the estimate, but the longer the calculation time.

[0115] The Metropolis-Hasting random walk implemented by the processing device reduces the expected variance L 预期 (λ, T) is estimated with a set of parameters toward the observed radiance L. 观察 (λ,T 估计 The convergence occurs when the estimated set of parameters forms the final vector. The coordinates.

[0116] Given the observed radiance L 观察 (λ,T 估计 The MCMC method was used to estimate the emission temperature T of the liquid metallurgical product 2. 估计 The value and spectral emissivity ε 估计 (λ,T 估计 The values ​​of ) were also estimated. In addition, the internal correction parameters K1 to K3 of spectrometers 9, 10, and 11, as well as the atmospheric water vapor content, were estimated. and carbon dioxide The concentrations of both.

[0117] The processing device executes the second algorithm to implement the fifth step E5, the sixth step E6, and the seventh step E7 of the method of the present invention. In the fifth step E5, the processing device will determine the observed radiance L. 观察 (λ,T 估计 Estimated emissivity ε 估计 (λ,T 估计 ) and temperature T 估计 With regard to T 参考，n =T 参考 Each reference radiance L from the database 参考，i (λ k T 参考，n The reference emissivity ε 参考，i (λ,T 参考，n ) and reference launch temperature T 参考 A comparison is then made. Using the comparison results obtained in step five (E5), the processing device determines the observed radiance L in step six (E6). 观察 (λ,T 估计 The best-fit reference radiance L bf (λ,T 参考 Finally, the processing device will be compared with the best-fit reference radiance L. bf (λ,T 参考 The relevant chemical components belong to the metallurgical product 2 that is being emitted. Therefore, the processing device has determined the chemical composition of the metallurgical product 2 that is being emitted.

[0118] Advantageously, the second algorithm is a known regression algorithm, preferably a multilayer perceptron, which is an artificial neural network comprising a number J interconnected neural layers, the artificial neural network being implemented by a processing device.

[0119] Each digital neuron in a layer of a neural network includes multiple input terminals and output terminals. The input terminals are configured to receive data from the output terminals of neurons in the previous layer, and the output terminals are configured to send output data to neurons in the next layer. More precisely, the input data is processed by the digital neurons using operators to produce output data. These operators are typically combination functions that sum all input data, with each input data being weighted by synaptic weights. Additionally, each neuron may include a non-linear transfer function, which is configured to produce output data if the weighted sum of the input data exceeds a threshold defined by a threshold function. Typically, the threshold function is a sigmoid function, whose formula is...

[0120] Each input to neuron j in a neural layer is influenced by specific synaptic weights. When a multilayer perceptron is programmed, the synaptic weights of the inputs are initially chosen randomly. As will be discussed later, these synaptic weights are adjusted during the training phase of the neural network.

[0121] Artificial neural networks are configured to predict and classify data, thereby performing regression analysis through the input data fed into the neural network—meaning through the input terminals of the neurons in the first layer. The injected data is a vector X with j coordinates. 0 It is processed continuously by the first neural layer and all intermediate neural layers.

[0122] Output data—meaning the data obtained at the output of the neural network—is the vector X representing the injected data. 0 The vector Y of the relevant j coordinates J .

[0123] Vector X 0 The first layer and all intermediate layers of the neural network each contain j coordinates, and each neuron has j neurons, each with j inputs, each influenced by specific synaptic weights. The final layer contains j neurons because the output vector Y... J It includes j coordinates.

[0124] To generalize the data regression implemented by the 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 the neural network is based on the input vector X that has been processed by the transfer functions of the neurons in layer j. j Generate output vector YJ X j The coordinates of the layer and the output vector Y of layer j-1 j -1 Compared to the input vector X at the same coordinates, but at layer j+1 j-1 The coordinates of layer j and the output vector Y j Compared to being at the same coordinates.

[0125] At the output of the neural network, the output vector Y J The output layer of the neural network processes the input vector X J =Y J-1 This is generated through processing. Therefore, data regression is achieved by data propagation in successive layers of a neural network.

[0126] Regarding steps five E5 to seven E7 of the method of the present invention, the input vector X 0 The coordinates are the estimated launch temperature T. 估计 and each center wavelength λ of N fuzzy groups m Estimated emissivity ε 估计，m (λ m T 估计 The input vector X has different values. 0 The neural network processes the data layer by layer to produce an output vector Y whose coordinates represent the proportions of different chemical compositions of metallurgical product 2. M In other words, the processing devices in steps five through seven are based on their estimated spectral emissivity ε 估计，m (λ m T 估计 ) and its estimated temperature T 估计 To determine the chemical composition of metallurgical product 2. However, such a process is only possible if the weights are adjusted correctly.

[0127] The key challenge of this method is to finely adjust the synaptic weights to achieve good data classification and / or prediction.

[0128] Reference radiance L 参考，i ,k(λ,T) 参考，n The database is obtained by testing at different emission temperatures T. 参考，n The following is a pre-constructed set of samples i with known chemical composition. For each sample, the observed radiance is measured, and the first algorithm described above is applied to the observed radiance to obtain a wavelength-dependent λ. k Several reference emissivity values ​​ε 参考，i，k (λ k T 参考 ).

[0129] Therefore, the processing device constructs a database of samples, each sample including an input vector X' related to the reference emissivity value at different reference emission temperatures. 0 and with X' 0 The output vector Y' is related to and associated with the known chemical composition. J In other words, the emissivity value at a specific emission temperature is directly related to the chemical composition of metallurgical product 2.

[0130] To train the neural network, implement the backpropagation function. (This is done in conjunction with the known output vector Y'.) J The relevant input vector X' 0 It is input into the neural network. X' 0 The neural network processes the data and obtains the calculated output vector Y. J Then, Y is calculated using a known loss function. J With Y' J The comparison is performed to determine the error value. Then, the processing device compares this error value with a predetermined threshold.

[0131] If the error value is greater than a threshold, then that error value, or the error gradient, is backpropagated in the neural network to slightly adapt to the weights of each neuron's input. Then, according to X' 0 Calculate the new output vector Y J And again, the loss function is used to adjust the new output vector Y. J With Y' J The comparisons are made, 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. Once the error value is less than the threshold, the neural network is trained.

[0132] The spectral system 1 and method of the present invention allow:

[0133] • To obtain electromagnetic radiation directly emitted by liquid metallurgical product 2;

[0134] • Based on the radiative transfer model implemented by the processing device of spectral system 1, the spectral emissivity ε within a predetermined wavelength range λ of system 1 is estimated according to the emitted radiance. 估计 (λ,T 估计 The temperature T of metallurgical product 2 估计 ,as well as

[0135] • The chemical composition of metallurgical product 2 is estimated by means of a regression model implemented by the processing device.

Claims

1. An apparatus for determining the chemical composition of a liquid metallurgical product (2) that emits electromagnetic radiation, said apparatus comprising at least: - a collection probe (3) configured to acquire said electromagnetic radiation emitted by said metallurgical product (2) in a predetermined wavelength range within - A spectroscopic device (8), which is connected to the collection probe (3) and configured to generate a spectral signal of the acquired electromagnetic radiation. - A processing device configured to calculate the observed radiance based on the generated spectral signal. And is configured to use an inference probability model to calculate the observed radiance. Determine the expected radiance to estimate the temperature. and the predetermined wavelength range of the metallurgical product (2) Spectral emissivity within , The processing device includes components within the predetermined wavelength range. Internal and Z reference temperatures Reference radiance of X samples i The database, where i varies from 1 to X and n varies from 1 to Z, includes the following for each reference radiance: - said predetermined wavelength range a reference spectral emittance associated with a reference temperature within said predetermined wavelength range and - The chemical composition of sample i, The processing device is configured to process the estimated spectral emissivity. and estimated temperature With reference temperatures included in the database and the predetermined wavelength range Inside and at the estimated temperature Reference spectral emissivity Compare and determine the observed radiance. Best-fit reference radiance , and based on the determined best-fit reference radiance To determine the chemical composition of the liquid metallurgical product (2).

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

3. The apparatus (1) according to claim 1 or claim 2, wherein, The optical spectrum device (8) is configured to generate a spectral signal corresponding to a spectral range of 200 nanometers to 12000 nanometers. between 200 nanometers and 12000 nanometers.

4. The apparatus (1) according to claim 1 or claim 2, wherein, The processing device is also configured to correct each portion of the spectral signal using at least one calibration constant calculated by the processing device and associated with the spectrometers under consideration (9, 10, 11).

5. The apparatus (1) according to claim 3, wherein, The processing device is also configured to correct each portion of the spectral signal using at least one calibration constant calculated by the processing device and associated with the spectrometers under consideration (9, 10, 11).

6. A method for determining the chemical composition of a liquid metallurgical product (2) emitting electromagnetic radiation using the apparatus (1) according to any one of claims 1 to 5, the method comprising the steps of: viii. Obtain (E1) the emission of the metallurgical product (2) within a predetermined wavelength range. The electromagnetic radiation within; ix. A spectral signal (E2) is generated by separating and measuring the spectral components of the electromagnetic radiation emitted by the metallurgical product (2); x. generating, from the spectral signal, a radiance luminance of the observation of the metallurgical product (2) (E3); xi. Based on the observed radiance This is achieved by implementing a radiative transfer model to extract unknown parameters, so that these unknown parameters can be correlated with the observed radiance. Fitting to estimate (E4) temperature and the spectral emissivity of the metallurgical product (2) within the predetermined wavelength range The unknown parameters include the emission temperature and emissivity within the wavelength range of the metallurgical product (2); xii. The estimated spectral emissivity and temperature With each reference radiance from the database Reference spectral emissivity and reference Temperature comparison (E5); xiii. determining a best fit reference radiation intensity (E6) to the observed radiation intensity (E6), and xiv. Determine the chemical composition (E7) of the metallurgical product (2) to be launched.

7. The method of claim 6, wherein, In step iv (E4), the atmospheric attenuation coefficient is also estimated based on the observed radiance, which is related to the attenuation coefficient.

8. The method of claim 6 or claim 7, wherein, Step iii (E3) is implemented after the following sub-steps: Converting the spectral signal into ideal blackbody radiation luminance ; The ideal radiance is corrected using at least one calibration constant calculated by the processing device and associated with the spectral device (8). The calibration constant is calculated using a calibration lamp with a known emission temperature and a known emissivity within the wavelength range.

9. The method of claim 6 or claim 7, wherein, Step iv (E4) is implemented after the following sub-steps: A vector is determined, wherein the coordinates of the vector are randomly selected variables that at least represent the emission temperature and the emissivity within the wavelength range; calculating an expected radiance based on the selected variable ; fitting an inference probability model by comparing the observed radiance and the expected radiance to randomly modify the expected variance by randomly modifying selected variables until the expected radiance towards the observed radiance converge by using this fitted inference probability model to estimate the emission temperature and the spectral emittance of the metallurgical product (2) in the wavelength range of the liquid state .

10. The method of claim 8, wherein, Step iv (E4) is implemented after the following sub-steps: A vector is determined, wherein the coordinates of the vector are randomly selected variables that at least represent the emission temperature and the emissivity within the wavelength range; calculating an expected radiance based on the selected variable ; An inference probability model is fitted by comparing observed radiance with expected radiance, and the expected variance is randomly modified by randomly changing selected variables until the expected radiance is reached. Radiance towards the observation convergence; by using this fitted inference probability model to estimate the emission temperature and the spectral emittance of the metallurgical product (2) in the wavelength range of the liquid state .

11. The method of claim 9, wherein, One of the coordinates of the vector includes the atmospheric attenuation coefficient, which is also estimated using a fitted inference probability model.

12. The method of claim 10, wherein, One of the coordinates of the vector includes the atmospheric attenuation coefficient, which is also estimated using a fitted inference probability model.

13. The method according to claim 9, wherein, Step iv (E4) is implemented using a reduced set of wavelengths, which is determined from the wavelength range using a trigonometric transfer function.

14. The method according to any one of claims 10 to 12, wherein, Step iv (E4) is implemented using a reduced set of wavelengths, which is determined from the wavelength range using a trigonometric transfer function.

15. The method of any one of claims 6, 7, 10-12, wherein, Steps v (E5) to vii (E7) are performed by the multilayer sensor implemented by the processing device.

16. The method of claim 8, wherein, Steps v (E5) to vii (E7) are performed by the multilayer sensor implemented by the processing device.

17. The method of claim 9, wherein, Steps v (E5) to vii (E7) are performed by the multilayer sensor implemented by the processing device.

18. The method of claim 13, wherein, Steps v (E5) to vii (E7) are performed by the multilayer sensor implemented by the processing device.

19. The method of claim 14, wherein, Steps v (E5) to vii (E7) are performed by the multilayer sensor implemented by the processing device.

Images