Estimation of element content in molten material at opening of metallurgical vessel

Abstract

The computer (203) estimates the content (273) of a specific chemical element in the molten material (303) obtainable at the opening (113) of the metallurgical vessel (103). The computer (203) receives characteristic-enhanced radiation data (153-P, 153-F) from a pyrometer (143-P) configured to monitor the opening (113), the characteristic-enhanced radiation data representing a thermal radiation 5 from the molten material (303) at the opening (113). The computer (203) operates on a processing module (253), which processes the radiation data (153-P). The module (253) has a regression model to estimate the content (273) of a particular chemical element, and the processing module (253) has been pre-trained.

Description

Technical Field

[0001] Generally, the present disclosure relates to metallurgical vessels or furnaces, and more particularly, the present disclosure relates to computer systems, methods, and computer program products for processing measurement data acquired during charging and during a casting process. Background Art

[0002] The metallurgical industry ("steel industry") is known to use a variety of large containers to contain liquid materials, at least during many stages of production. In short, a metallurgical container is a device used to produce and transport metals. Typically, the container holds the material at a relatively high temperature for several hours to melt the material, initiate a chemical reaction (e.g., convert ore into metal), etc. An opening in the container allows the molten material to flow out of the container (or into the container).

[0003] A typical example for a metallurgical vessel includes a furnace. A furnace has several openings and sections. Again in simple terms, there is an inlet opening at the top of the furnace and a corresponding section to supply material (such as a "raw material inlet", usually not yet molten), and one or more outlet openings at the button (such as a "tap-hole") and a corresponding section (usually called a "casting room").

[0004] A prominent example of such a furnace is the blast furnace used to produce iron, or more precisely "pig iron". The blast furnace has additional openings, such as tuyere. While the chemical reaction takes place, the outlet opening at the lower part of the furnace is kept closed by clay (also called "mud" or similar). The material does not leave the furnace.

[0005] From time to time, operators use drilling machines to remove the clay. This process is known as "letting out" and opening the let-out hole.

[0006] In a subsequent processing step (commonly referred to as "casting" or "pouring"), the molten material (i.e. metal and slag) then flows through tapholes into runners and further to other containers. Casting typically takes several hours (to remove a volume of, for example, 2,500 cubic meters of molten material from the furnace).

[0007] During casting, operators can check samples of the molten material for chemical composition, temperature, appearance, etc. Operators can collect material samples through tools such as immersion lances, for example from runners or from skimmers (mechanical devices that separate metal from slag).

[0008] Sensors are distributed at different locations in the furnace, and they can provide measurement data at virtually any time. The operator uses the measurement data to fine-tune the operation of the blast furnace, which is usually with an optimization goal in mind, such as minimizing energy consumption or minimizing emissions to the environment (such as CO2 emissions). In other words, the data is related to the amount of coal or other fuel in the material, the energy used to heat the furnace, etc.

[0009] Temperature measurements at the taphole (or other location) are particularly important. They are related to the temperature of the molten metal (such as the "hot metal temperature" HMT) and the temperature of the slag.

[0010] Other measurements are made on the chemical composition of a material, such as determining the amount of silicon (Si) in a metal.

[0011] However, there are challenges, including:

[0012] ● It is not possible for the operator to take samples all the time during casting. Taking material samples presents a considerable safety risk.

[0013] ●On the other hand, non-contact measurement tools may not be accurate all the time. Two reasons should be mentioned:

[0014] The taphole may be obscured by smoke or other material, and the materials at the taphole (ie, metal and slag) have different emissivities (effectiveness in emitting energy as thermal radiation) at different wavelengths, which can affect accuracy.

[0015] • Alternative measurement techniques such as Laser Induced Plasma Spectroscopy (LIPS) may create additional safety risks (due to the application of laser and generation of plasma).

[0016] A recent publication by Pauna et al. describes the application of spectrometers during steel processing (“Optical emission spectroscopy as a method to improve the process automation of electric arcfurnaces and ladle furnaces” Science Direct, IFAC Papers On Line 55-2 (2022) 78–83). Summary of the invention

[0017] Embodiments of the present invention are directed to a computer system and a computer-implemented method for estimating the content of a specific chemical element in a molten material available at an opening of a metallurgical vessel.

[0018] The pyrometer is mechanically arranged at the container within the monitoring distance. The pyrometer monitors the opening, and the computer receives radiation data from the pyrometer. The radiation data represents thermal radiation at at least two wavelengths from the material at the opening.

[0019] The computer operates a processing module that processes the radiation data into feature-enhanced radiation data. Feature enhancement improves estimation accuracy. The processing module has a regression model to estimate the content of a specific chemical element (i.e., "content data"), and the processing module has been pre-trained.

[0020] In an embodiment, the pyrometer may measure radiation from the taphole of the blast furnace. To further improve accuracy, additional sensor data may optionally be processed.

[0021] A computer-implemented method is provided for estimating the content of a specific chemical element in a molten material (available at an opening of a metallurgical vessel).

[0022] The computer receives radiation data from a pyrometer configured to monitor the opening, the radiation data representing thermal radiation for at least two wavelengths from the molten material at the opening.

[0023] The computer operates a feature enhancer module to identify features from the radiation data according to a plurality of feature enhancement rules and provide feature data. The radiation data and the feature data are collectively referred to as feature-enhanced radiation data. The computer operates a processing module that processes the feature-enhanced radiation data, wherein the processing module has a regression model to estimate the content of a specific chemical element, wherein the processing module has been pre-trained.

[0024] Optionally, the step of operating the processing module includes operating a module that has been trained using training data collected during the data collection phase, wherein the training data is a combination of historical radiation data and historical content data obtained by measuring samples collected from the molten material at the opening. The training data also includes feature data, the feature data being identified according to a plurality of feature enhancement rules.

[0025] Optionally, when the data collection phase has been completed, the feature data has been identified according to a plurality of feature enhancement rules. Considering time, there are in principle two options: (i) the feature data can be stored, for example, together with the historical radiation data (as de facto historical feature data), or (ii) the feature data can be calculated according to the rules when needed (such as for training a processing module, for operating the processing module).

[0026] Optionally, in a further receiving step, the computer receives sensor data from a sensor configured to monitor the container; and in the step of operating the processing module, the computer processes the feature-enhanced radiation data in combination with the sensor data.

[0027] Optionally, after receiving the sensor data, the computer adapts the sensor data to obtain intermediate data, so that the computer executes the step of operating the processing module using the intermediate data.

[0028] Optionally, the computer performs the adaptation with the aid of a machine learning tool. The tool may be an autoencoder.

[0029] Optionally, the computer performs the additional step of receiving one or more images showing the opening or showing the path between the opening and the pyrometer from a camera. The computer classifies the one or more images to identify phenomena. The phenomenon selectively includes at least one of the following: accumulations, smoke, or reflections located in the path; the degree of alignment of the pyrometer with the opening, and the proportion of metal and slag on the image. Based on the identified phenomenon, the computer classifies the radiation data from the pyrometer as reliable radiation data or unreliable radiation data. As mentioned, the computer processes the radiation data to estimate the content of specific chemical elements, but investigating reliability can be regarded as a "quality assurance measure" that ultimately causes the computer to automatically ignore certain estimates. For unreliable radiation data, feature enhancement is not required.

[0030] Optionally, the molten material is a composition comprising a molten metal or a metal alloy, and wherein the specific chemical element is selected from the group consisting of carbon, silicon, iron and sulfur.

[0031] Optionally, the method steps are applied to a metallurgical vessel, the metallurgical vessel being a blast furnace, wherein the opening comprises a tap hole and a flow channel to carry material away from the blast furnace during casting.

[0032] Optionally, the computer performs the further step of presenting said estimate together with the content data to an operator of the blast furnace.

[0033] In an iterative manner, this method can be used to control the operation of the container.

[0034] A computer program product which, when loaded into a memory of a computer system and executed by at least one processor of the computer system, causes the computer system to perform the steps of a computer-implemented method.

[0035] The computer system includes a plurality of modules that perform the steps of the computer-implemented method.

[0036] A system is provided for estimating the content of a specific chemical element in a molten material available at an opening of a metallurgical vessel.

[0037] A pyrometer is suitable for monitoring an opening and providing radiation data, which represents thermal radiation for at least two wavelengths from a molten material at the opening. An imaging device is suitable for monitoring an opening or a path between an opening and a pyrometer to provide one or more images showing the opening or path. A computer has a feature enhancer module, an estimation module, and an image classification module. The feature enhancer module is suitable for processing radiation data into feature data according to a plurality of feature enhancement rules. As used herein, radiation data and feature data together become (i.e., collectively referred to as) feature enhanced radiation data. The estimation module is suitable for processing feature enhanced radiation data using a regression model to provide an estimate of preliminary content data for a specific chemical element in the molten material. The estimation module has been trained by training data collected during a data collection phase. The collected data is a combination of historical radiation data and historical content data (obtained by measuring a sample collected from the molten material at the opening). The training data also includes feature data, which is identified according to a plurality of feature enhancement rules. The image classification module is suitable for classifying one or more images to identify phenomena at openings or paths, and then providing a reliability classification RC for the radiation data based on the identified phenomena, so that the computer outputs the preliminary content data as content data only when the reliability classification RC meets predefined rules.

[0038] Optionally, the phenomenon selectively includes at least one of: accumulation, smoke, or reflections located in the path; the degree of alignment of the pyrometer with the opening; and the proportion of metal and slag on the image. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Embodiments of the present invention will now be described in detail with reference to the accompanying drawings, in which:

[0040] Figure 1 A metallurgical vessel is shown in a very simplified side view and an operator is shown using the content data to control the operation of the vessel;

[0041] Figure 2 An example of a blast furnace as a vessel during the data collection phase is shown, the blast furnace being equipped with a pyrometer;

[0042] Figure 3A and Figure 3B A computer having a processing module during a training phase is shown, wherein Figure 3B A feature enhancer module is also shown;

[0043] Figure 4 The blast furnace is shown during the estimation phase, the blast furnace still being equipped with pyrometers;

[0044] Figure 5A and Figure 5B A computer-implemented method is shown for estimating the content of a specific chemical element in a molten material obtained at an opening of a metallurgical vessel, wherein: Figure 5A An overview of the data collection phase, the training phase, and the estimation phase is also shown, and wherein, Figure 5B Optional steps are also shown;

[0045] Figure 6 A container associated with a pyrometer and camera device is shown to symbolize the sensors providing data to the computer;

[0046] Fig. 7A and Figure 7B shows a time series with radiation data and sensor data, where Fig. 7A shows radiation data and sensor data, and wherein Figure 7B It is shown that the feature enhancement rules are applied to further obtain feature data;

[0047] Figure 8 An image from a camera device is shown, wherein metal and slag are differentiated by classification, so that the classification also provides sensor data;

[0048] Fig. 9 An image with phenomena from a camera device is shown;

[0049] Fig.10 shows an overview of a blast furnace and a computer; and

[0050] Fig.11 A general purpose computer is shown. DETAILED DESCRIPTION

[0051] Writing conventions

[0052] The description sometimes refers to a human operator playing different roles, such as the role of a container operator, the role of a sample operator (or more precisely, the operator who collects the sample), etc. These roles can be played by the same person, or by different people.

[0053] The specification occasionally refers to time series with data. The symbol {} represents a univariate time series, and the symbol {{}} represents a multivariate time series. For convenience, a more detailed description of time series is included at the end of this specification. Fig. 7A and Figure 7B To give.

[0054] As used herein, the notation "computer" (in the singular, with or without a reference number) represents a computing function or the functionality of a computer-implemented module (such as a processing module). The computing function may be distributed to different physical computers.

[0055] The figures also show a computer program or a computer program product. The computer program product, when loaded into the memory of a computer and executed by at least one processor of the computer, causes the computer to perform the steps of the computer-implemented method. In other words, the program provides instructions for a computer-implemented module.

[0056] From different perspectives, the drawings illustrate modules of a computer system, which include a plurality of computer-implemented functional modules that, when executed by the computer system, perform the steps of a computer-implemented method.

[0057] Although the description focuses on computers performing computer-implemented methods, the computers and methods are sometimes placed in the context of industrial systems because the metallurgical vessel is a system component.

[0058] The description will refer to techniques known in the art, among which:

[0059] The controller may implement a control loop in an industrial system. The specification thus explains a use case where a computer provides a result (such as content data) when executing the method. The control loop does not have to be implemented by the controller, in some cases it may be sufficient if a container operator views the content data and interacts with the container accordingly.

[0060] ● Assisted Machine Learning (ML) tools provide intermediate data to computers by utilizing commercially available technologies, such as classifying images to obtain classification data as intermediate data.

[0061] Therefore, implementation details for the controller and for the auxiliary ML tools are available to those skilled in the art.

[0062] In the context of this patent application, industrial systems (or industrial machines) are not considered to be computer-implemented modules (since they do not perform method steps).

[0063] The specification writes "sampling interval" as the time interval at which data becomes available in digital form as used in signal processing, but "time point at which sample is collected" is used for the time point when a physical sample is collected by a sampling operator.

[0064] The term "element" refers to a chemical element.

[0065] Overview

[0066] Figure 1The metallurgical vessel 100 is shown in a very simplified (and almost symbolic) side view. Figure 1 Also shown is a container operator 190 ("operator") who uses the content data 270 / 370 to control the operation of the container 100. The container 100 is part of an industrial system having sensors, actuators, computers, data communications equipment, etc., and having many other components or equipment. For simplicity, such system details are omitted from the illustration.

[0067] As shown on the left side by part A of the figure, the container 100 has an opening 110 for the molten material 300 to exit the container 100. Part B shows a slight variation, where the container 100' has an opening 110' for receiving the molten material 300.

[0068] The opening 110 may be further divided into a hole 120 and a channel 130 (shown only for portion A), and the flow of the molten material 300 from left to right is shown.

[0069] The molten material 300 may be further differentiated into metal 350 and slag 360, shown by solid and dashed lines, respectively. This differentiation may be more relevant for the vessel 100 in section A than for the vessel 100 in section B.

[0070] The functions of the containers 100 ′ and 100 may be combined, for example into a container for transporting materials within a factory.

[0071] The operator (of the container) 190 needs to know the content of certain chemical elements in the material 300 (finally differentiated into metal 350 and slag 360). The content is represented by content data 270 / 370, which is here symbolized in the form of a pie chart (not to scale).

[0072] The figure uses reference numeral 370 for content data obtained by fairly conventional methods, such as collecting samples and subsequent laboratory analysis, and the figure uses reference numeral 270 for content data obtained by computer-implemented methods.

[0073] The content of the elements in the metal 350 may be given in percentages for elements such as carbon C, silicon Si, iron Fe and sulfur S, among others. The content of the slag 360 may be given for other chemical elements. The operator 190 uses the content data 270 / 370 to control the operation of the vessel 100 (e.g., by changing the supply material, etc., or changing other operating parameters).

[0074] Control is not limited to the vessel 100, and the content data 270 / 370 can be used to control other equipment that further processes the material. For example, molten material 300 (e.g., iron from a blast furnace) from the vessel 100 is typically directed to other vessels (e.g., vessel 100') that further process the material (e.g., to produce steel in a steel mill).

[0075] In other words, the content data 270 / 370 is relevant for controlling the container 100 (providing the material 300), but it is also relevant for controlling further processing steps (by controlling further equipment, such as the container 100').

[0076] In the following, the specification explains a computer-implemented method for estimating content data 270 .

[0077] The description is detailed by using the example that the vessel 100 is a blast furnace. The skilled person is very familiar with such equipment and knows various implementations of tapholes (i.e., the example for holes 120) and runners (i.e., the example for guide grooves 130). Real-world furnaces may have more than one taphole, more than one runner, etc. Additional implementations typically include a so-called skimmer (for separating the metal 350 from the slag 360). The tapholes and runners transport the molten material 300 (i.e., metal 350 / slag 360) out of the blast furnace during casting.

[0078] The method - although computer-implemented - also uses data obtained using traditional tools (ref. Figure 2 ), and use phases to distinguish. The term "phase" is used here in the sense of "time interval".

[0079] Although Figure 1 The term is referred to by calling it **0, but other diagrams distinguish terms by phase (refer to Figure 5A To further summarize):

[0080] ● In the data collection phase **1, the industrial system (using container 101) collects data to be used for training, the details of which are outlined in Figure 2 middle.

[0081] In training phase **2, a computer-implemented tool - shown as processing module 252 - is being trained, the details of which are outlined in Figure 3A and Figure 3B middle.

[0082] ● In the estimation phase**3, the computer performs a method for estimating the content data,

[0083] The details are outlined in Figure 4 (device, computer 203) and Figure 5A and Figure 5B (Method 403).

[0084] In other words, throughout the specification, reference numerals labeled **1 / **2 / **3 represent items that are similar but perform different functions at these stages. Figure 1 The vessel 100 in FIG. 1 may be physically identical to the vessel indicated by 101 or 103 in the other figures. During all stages, the vessel may provide molten material, but in stage **3, the computer estimates the elemental content of the material.

[0085] Likewise, the focus of the description will be on computers (and computer-implemented methods), and it is not important whether the vessel 100 is a blast furnace or other metallurgical vessel.

[0086] challenge

[0087] But in most scenarios, there are technical challenges, and the specification lists only some of them:

[0088] The molten material 300 in the vessel 100 may be dangerous to a human operator 190. The molten material has a temperature on the order of 1,000°C or higher, so the operator 190 must not come into physical contact with the molten material at any time. Protective clothing for the operator may be ineffective. Since the openings 110 (especially the tap hole) may be accidentally opened, personnel should not remain nearby, but they may need to collect samples that are only available from the openings (see Figure 2 and Figure 4 , sample operators 171 and 173).

[0089] Determination of content data 370 typically involves sending a sample of the material to a laboratory (refer to Figure 2 ),

[0090] However, the results containing the content data arrive after a certain delay. This delay makes the control of the container more difficult (see Figure 2 Delay in the laboratory 181).

[0091] ● Sample collection is limited to once or twice during a single casting process (taking several hours),

[0092] Therefore continuous monitoring is not possible. However, continuous monitoring is preferred for real-time control of the operation of the container 100 (and other equipment) (see Figure 4 δT_COM in is relatively short).

[0093] ●Inspection of the molten material 300 inside the container 100 is complicated for other reasons (e.g., the opening of the container is usually closed most of the time, material is available in different layers within the container, but the material composition at the opening is what is important).

[0094] Solution Overview

[0095] The operator constantly sends samples of the material to the lab (i.e., the traditional method of measuring the content), but the computer-implemented method supports the operator by estimating the content. The computer uses data from the sensor, not from the physical sample. A pyrometer is an example of a sensor for providing such data. To keep the description as simple as possible, the specification refers to the pyrometer (and the data from the pyrometer) as a sensor, which collectively represents multiple sensors (of different types and with different properties or items to be measured).

[0096] The specification distinguishes between measurement data (ie, data from a sensor) as "radiation data" when the sensor is a pyrometer and "sensor data" when the sensor is a non-pyrometer.

[0097] This "non-pyrometer" sensor is explained and discussed below. Figures 6 to 9 ), where a camera is a typical example. Generally, the specification and drawings use reference numeral 140 for a sensor, and reference numeral 150 for measurement data (from the sensor).

[0098] Reference numerals 140 / 150 may use abbreviations, such as “P” (for “pyrometer”) or “C” (for “camera”).

[0099] Sometimes reference numbers will distinguish between the various stages.

[0100] Figure 2 A blast furnace 101 (i.e., as an example of a vessel) is shown during data collection phase **1, which is equipped with a pyrometer 141-P. In short, the pyrometer measures infrared radiation (IR) of one or more wavelengths λ passing through an optical system. The term "one wavelength" is understood to be the range between a minimum wavelength and a maximum wavelength (i.e., bandwidth). An equivalent term to "wavelength" would be the frequency of the radiation (i.e., calculated using the reciprocal value).

[0101] The pyrometer 141-P is configured to monitor the opening 111, and the pyrometer 141-P provides radiation data 151-P. Radiation can be measured as "spectral energy density", and those skilled in the art are familiar with the underlying principles: More simply, the light emitted from a body depends on the temperature of the body. The emissivity is a function of the body's temperature and its elemental composition, and the emission can be described by a formula with the so-called Planck constant in its components.

[0102] Depending on the implementation, the pyrometer 141-P is located within a monitoring distance from the opening 111, typically a few meters. A minimum monitoring distance will avoid thermal damage to the pyrometer. A suitable monitoring distance allows the radiation to be measured while limiting external influences on the measurement. As the distance increases, interference with the measurement also increases. The pyrometer can be mounted on a holder so that its relative position (relative to the container) remains substantially unchanged ( Fig.10 A pyrometer is shown on a holder support or bracket. The support allows adjustment so that the pyrometer can capture radiation from the opening.

[0103] The radiation data 151-P represents the thermal radiation emitted by the molten material 301 at the opening 111. Since the emission that produces the radiation depends on the wavelength, the pyrometer 141-P makes the radiation data 151-P available for at least two wavelengths. Therefore, the pyrometer 141-P is a multi-wavelength pyrometer. The data is available in a computer readable format for processing by a computer, and the data represents, for example, the intensity of the radiation at specific wavelengths (e.g., λ1 and λ2).

[0104] Those familiar with pyrometers also refer to opening 111 as a "target" because the primary field of view of pyrometer 141-P is directed toward opening 111. Fig. 9 An alternative method of adjusting the pyrometer 141 -P to the opening 111 (so that the pyrometer receives radiation from the entire opening and maintains a constant relative position) is explained.

[0105] The radiation data 151-P (e.g., λ1 and λ2) that becomes available over time will be available in a computer readable format of a multivariate time series {{λ}}1, with univariate time series {λ1}1 and {λ2}1, or in other notations {λi}1, where index i = 1 to N. The number "1" on the right represents data collection phase **1.

[0106] The number of wavelengths (i.e., "first and second") does not necessarily have to be N=2. Figure 2 {{λ}}1 is shown with N=3 lines (ie 3 signals or "channels"). It is contemplated to use a state-of-the-art pyrometer with N=5 channels (ie λ1 to λ5).

[0107] The pyrometer 141-P will provide data values ​​for λi at sampling intervals such as ΔT, which are generally equidistant. For ease of illustration, it may be assumed that ΔT is one second. This is by way of example only, and shorter ΔTs, such as 125 milliseconds, may also be considered.

[0108] General pyrometer

[0109] Commercially available pyrometers are usually used to measure the temperature remotely. For example, a provider is Paul Wurth SA, 32, rue d'Alsace, L-1122 Luxembourg.

[0110] In the context of a blast furnace, US 4,619,533 explains a pyrometer for measuring the bath temperature of the furnace through the tuyere.

[0111] It is worth noting that - for the present method of estimating the content - it is not necessary to calculate the temperature. Nevertheless, the pyrometer (or more precisely: the electronics in the pyrometer) provides temperature data and presents the temperature to the operator (usually on a large display). The technician can directly intercept the radiation data 151-P from the optical sensor of the pyrometer (refer to Figure 4 153-P in the .

[0112] Physical samples

[0113] Figure 2 Further shown is a sample operator 171 taking a physical sample 311 (of the material 301 available at the opening 111).

[0114] The description and drawings are not described in detail here, because collecting samples is known in the art. Those skilled in the art will further distinguish between metal samples (collected from molten metal) and slag samples (collected from slag). The sample operator 171 will collect the sample from the flow channel (refer to Figure 1 130) but not directly from the discharge hole (from Figure 1 The sample will be collected at 110 in the furnace, or after the skimmer (which separates the metal from the slag). Fig.10 An implementation method is shown.

[0115] It is known that the sample is transported to a laboratory (here, laboratory 181). The laboratory delivers content data 371 as measurement results, also shown in a time series. Since the time between consecutive acquisition samples is relatively long (i.e., in hours) compared to ΔT for λi (at the pyrometer), the figure symbolizes the content data 371 by points (rather than lines).

[0116] The laboratory 181 usually provides content data 371 for different chemical elements, but not for all 100+ chemical elements, but for the elements expected to appear in the molten material, such as Fe, C, S, Si, Ti, Mn, P, Cu, Cr, etc. The figure shows a multivariate time series {{content}} and a univariate time series (variables correspond to elements).

[0117] Continuing with the blast furnace example, a single casting process has an average duration of about three hours, but the content of chemical elements may vary during casting. Although the laboratory can provide content data 371 with relatively high accuracy (i.e., with relatively narrow tolerance bands), the data from the laboratory may no longer correspond to actual conditions by the time the data arrives from the laboratory. There is a delay between taking a physical sample (by operator 171 at t1, t2, etc.) and having the result as content data 371.

[0118] By way of example, a multivariate time series with content data 371 would be available in the following way:

[0119] {{Content}}1={{Content-C}1,

[0120] {Content-Si}1,

[0121] {Content-Fe}1,

[0122] {Content-S}1}.

[0123] The number "1" also represents the data collection phase. These abbreviations are the element symbols for carbon, silicon, iron and sulfur, respectively. The figure shows {{content}}1 with the time point (ie the time point when the sample 311 was collected).

[0124] It is convenient to distinguish between metal and slag. In this case, we will have a first multivariate time series {{Content}}1 for metal and a second multivariate time series for slag. For other elements, slag can be distinguished in other ways.

[0125] The graph shows the content data 371 for individual sample collection time points t1, t2, etc. in the form of percentage (such as weight wt%, mass, or volume selected by the technician). These time points represent the time when the physical sample was collected, not the time of delayed availability of the results. For ease of illustration, the points represent the content value of only one element.

[0126] Although the description refers to multivariate time series {{content}}, it is also possible to restrict the analysis to a single element in a univariate time series.

[0127] The time series {{λ}}1 and {{content}}1 (at least for one single variable for one chemical element) form the training dataset, which will be used in the following training stage **2.

[0128] Control loop

[0129] Figure 2 Also shown as an optional arrangement is a controller 501 which may use the content data 371 as an input to implement the control loop mentioned above.

[0130] The controller 501 can use the content data 371 to control the container and control other equipment. However, the delay may conflict with the efficiency of the control. For example, when the content data 371 becomes available (i.e., the time point t1 or t2 when the sample was collected plus the delay), the casting may have been completed (the container will be empty), so the control of the container may no longer be applicable. Only the control of other equipment is still possible.

[0131] Training the processing module

[0132] Figure 3A The computer 202 with the processing module 252 is shown during a training phase **2.

[0133] On the left, the figure repeats the Figure 2 The {{λ}}1 (radiation data 151 -P) of FIG. 2 is combined with {{content}}1 (content data 371 from the laboratory), but with details added. The combination of radiation data 151 -P and content data 371 becomes training data 241 .

[0134] As mentioned, ΔT is the sampling interval (for the radiation data). The content data {{content}}1 is presented in clusters (clustered according to temporal proximity during casting, dashed circles), mainly for the following reasons:

[0135] ● The opening 111 (of the container 101) is not always open, but only temporarily opened (for a blast furnace,

[0136] Only opened during casting). No physical samples 311 were taken at other times (because they were not available).

[0137] • The number of samples collected during a single casting process is limited (to one or two, sometimes three, samples per casting).

[0138] ●Time distance ( Figure 2 The time between t1 and t2, between t2 and t3, etc., or "δT_LAB") is therefore not constant. The delay is not important because t1, t2, etc. can be easily matched to the time points for the radiation data.

[0139] Radiometric data may be provided regardless of the casting, and will be available even if the casting has not yet begun or has been completed. Data not relevant to the casting will be excluded from the training data 241.

[0140] Nevertheless, the content data {{content}}1 can be considered as the ground truth for training. In other words, {{content}}1 can be the reference data for training. Although the processing module 252 is shown by a single line at the output, there may be several outputs, for example, separated by element (reference C, Si, Fe, etc.).

[0141] The data is collected over a time interval called T_historical. As time goes by, this interval will become longer. It is convenient to assume that T_historical is greater than at least one year.

[0142] But it is understood that the interval T_historical will be divided into shorter time frames so that the training data 241 is provided to the processing module 252 in the form of multiple repetitions M. To continue the example of the blast furnace, these shorter time frames "T_cast" can be related to various casting processes. T_cast can start when the drilling machine starts to open the opening hole, and T_cast can end when the furnace becomes empty (in the sense that the molten material stops flowing out of the furnace).

[0143] It is common for each cast to be numbered consecutively (by a "unique cast ID"), and for further details to be recorded, such as the identity of the specific taphole (for furnaces with multiple tapholes), the volume of molten material obtained, the size of the taphole (drill length) and others.

[0144] A technician can review historical data, such as by comparing data from various foundries, and can ignore data from certain foundries, or can ignore data that is not relevant to a foundry. For example, a confidence score can be assigned to a particular foundry. For example, for sample 311 (reference Figure 2 ) and therefore content data 371 is not present will be ignored for use as training data (or excluded from the training data set). In other words, certain sections of T_historical are ignored, so T_cast*M <T_historical。

[0145] The application of confidence scores is only an example for ignoring data. In the following, the specification will explain the concept of reliability classification RC which can also be applied (refer to Figure 6 ).

[0146] Content data 371 can be considered as an "annotation" of radiation data 151-P, but without the need for human expert supervision. Content data 371 varies with each casting.

[0147] In a single training run, the module 252 will therefore receive the training data 241 in the M data sets and obtain the module internal values ​​accordingly. These values ​​are different for different architectures, as will be explained below.

[0148] Using feature-enhanced radiometric data to train the processing module

[0149] Figure 3B The computer 202 with the processing module 252 and the additional feature enhancer module 282 is shown during the training phase **2. In other words, Figure 3B yes Figure 3A A copy of Figure 3B A booster module is also shown.

[0150] Since the rules can be enhanced based on multiple features (ref. Figure 7B ) enhances the radiation data 151-P by features so that the features can be identified by the feature enhancer module 282 (the feature enhancer module 282 is part of the computer 202 or is located outside the computer 202). The radiation data 151-P becomes feature-enhanced radiation data 151-F. In other words, the combination of radiation data and feature data is referred to as "feature-enhanced radiation data". Fig. 7A and Figure 7B to explain the details.

[0151] The feature enhanced radiance data 151-F need not be "historical" since enhancement may be a step prior to training. However, the feature enhanced radiance data 151-F may be stored (and used, for example, as training data) when needed.

[0152] By referencing the reference numeral "283" in parentheses, the figure shows that the feature enhancer 282 / 283 can also operate in stage **3.

[0153] Use of feature enhancer module 282 (and also module 283) is optional, but feature enhancement can improve the estimation accuracy (ie, estimate the content of the chemical element with higher resolution).

[0154] The implementation of feature enhancement rules can include using software libraries, building feature enhancer modules through autoencoders, and other implementations.

[0155] It is also possible to implement this module as a machine learning module that is being trained to select multiple feature enhancement rules from a large number of rules. Automatic encoding is also an option. In other words, the enhancement rules can be obtained through training.

[0156] Model Architecture

[0157] The processing module 252 uses an underlying model. The model can be based on so-called "deep learning", and those skilled in the art are familiar with choosing an appropriate model. The following paper provides useful information: Amal Mahmud, Ammar Mohammed: "ASurvey on Deep Learning for Time-Series Forecasting" January 2021, DOI: 10.1007 / 978-3-030-59338-4_19, in the book "Machine Learning and Big Data Analytics Paradigms: Analysis, Applications and Challenges" (pp. 365-392).

[0158] The paper mentions architectures such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Deep Autoencoders (AE), Restricted Boltzmann Machines (RBM), and Deep Belief Networks (DBN).

[0159] Due to the multivariate time series nature of the data, RNN, LSTM, GRU, or CNN applied to time series are particularly suitable architectures.

[0160] Alternatively, a deep autoencoder can be employed to identify features that can be used to train the architecture explained in papers such as Karl-Philipp Kortmann, Moritz Fehsenfeld, Mark Wielitzka, “Autoencoder-based Representation Learning from Heterogeneous Multivariate Time Series Data of Mechatronic Systems”, arXiv:2104.02784.

[0161] Other traditional machine learning algorithms such as neural networks, XGBoost, random prediction, etc. are also suitable for training modules where the amount and size of historical data is limited for deep learning architectures.

[0162] Model architecture is conveniently represented by "input" and "output" (where data flows from left to right, as in e.g. Figure 1 The estimation stage **3 is explained in Mahmoud et al. in . Module 252 also has its "input" and "output". During training **2, radiation data 151-P is provided to the "input" and content data 371 is provided to the "output".

[0163] The processing module 252 may be trained for different chemical elements.

[0164] Figure 4 The blast furnace 103 equipped with a pyrometer 143-P is shown during an evaluation phase **3.

[0165] Figure 2 To FIG. 3 regarding the device (such as Figure 2 The explanation of the pyrometer 141-P) and time series in FIG. 1 also applies to the training phase and the estimation phase. In other words, and by way of example, Figure 4 The pyrometer 143-P in the embodiment is configured to monitor the opening 113 and provide radiation data 153-P (for at least two wavelengths λ1 and λ2). The data corresponds to the thermal radiation at at least two wavelengths from the molten material 303 at the opening 113, such as Figure 2 As explained in .

[0166] In executing the computer-implemented method, the computer 203 operates a processing module 253 (see step 453), which processes the radiation data 153-P (at least the data available before the time point t_current (i.e., the current time point)). The processing module 253 has a regression model to estimate the content of a specific chemical element (the architecture allows regression).

[0167] As already explained with reference to FIG3 , the processing module 253 has been pre-trained. The time required for the computer to execute the method is negligibly short (cf. Figure 2 , which is short compared to δT_COM and relatively short compared to ΔT), so that the computer can repeat the method over and over. Thus, by receiving updated radiation data, the computer can update the estimate.

[0168] Computer 203 provides content data 273, but without safety risk to the operator (provided that alternative methods, such as LIPS mentioned above, are not applied). Container operator 193 does not participate in providing content data 273. Sample operator 171 (who will collect samples) does not participate in providing content data 273. (Note that each new sample collected and evaluated by the laboratory can be used as training data, see Figure 3A and Figure 3B ).

[0169] The container 101 / 103 and the pyrometer 141-P / 143-P may even be physically identical.

[0170] Figure 4 It is also shown in dashed lines that the computer 203 can optionally receive sensor data 153-X, which will be used below. Figures 6 to 10 to explain the details. Figure 4 T_cast (explained above) is also shown, and a time interval T_back is also introduced.

[0171] To keep the diagram simple, Figure 4 The feature enhancer module 283 is omitted (refer to Figure 3B ), as mentioned, feature enhancement may optionally be used to improve accuracy. In this implementation, processing module 253 will receive feature-enhanced radiation data 153-F (in brackets in the figure) instead of data 153-P.

[0172] Control loop

[0173] Optionally, the controller 503 controls the operation of the vessel 103. In short, the controller 503 can process the content data 273 to control signals for controlling the vessel 103 (i.e., the furnace) or for the industrial system. It does not matter whether the vessel operator 193 uses the control signals to adjust the operating parameters, or whether the controller 503 acts automatically on the vessel 103. In any case, the content data 273 can be regarded as data describing the state of the technical system.

[0174] Figure 4 The controller 503 and the control signal are symbolized by the arrow drawn with a dashed line from the output of the computer 203 to the vessel 103. In other words, a control loop can be established with the pyrometer 143-P and the computer 203. The content data 273 will be used as input data for the loop. The controller 503 will process the content data 273 to control the signal. The vessel 103 (and optionally other parts of the industrial system) will be the recipient of the control signal (for setting operating parameters for the industrial system, etc.).

[0175] Since those skilled in the art can configure the controller 503 accordingly, no further explanation is needed here, and suitable control targets are mentioned only by way of example:

[0176] • In a first example, the control loop aims to minimize energy consumption (of the container 103, or of the entire industrial system).

[0177] • In a second example, the control loop aims at providing the vessel with material having a content optimized for subsequent activities (eg providing pig iron with predefined element content percentages so that the iron is optimized for conversion into steel in a subsequent processing step).

[0178] • In a third example, the control loop is intended to reduce the impact of an industrial system on the environment, such as to reduce the emission of substances (such as CO2) to the environment.

[0179] It is possible to implement multiple control loops and the content data 373 will be used as input data for multiple controllers.

[0180] Discussion about the stage

[0181] Figures 2 to 4 Equipment for separate phases are shown, but these phases are merely a convenient convention to describe the actions to be performed in a sequential order.

[0182] During the data collection phase (ref. Figure 2 ) during the operation container 101 is used in the training phase (reference Figure 3A and Figure 3B ) in which the processing module 252 is trained using the training data obtained in the data collection phase.

[0183] The processing module 252 is trained in the training phase to estimate the content (refer to Figure 4 , to provide content data 273).

[0184] However, in real-world scenarios, data collection for training purposes (such as in the data collection phase) is possible more or less at any time. Figure 4 ) provides content data 273, samples will continue to be manually collected (by sampling operator 173) to obtain additional training data and evaluate the content data 273 (from module 253).

[0185] The operation of the containers during the data collection phase (such as container 101, in the continuation of this phase) and during the estimation phase (such as container 103) is independent of the training (refer to Figure 3A and Figure 3B ) and proceed.

[0186] Sample operator 173 is still actively collecting sample 313, ref Figure 4 Although the time series {{λ}}1 and {{content}}1 can be considered as "historical data", the continuous operation of the container 101 / 103 (even during the estimation phase **3) provides data that can be used for the dual purpose of determining the content (refer to Figure 2 , content data from the laboratory 371), and provide additional training data (reference Figure 3A and Figure 3B ).

[0187] method

[0188] Figure 5A and Figure 5B A computer-implemented method 403 ( Figure 5A The dotted box in the figure) is used to open the metallurgical container (ref. Figure 4 Available molten material (103, 113) Figure 4 In other words, method 403 is related to the estimation stage **3. The estimation of the content can produce content data 273 (also refer to Figure 4 ).

[0189] For convenience, Figure 5A Put method 403 in the context of other stages. The figure illustrates the stages in columns from left to right: Data collection stage **1 (generating training data 241, Figure 3A and Figure 3B ), training phase **2 (generating a trained processing module 252, also refer to Figure 3A and Figure 3B ), and estimation stage **3 (using method 403).

[0190] In more detail, the computer-implemented method 403 includes step 413 "receiving data" and step 453 "operating a processing module".

[0191] The computer 203 receives (step 413 ) radiation data 153 -P (ie, optionally feature-enhanced radiation data 153 -F) from a pyrometer 143 -P configured to monitor the opening 113 , the radiation data representing thermal radiation from the material 303 at the opening 113 for at least two wavelengths λ1 , λ2 .

[0192] The computer 203 operates the processing module 253 (step 453), which processes the radiation data 151-P. The processing module 253 has a regression model to estimate the content of a specific chemical element.

[0193] Although the specification writes "element" in the singular, the method may also be performed on multiple elements (refer to examples C, Si, Fe, and S), such as performed on each element individually (serially or in parallel) or in combination.

[0194] The processing module 253 has been pre-trained, as already explained (refer to Figure 3A and Figure 3B ).

[0195] Optional Steps

[0196] Figure 5B The method 403 may optionally include further steps 423, 433, 443, and 463 as shown in dashed boxes. The optional steps are introduced for optional application of sensor data (details are provided below). Figures 6 to 9 ). Applying optional steps may be advantageous in light of timing and accuracy as will be explained below.

[0197] The figure shows on the left side a method 403 with optional steps (dashed box), the figure also showing on the right side the application of these steps in a simplified sequence of steps.

[0198] The computer 203 optionally receives 423 sensor data from a sensor configured to monitor the container 103 (and / or to monitor the industrial system). Fig. 7A 150-X in , or {{X}} in other symbols).

[0199] In the same dashed box, Figure 5B A further optional step 433 "Adapt the sensor data" is also shown. The computer can assist with ML tools (refer to Figure 6 The adaption step may be implemented by using the adapter 260 in FIG. 1 as an example) or other methods. The adaptation may generate intermediate data. The following will describe an example of adapting image data to identify phenomena and obtain classification data.

[0200] In the "Training-based Applications" column, Figure 5B It is shown that - upon receipt of sensor data (150-X) and (still optionally adapted) - step 453 of operating the processing module may be performed by processing the sensor data.

[0201] Additional steps are step 443 "Evaluate radiation data" (or "Evaluate sensor data") and step 463 "Evaluate content data". Evaluation steps 443 and 463 are optionally performed after the data to be evaluated becomes available. In short, the result of the evaluation can allow the use of sensor data in subsequent steps (such as process 453) or block the sensor data, or the result of the evaluation can "allow" or "block" content data.

[0202] In the Rule-Based Application column, Figure 5BIt is shown that - also in case of receiving sensor data and still optionally adapting - step 443 may follow to evaluate the sensor data. The description will explain examples below.

[0203] The same column "Rule-based application" shows that in step 463, the content data is further evaluated based on the sensor data. Figure 6 Explain examples of this application where preliminary content data will be forwarded (allowed) or blocked.

[0204] Timing and Accuracy

[0205] The computer may repeatedly execute method 403 ( Figure 5A ) to estimate the "new" content (i.e., to provide content data 273) that will be generated each time the method is repeated. Figure 4 : Each point corresponds to a method execution. For ease of explanation, it can also be assumed that the content data 273 become available at time points that are substantially equidistant at an output interval δT_COM (“COM” stands for “computer”).

[0206] Method duplication can be done with (ref. Figure 4 The availability of data at the input of the computer 203 is synchronized. For example, since the radiation data 153-P becomes available every ΔT, the estimate for the content data 273 may be available at approximately the same interval δT_COM as ΔT.

[0207] A person skilled in the art configures a computer to have sufficient processor and memory performance so that the runtime of a single method execution (related to the interval δT_COM) can be neglected.

[0208] In short, whether you run a computer to perform the method or run a lab, you transform input information into output information, but the method does not virtualize the lab, but uses data that is not derived from physical samples. In addition, the "latency" for the method can be ignored.

[0209] In any case, the interval δT_COM (ie, in seconds) will be much shorter than the interval δT_LAB for the content data 371 from the laboratory 181 (cf. Figure 2 , for example, about 20 minutes).

[0210] In view of the optional controller 503 ( Figure 4) can implement a control loop that may not require the method 403 to be repeated at a relatively high rate. As explained above, the content data 273 can be used to control the operation of the container (and the operation of the industrial system to which the container belongs). It can be assumed that the control signal used to modify the operation of the container will need to be at a time interval longer than δT_COM but shorter than δT_LAB.

[0211] But despite this advantage in "speed", which means the availability of content data 273 at δT_COM < δT_LAB and virtually no delay, at least the following accuracy aspects should be considered:

[0212] ●Tolerance band, and

[0213] ●Reliability (in other words, the question of confidence in the data).

[0214] With respect to the tolerance band, the content data 273 may be given as a percentage (or other means), and the container operator 193 will view the data with the knowledge of the tolerance band. By way of simplified example, {content-Si} may be a percentage between 0.5% and 1.5%, and the tolerance band may be assumed to be plus / minus 0.2% (or 0.4% from the lower limit to the upper limit).

[0215] Of course, the trained processing modules 252 / 253 will not be more accurate than their laboratory master data (i.e., Figure 3A and Figure 3B The training data 241 with the content data 371 in the training data 241 is considered more accurate). The tolerance band may be wider (eg, 0.4% mentioned in the content data 273) than that used for laboratory measurements (eg, 0.1% in the content data 371).

[0216] However, content data 273 acquired in a real-time scenario (from a computer) with relatively lower accuracy (i.e., larger tolerance band) may be more suitable for control purposes than content data 373 acquired (from a laboratory) with relatively higher accuracy (but shorter tolerance band).

[0217] Regarding reliability, note the following: Despite the relatively high output rate ( Figure 4 δT_COM ratio Figure 2 δT_LAB in is relatively short), but the radiation data 153-P may not always be accurate.

[0218] A pyrometer can provide radiation data that is still within the usual tolerance range between minimum and maximum values ​​(i.e., values ​​for a particular wavelength), but the data may sometimes not correspond to the radiation. As will be explained, a computer can process the sensor data (reference Fig. 7A150-X) in the , so that the radiation data can be distinguished by reliability classification (or RC for short) (refer to Figure 6 ) RC can be a binary classification, for example, classified as "reliable" or "unreliable", or it can be classified by degree of reliability.

[0219] This specification will explain a method of obtaining RC by processing an image (ie, the additional data is an image).

[0220] Overview of sensors and sensor data

[0221] Method 403 has been explained for processing radiation data 153-P (during training: data 151-P; also as feature-enhanced radiation data 153-F in implementation), but industrial systems have more sensors. Processing data from such sensors can improve the accuracy of content data 273. Figure 5B The steps of receiving sensor data 423 , adapting the sensor data 433 , and applying it (by processing in step 453 and by evaluating in steps 443 , 463 ) have been described.

[0222] Below, the specification gives an overview of the sensor ( Figure 6 ) and gives an overview of the time series used for measurement data (i.e., data from sensors) ( Fig. 7A and Figure 7B ). The specification then discusses how to use the sensor data to evaluate the content data 273 (refer to Figure 5B Optional step 463 in ). Technicians can also use the teachings to evaluate the radiation data / sensor data.

[0223] Note again that using sensor data is optional.

[0224] When the computer processes the radiation data through the processing module (such as Figure 5A The computer may use the sensor data in at least two applications:

[0225] ● (Rule-based applications) Computers can use sensor data (reference Fig. 7A 150-X)

[0226] To evaluate the content data 273 by applying predefined rules (see step 463).

[0227] This scenario will be classified by phenomenon and reliability RC (ref. Figure 6 ) for explanation.

[0228] ● (Training-based approach) The computer can use the sensor data (reference Fig. 7A 150-X in ), refer to Figure 5B In this case, the sensor data is processed like the radiation data, and the data collection and training explained above apply. In other words, the sensor data can be used in parallel with the radiation data (see Fig. 7A ), or sensor data is used in combination with radiation data.

[0229] In both applications, the sensor data can be used to narrow the tolerance band (of content data 273) (because the processing in step 453 uses more data), and the sensor data can be used to improve the reliability of content data 273 (or at least provide a reliability classification because additional evaluation is required).

[0230] Computers can use some sensor data in two applications. This specification uses image data as an example of sensor data for use in both rule-based applications and training-based applications (see Figure 5B ). For specific sensor data, these methods are not exclusive but complementary.

[0231] Since the stage distinction is not important, Figure 6 **0 notation is used up to Figure 7. In other words, these figures apply to stages **1 (collection of data) and **3 (estimation of content).

[0232] Assistive Machine Learning Tools

[0233] Thus far, the specification has explained the processing modules 252 / 253 (during training and estimation) in detail as machine learning tools. In the training-based approach, historical sensor data will become Figure 3A and Figure 3B Part of the training data 241.

[0234] However, for adapting the sensor data in step 433 , further tools, also referred to below as “auxiliary ML tools”, may be used.

[0235] The traditional stage distinction **1, **2, **3 does not necessarily apply to these auxiliary tools. Some of these auxiliary tools use image processing, and they can be trained in other ways. Alternatively, the auxiliary tools can use ready-made models. For example, an auxiliary ML tool for image processing can be trained in a stage that does not necessarily correspond to stage **2.

[0236] Figure 6Shown are a vessel 100 , a pyrometer 140 -P, and a camera 140 -C which provides data to a computer 200 (from left to right). Figure 6 The concept is shown and there may be more than one pyrometer or more than one camera. The camera 140-C providing the image is just an example of a dual purpose image that is used for:

[0237] ● monitoring the phenomenon to obtain a reliability classification RC (e.g., rule-based application to sensor data to evaluate the data, see steps 443, 463), and ● providing the sensor data for processing by module 253 (see Figure 5B In the optional step 423,

[0238] as an example of a training-based application for sensor data).

[0239] The computer 200 has a processing module 250 for processing the radiation data 150-P (optionally, the feature-enhanced radiation data 150-F) into content data 270 (refer to modules 252 / 253, Figure 5A Method 403 in ). This figure is not phase-related and therefore uses the reference numerals in style **0.

[0240] The container 100 has an opening 110 (see Figure 1 ) and - corresponds to Figure 2 and Figure 4 - Pyrometer 140-P is monitoring radiation (dashed arrow). Camera 140-C captures image 150-C of opening 110 (or its surroundings), but ignores certain wavelengths. In other words, pyrometer 140-P is monochromatic, while camera 140-C is polychromatic. Image capture generates sensor data, and the computer receives this sensor data in step 423 (refer to Figure 5B ).

[0241] Image 150-C illustrates phenomenon 160, which, in some cases, may result in the following: Figure 5A ) is unreliable. By adapting (refer to Figure 5B In step 433 of the above, the phenomenon 160 is detected. In short, the presence or absence (or degree) of the phenomenon 160 is given by intermediate data.

[0242] By way of example, the following phenomenon 160 is mentioned:

[0243] There may be dust (or other particles) in the space (or path 110 / 140) between the opening 110 and the pyrometer 140-P. This dust will attenuate the radiation, so that the pyrometer 140-P may detect erroneous radiation. Or worse, the pyrometer 140-P will output radiation data 150-P that the computer 200 cannot process (or will process as incorrect content data 270). This is especially expected during casting (hot material will interact with air). In the metallurgical industry, dust or other particles may be referred to as "accumulation" or "fumes".

[0244] • The pyrometer 140-P may not be aligned with the opening 110, it may potentially only be viewing a portion of the opening. A perfectly aligned pyrometer would appear to receive radiation from the entire opening.

[0245] • The pyrometer 140-P may actually be looking at the slag and not the metal, but the content should be estimated for the metal and not for the slag.

[0246] Phenomenon 160 should be understood in a more abstract sense, and phenomena do not necessarily have to be located in a path.

[0247] If properly detected, this undesirable shadowing and other adverse effects of the phenomenon 160 can be accounted for and ultimately mitigated. If dust, etc. is detected, the content data 270 can be ignored, or if misalignment is detected, the pyrometer 140-P may simply be readjusted. To detect this phenomenon 160, a technician may use an adapter 260 implemented by an auxiliary ML tool.

[0248] But phenomenon 160 may not be limited to dust and misalignment, phenomenon 160 may include other things that can be converted into data.

[0249] Method 403 (reference Figure 5B ) may optionally include the following: In a receiving step, the computer 200 receives one or more images 150-C showing the opening 110 (or showing the path 110 / 140 between the opening 110 and the pyrometer 140-P). In a classification step, the computer 200 classifies the one or more images 150-C to identify the phenomenon 160. This classification may be achieved by an adapter 260, which is implemented by an auxiliary ML tool. In a further classification step, the computer may classify the radiation data 150-P from the pyrometer 140-P as (un)reliable radiation data based on the identified phenomenon 160. In other words, the adapter 260 provides a reliability classification RC. This classification in the adapter may also be achieved by an auxiliary ML tool.

[0250] Figure 5BOptional steps have been shown, more generally with: receiving 423, adapting 433 (to classify the phenomenon), and evaluating 443 (here for evaluating the radiation data).

[0251] Again, it does not matter whether it is phase 1 or phase 3. Reliability classification RC can be related to both data collection phase 1 and estimation phase 3. For example, unreliable data may not be suitable for training.

[0252] although Figure 6 RC and content data 270 are shown as separate outputs from computer 200, but content data 270 is presented to operator 193 (see Figure 4 , for content data 273) can be done conditionally, such as presenting only content data that is "reliable" for RC. In other words, this is only for Figure 5B Another example of an optional evaluation in step 463. The data evaluation steps can be interrelated, for example, evaluating the radiation data as unreliable (step 443) may also mean that the content data is also evaluated as unreliable (step 463, "switch on", not forwarded).

[0253] reliability

[0254] In other words, the reliability classification RC may be used as a criterion for using the content data 270 (for control or other purposes) or ignoring them (eg, negative for a binary RC).

[0255] Based on the RC, the operator 193 can take some corrective actions, such as: taking steps to remove smoke from the path, realigning the pyrometer, ignoring content data for cases with a relatively high fraction of slag (e.g., more slag than metal at the taphole).

[0256] However, processing the image to obtain a reliability classification RC is only an example for using sensor data (i.e., data that is not radiation data). The specification now explains some methods for using sensor data in content estimation (based on the application of training, steps 423 and 453). In view of the optional step 423 introduced above (refer to Figure 5B ), the sensor data (with or without adaptation) is being processed by module 253 (and this module has been trained using historical sensor data).

[0257] system

[0258] Figure 6Also conveniently shown is a system 1000 for estimating the content of a specific chemical element in a molten material available at the opening 110 of the metallurgical vessel 100. An embodiment is presented here with the camera 140-C and the classification module 260 as optional components.

[0259] The pyrometer 140-P is adapted to monitor the opening 110 and provide radiation data 150-P. As already explained above, the radiation data represents the thermal radiation from the molten material at the opening 110 for at least two wavelengths λ1, λ2.

[0260] The camera 140-C is adapted to monitor the opening 110 (or the path 110 / 140 between the opening 110 and the pyrometer 140-P, or both). The camera 140-C provides one or more images 150-C showing the opening (or the path, or both). Figure 5B In particular, computer 200 receives image 150 -C as sensor data, step 423 .

[0261] The computer 200 has an estimation module 250 (i.e., a processing module in the above explanation) and the computer 200 has an image classification module 260 (implemented as an auxiliary ML tool). The estimation module 250 is adapted to process the radiation data 150-P using a regression model to provide an estimate of preliminary content data 270' for a specific chemical element in the molten material. The estimation module 250 has been trained using training data collected during the data collection phase **1, such as using Figure 3A Interpreted: The data collected is a combination of historical radiation data and historical content data obtained by measuring samples taken from the molten material at the opening.

[0262] The image classification module 260 is adapted to classify one or more images 150-C to identify phenomena 160 at the opening 110 (or at the path, or at both the opening and the path), and to subsequently provide a reliability classification RC for the radiation data according to the identified phenomena 160. The computer 200 outputs preliminary content data 270′ as content data 270 only when the reliability classification RC meets predefined rules. Figure 6 The conditional forwarding of data is shown through a switch metaphor under the control of the RC. A RC that meets the rules will turn on the switch, otherwise the switch will be open (to block the preliminary content data 270'). Figure 5B In particular, the computer fits the image (step 433, classification as an example for fitting) and applies the fitting results to evaluate the content data (step 463).

[0263] Phenomenon 160 has been explained, noting that container 100 does not belong to system 1000 .

[0264] Feature enhancement may optionally be used, and Figure 6 The feature enhancer module 280 is shown by the dashed box (enhancing the radiation data 150-P into feature-enhanced radiation data 150-F, as a combination of 150-P and feature 150-Q, see Figure 7B ).

[0265] Radiation and other data

[0266] Fig. 7A A time series with radiation data 150 -P (also referred to as {λ1}, {λ2}, etc.) and with sensor data 150 -X is shown.

[0267] As mentioned above, the pyrometer need not be the only sensor, and by way of example only, the sensor may be one or more cameras (see Figure 6 , for rule-based applications to derive RC from images), one or more microphones, sensors for measuring the viscosity of the material, the speed at which the material moves in a flow channel or the like, and sensors for measuring specific gaseous species near the opening.

[0268] Temperature data can be one of the sensor data, and it is noted that the main purpose of the pyrometer is to measure temperature. Therefore, the pyrometer can also provide temperature data.

[0269] By way of example, {X1}, {X2}, and {X3} are univariate time series that represent measurement data available as scalars. Two or more variables can be combined into a vector. For example, {{X1}, {X2}} would be a double scalar vector.

[0270] By way of example, {X4} represents data obtained from a microphone, which may represent noise intensity (eg, such as sound pressure level), sound frequency, or other sound-related observations (eg, characteristic sounds occurring during casting).

[0271] By way of example, {X5} represents the image from the camera device (reference Figure 6 As explained, the image is used to be adapted, for example, by classification (refer to Figure 5B Step 433 in FIG. 4 is a prominent example of intermediate data.

[0272] Since those skilled in the art are familiar with installing sensors to industrial equipment, the description and drawings focus on adapting data from the sensors without necessarily describing the sensors in too much detail.

[0273] Using sensor data during the training phase **2 and during the estimation phase **3 can improve the accuracy of the estimate (content).

[0274] Computer 200 (having Figure 6 The processing modules 252 / 253, 250 in FIG. 250 receive the radiation data 150-P (ie, {{λ}} having variables corresponding to different wavelengths), as already explained with respect to method 403 (see FIG. 250 ). Figure 5A ).

[0275] Receiving sensor data 150 -X (or {{X}}, shown here as {X1} to {X5}) in step 423 (and processing sensor data 150 -X) may improve the accuracy of content data 273 (such as by making the tolerance band narrower).

[0276] From a different perspective, the computer processes (in step 453) the radiation data 150-P (from the pyrometer), but in this context, such as, for example,

[0277] • the (un)availability of data that may indicate content, with reference to the evaluation of radiation data or sensor data (step 443), which may prevent the data from being processed,

[0278] ● There are phenomena that may interfere with the operation of the pyrometer (refer to Figure 6 and Fig. 9 , phenomenon 160, such as dust or misalignment), where the phenomenon data can be input to the processing module 253 or can be used for evaluation, such as by Figure 6 As explained in the image examples in

[0279] • Possibility of detecting contradictions (opening closed while the computer receives data from the pyrometer).

[0280] The sensor data 150 -X does not have to be acquired directly from the sensor, but can be obtained through adaptation.

[0281] It is also possible to use RC as additional input data. In this case, RC will not only be used for evaluation (refer to steps 443 and 463), but will also be processed in step 453.

[0282] Feature-enhanced radiometric data

[0283] exist Fig. 7A In FIG. 1 , the dotted box points to the feature-enhanced radiation data 150 -F, which is used for an optional implementation, but in order to simplify the description, the following will mainly use Figure 7B Provide explanation.

[0284] image

[0285] Figures 8 to 9 An image 150-C from camera 140-C is shown. Image processing may be applied (via processing modules 252 / 253) or by auxiliary ML tools such as Figure 6 The result of the image processing can be used as sensor data (refer to Fig. 7A ), the result of image processing can be used as a reliability discriminator (ref. Figure 6 and Fig. 9 , evaluation), the result of image processing can be used as a control signal for the pyrometer (ref. Figure 8 , for metal / slag ratio, refer to Fig. 9 , for alignment control), or other purposes.

[0286] When the camera device 140-C is directed to the opening 110 (ie, for the blast furnace example, for the tap hole, flow channel, etc., refer also to Figure 1 ), image 150-C - at least in many cases - shows opening 110 and the path between the opening and the camera device.

[0287] Image processing may be advantageous for another reason: an image (received for one ΔT) typically comprises millions of pixels, whereas the radiation data from a pyrometer will comprise a few scalar values ​​for each wavelength (e.g., two values ​​for λ1, λ2, and optionally for another length such as λ3, λ4, λ5, etc.), but not millions. Sensor data (ref. Fig. 7A ) are also relatively sparse. Therefore, the availability of data volumes will be uneven. Image processing can reduce (i.e., by adaptation, step 433) the dimensionality: a megapixel image becomes a "short" data vector, but this information helps in estimating elemental content. (To put it metaphorically, image data is adapted to play in the same league as pyrometer data).

[0288] Metal and slag share

[0289] Figure 8 An image from a camera is shown where metal and slag are distinguished by classification, again using an auxiliary ML tool trained accordingly. The training does not necessarily have to be performed for a specific vessel 100, but potentially for any image showing metal and slag, typically supervised training.

[0290] Figure 8 A series of images 150-C taken at consecutive time points are shown, similar to Fig. 7A{X5} in . The figure shows that the auxiliary ML tool distinguishes the material 300 into metal 350 (black diamond symbol) and slag 360 (black circle). Image processing allows to distinguish the image area (or area of ​​interest, refer to the dotted line). Image processing converts the image data into intermediate data.

[0291] This individual differentiation of the individual images can serve as a basis for classification (an example of adaptation in step 433), such as explained for the following example:

[0292] • Images can be categorized according to percentages or shares, representing slag and metal. This graph shows slag / metal shares of 50 / 50, 20 / 80, 60 / 40, and 0 / 100 in percentage form.

[0293] ●Images can be classified based on whether there is a "boundary" between the slag and the metal. From left to right,

[0294] There is a single boundary between the slag and the metal (such as "one" and "one" in terms of the data), multiple boundaries, or no boundary at all (in the example of only metal). Information about the number of boundaries is intermediate data {X7}. It can be processed by the processing module 250 (in step 453).

[0295] ● The image sequence can be classified according to the change, for example, the image sequence can be classified according to the change of the slag / metal fraction. The metal fraction increases from the first image to the second image (from 50 to 80), the metal fraction decreases from the second image to the third image (from 80 to 40), and then increases again to the fourth image. In other words, the intermediate data (explained here by taking the fraction as an example) can be obtained by an operation similar to mathematical deviation.

[0296] ● The image sequences can be classified in other ways. For example, the computer can detect the number of regions.

[0297] The first and second images have two regions (one for slag and another for metal), but the third image has multiple slag regions, and the slag in the fourth image has disappeared.

[0298] Changes in regions can also be encoded in intermediate data.

[0299] Classifying images is used for ( Figure 5B , an example of adapting data in optional step 433 of method 403. Depending on the application,

[0300] ●(Training-based application) The intermediate data {X6}, {X7}, etc. can then be processed by the processing module 250 (step 453, in parallel with the radiation data, in combination with the radiation data), ●(Rule-based application) The intermediate data can then be used for evaluation (for example, the rules will provide threshold conditions for the slag / metal fraction).

[0301] The technician can use other image processing techniques, such as obtaining a histogram (color distribution), etc. The color distribution can even be changed by filtering out certain colors. Similarly, image classification is only an example for the adaptation in step 433.

[0302] Image classification can also be part of determining the reliability classification RC. In the case of RC, the computer will apply rules (to make an assessment). For example, a rule can turn the RC into "negative", that is, the preliminary content data will not become content data (refer to Figure 6 270 ′ and 270 ). As an example, for a slag / metal fraction classified as 90% slag (or more), no estimation of the elemental content in the metal can be performed.

[0303] Another phenomenon

[0304] Fig. 9 Image 150 -C is shown in a different variation, having an additional phenomenon.

[0305] It can be assumed that the pyrometer 140-P and the camera 140-C can be adjusted together, so that any repositioning of the pyrometer relative to the opening is also a repositioning of the camera. In this case, the image also shows the field of view of the pyrometer (the technician can take into account parallax, if any). To symbolize the adjustment, the image 150-C is shown with crosshairs (dashed lines).

[0306] In this example, the pyrometer 140 -P would be pointed toward the tap hole 120 (by way of example, but could also be pointed toward the flow channel or other portions).

[0307] In case (A), image processing (by auxiliary ML tools) will result in the class "Best Fit", in case (B) the class "No Fit", and in case (C) the result will be "Dust" (symbolized by the S-shaped line, obscuring the bleed hole). In cases (B) and (C), the bleed hole will not be visible (to the pyrometer), making the measurement unreliable (RC "Negative").

[0308] Again, auxiliary ML tools can be used here as well, it does not matter where they have been trained. In theory, the detection of misaligned objects in images (i.e., a specific classification) can be trained by using "synthetic" images or by images taken from a simulated device.

[0309] It is possible to give the operator an appropriate warning, such as re-adjusting the pyrometer in (B) (the circle will be moved to the crosshairs), or waiting until the dust disappears in (C). This result can be used as a control signal for the processing module 252 / 253: in this case, do not allow the content value to be processed using data for training, etc.

[0310] Return briefly to Figure 7 to Figure 8 , misalignment detection may cause the computer to process additional data (e.g., as binary data best fit / deviation). Modules 252 / 253 will eventually learn that data from a misaligned pyrometer may have to be ignored, but issuing an alarm (causing readjustment) would be appropriate.

[0311] Note that the pyrometer is automatically aligned based on image detection.

[0312] In the case where the pyrometer and camera are not adjusted together (as they may be supported independently), then the best fit may be a "false positive" and the deviation may be a false negative, but this situation can be taken into account by those skilled in the art.

[0313] radiation

[0314] As mentioned above, a pyrometer measures infrared radiation (IR) at one or more wavelengths passing through its optical system. Typically, IR has a wavelength from 700 nanometers (nm) to 1 millimeter (mm), but ranges outside the IR are also contemplated.

[0315] One criterion for selecting an appropriate wavelength is the emissivity of the surface of the material mentioned above (ie, material 300) which is its effectiveness in emitting energy in the form of thermal radiation.

[0316] Therefore, it seems reasonable to use longer wavelengths (more than a millimeter) or shorter wavelengths (i.e., visible light).

[0317] Note that using images (such as image 150-C from camera 140-C, Figure 6 ) implies that the computer (Figure 3 to Figure 4 202, 203) in the figure process data related to the radiation of visible light.

[0318] Autoencoder

[0319] As explained, the steps of adapting sensor data ( Figure 5B 433) can be performed by auxiliary ML tools.

[0320] Another example of implementing such a tool is an autoencoder. Using an autoencoder can be advantageous because the workload of human experts in annotating relevant features can be reduced. Autoencoders are discussed in papers such as the one by Kortmann et al. identified above.

[0321] Example

[0322] Fig.10 An overview of a blast furnace 600 is shown, with a computer 690 making an estimate of the content of a particular chemical element. Thus, the computer may use the image data for the following functions (alternatively or in combination):

[0323] ●As above, use Fig. 7A As explained with respect to system 1000 , computer 690 may use the image data to obtain and apply RC.

[0324] ●As above, use Figure 5B As explained in optional steps 423 and 433 of , the computer 690 can use the image data to provide intermediate data for processing (step 453). For example, the intermediate data can indicate the slag / metal ratio.

[0325] On the left, the figure shows a blast furnace 600 (an example of a metallurgical vessel with an opening) with a tap hole 610 during casting. Material is shown flowing from left to right, from the furnace to multiple runners. Molten material (slag 660 and iron 650) moves from the furnace through the tap holes to the runners. As shown in the center of the figure, the slag and cast iron are mixed in the initial runners, but are quickly separated (by a skimmer) into slag and cast iron (flowing into separate runners). The figure also shows (by vertical arrow 670) the manual sampling performed at the cast iron runner (reference Figure 2 Operator 371 in.

[0326] Similar diagrams with furnaces, tapholes, runners, and skimmers can be found in textbooks, but Fig.10 A pyrometer 640 is added (cf. pyrometers 151 / 153 and processing unit 690, cf. computer 200 executing method 403). The camera device may be mounted on the same support together with the pyrometer. An image 650 is symbolized at the top of an arrow pointing from the pyrometer to the taphole. In other words - assuming that the pyrometer is correctly aligned with the taphole - the image shows slag and iron in different proportions. The image may also show phenomena such as dust etc. It has been explained in detail above how to adapt the image to intermediate data and to use the intermediate data for purposes such as evaluation.

[0327] The pyrometer 640 and, optionally, the camera and other sensors are adapted to operate in such conditions when the radiation delivers relatively high temperatures from the emission aperture to its vicinity. Technicians can provide cooling for the electronics, such as by water cooling, or can provide appropriate shielding. Fig.10 This protection is symbolized by a protective cover mounted to the left side of the pyrometer / camera assembly.

[0328] Time Series

[0329] The specification sometimes refers to a multivariate time series representing a plurality of measurement data. Fig. 7A An example is shown in . In general, {{X}} represents a multivariate time series with i=1 to N variables {Xi}. The index i is the variable index and N is the variable number. {{X}} includes multiple univariate time series. The variables {Xi} are given in the form of univariate time series. In an alternative notation, the univariate time series can be given as a sequence of data samples.

[0330] M is the number of samples in the observation interval WINDOW (i.e., the time length of the time series), and a single sample is identified by the index m. The sampling interval has a duration of WINDOW=ΔT*M. ΔT represents the sampling interval. ΔT can be the same for all i. This is convenient for illustration, but is not actually necessary. Different univariate time series can use different sampling intervals. For example, temperature will be measured every minute, Δt=60 seconds, but radiation will be measured every second, ΔT=1 second (as explained).

[0331] At each time point tm, data may be available, which represents the value for each variable Xi. Since the semantics is not important, it is OK to normalize the values. For example, for a temperature range of 100°C to 1,100°C, the normalized values ​​may be processed as "0" and "1". Technicians may apply pre-processing to filter out infeasible variable values. For example, a defective sensor may sometimes output excessive values ​​that can be ignored.

[0332] The time point tm is given at the end of each sampling interval ΔT. This is just a convenient convention, in other words, tm represents the sampling interval ΔT at the end of tm.

[0333] Since the specification distinguishes multivariate from univariate time series by writing {{}} or {}, the specification sometimes omits "multivariate / univariate" or puts them in brackets respectively.

[0334] Radiation data that conveys additional information

[0335] As mentioned above, emissivity is a function of the temperature of the body and its elemental composition. Figure 2 151-P in Figure 4 153-P, and Figure 6 The 150-P in φ is related to the emissivity, although the emissivity does not have to be calculated.

[0336] Since the radiation data is wavelength specific, the data can be viewed as a spectrum with N spectral lines (refer to N wavelengths, N frequencies). The radiation intensity varies with time, and therefore the radiation data is processed by computer (in stages **1 and **3) as a multivariate time series {{λ}}.

[0337] Since radiation transfers energy, there are two well-known aspects of energy, which are simplified here: First, in principle, radiation at shorter λ transfers more energy than radiation at longer λ. Second, the value of a single λ also corresponds to energy, and this energy can vary over time, where lower values ​​correspond to lower energy, and where higher values ​​correspond to higher energy.

[0338] The second aspect is shown in Figure 2 {λ1}, {λ2}, etc. in , which have values ​​that change over time (at the ordinate). Changes over time can be conveniently discussed with curves or graphs, but note that the computer does not have to present such visualizations to the user.

[0339] Figure 7B The radiation date 153-P is shown as two univariate time series for the (first and second) time series {λ1}, {λ2}, and Figure 7B The characteristic data 153-Q is also shown in time series. For convenience, the ΔT interval is applied to all data 153-P and 153-Q. The interval is given as tm from t1 to t17.

[0340] This figure gives some examples of feature enhancement according to multiple feature enhancement rules. Computer (reference Figure 3B Enhancer module 282 / 283 in Figure 6 Module 280 in ) applies rules by processing the radiation data 153-P. Rules (and their corresponding features) can be distinguished as rules by the following operations: the computer (i) processes one univariate time series or (ii) processes two or more univariate time series.

[0341] In this example, {λ1} has rising and falling values ​​(bold line), and the value is sometimes above the reference value and sometimes below the reference value. The reference value can be a threshold value (here given by the horizontal line). It is also possible to normalize the data to such a threshold value or other threshold values. For example, the reference value can be an average value, so that the value is sometimes above the average value ("positive") or below the average value ("negative"). {λ2} also has rising and falling values, but with different times (in this example, the bold line is smoother).

[0342] By way of example, the computer applies the first rule, and the first feature should be the sign of the derivative. This rule can be implemented by comparing consecutive values: if "value(tm+1)>value(tm)", then the derivative is positive, or if "value(tm+1)<value(tm)", then the derivative is negative.

[0343] The computer may apply a second rule, and the second feature should be the identification of a threshold crossing, here representing the time interval t4, t7, t13, 17, during which the value of {λ1} crosses the threshold (in either direction).

[0344] While the first and second features can only be derived from {λ1} (in case (i)), the third feature should be determined by processing both {λ1} and {λ2} (case (ii)). By applying the third rule, the feature enhancer will identify the third feature. In this example, the computer identifies the time intervals where the values ​​of {λ1} and {λ2} are both rising, here the time intervals are t1, t2, t5, t6 and starting from t14.

[0345] The first rule, the second rule and the third rule mentioned above are only examples. The feature recognition of data in the time series does not need to be explained using graphs. This specification will continue by discussing the features without referring to graphs.

[0346] The radiance data is pre-processed to identify features that are local properties and that may be specific to a single wavelength. In other words, changes in lambda values ​​may follow a detectable pattern in the curvature. By way of example, for a single univariate time series {λi}, such a pattern may be of the following form: ● The value of the time series may increase or may decrease (i.e., the first derivative) within an interval of time points,

[0347] Or it can remain unchanged (first derivative is at zero).

[0348] • A time series may show this increasing or decreasing change (ie, the second derivative).

[0349] • The time series may have inflection points (this is just another way of looking at the second derivative).

[0350] • A value may reach zero (and remain at zero for a detectable number of time points), and the same applies to reaching a threshold.

[0351] ●These values ​​can reach a maximum (or minimum) value, regardless of whether or not the absolute value, etc. is considered.

[0352] The radiometric data can also be pre-processed to identify features that are still local in nature, but can be specific to two or more separate wavelengths. Continuing with the example, the features can be detected as follows:

[0353] Over the interval of time points tm, the value for the first wavelength λ1 increases, and at the same time, the value for the second wavelength λ2 decreases (the value for the first wavelength λ1 decreases, and at the same time, the value for the second wavelength λ2 increases). In this sense, the first-order derivatives of these two values ​​will have different signs (positive and negative, or negative and positive).

[0354] For a pair of wavelengths, there may be a second order derivative with signs such as "++", "+-", "-+", "--".

[0356] These values ​​may reach zero in a particular order (eg, first λ1, then λ2).

[0357] The values ​​of λ1 and λ2 may reach the maximum value substantially at the same time, at different times, etc.

[0358] Such features can be detected by a computer (e.g., by a feature detector or feature enhancer), and the detected features can be considered an enhancement to the radiation data. Feature-enhanced radiation data 150-F (see Fig. 7A and Figure 7B ) can be processed like radiation data (at all stages **1, **2, and **3). In other words, feature-enhanced radiation data 150-F (similar to radiation data 150-P) represents thermal radiation from the molten material, but it represents the radiation with greater accuracy, allowing estimates (i.e., elemental content) to be derived with greater accuracy.

[0359] Therefore, the computer executes the method steps accordingly (refer to Figure 5A and Figure 5B ): In the receiving step 413, the computer receives the feature-enhanced radiation data. In the operation step 453, the computer processes the feature-enhanced radiation data. The same principle is also applied to training (refer to Figure 3B for discussion).

[0360] Typically, feature-enhanced radiometric data ( Fig. 7A 150-F and Figure 7BThe examples in are for example only) are data representing local characteristics for the radiation data. The feature enhanced radiation data is wavelength specific data (150-F is shown only for λ2). The term "local" refers to a time interval within a time series.

[0361] Additional examples include the following:

[0362] ● The value of a particular wavelength λi can have a local minimum or a local maximum (in terms of the “local” time interval

[0363] The characteristic enhanced radiation data indicates the time point of the minimum or maximum value.

[0364] The value of a particular wavelength λi may oscillate, and the oscillation can be detected at specific time intervals.

[0365] Fig. 7A A 150-F signal for this temporary oscillation of λ2 is shown. The characteristic enhanced radiation data indicates the start and end times of the oscillation, and optionally the level (relatively low or high oscillation, etc.)

[0366] Because the value of a particular wavelength λi will typically change over time, it is possible to calculate a first derivative (or even a second derivative). For example, an increase in value is a feature, and it will correspond to a positive but more or less stable (first order) derivative. The computer can use this feature (i.e. an indication of the derivative) as a processing input. Changes over time are also considered slopes (curves or graphs), and the skilled person is familiar with classifying them (e.g. steep slopes).

[0367] • Changes over time do not have to be processed per specific wavelength λi, but applying local curve fitting techniques may be advantageous. Such a fitted curve may ignore outliers etc. This is just another example where feature enhancement can improve accuracy: the outlier is simply ignored.

[0368] ●While the first univariate time series changes in a particular pattern over time (such as changing with a steep slope), the second univariate time series may change in a similar pattern (such as also changing with a steep slope), or may change in an opposite pattern (other slope direction). This pattern similarity can also be detected as a feature.

[0369] ● Pattern similarity may involve time differences (similar to the phase shift between voltage and current in electrical technology), and the time difference can also be used as a feature.

[0370] ● Since the univariate time series (here: lambda) represents the radiated energy, the ratio of the values ​​between the two time series can also be used as a feature (e.g., for equal tm, λ1(tm)

[0371] value at λ1(tm) / value at λ1(tm)).

[0372] Time windows and additional statistics

[0373] The specification has explained features by example, in terms of considering the time aspect of one or two (or even more) time series, considering comparing values ​​and other aspects. In general, feature identification can apply statistics. Therefore, the following is just an overview of additional options:

[0374] • Features may evolve over time. An example of such a time evolution for a feature is a change over a period of time longer than ΔT.

[0375] • It is possible to apply a moving time window. Statistics (e.g., mean or average, median, standard deviation, quantiles, skewness, polynomial fit parameters, or any curve fit) can be calculated from the distribution of values ​​within such a window.

[0376] • It is also possible to apply a moving window and to obtain the signal spectrum. By way of example, the following transforms may be applied: Fast Fourier Transform (FFT) or Wavelet Transform.

[0377] ● Applying a moving window has additional aspects in computing the signal autocorrelation and computing associated properties (e.g., time shifting so that the autocorrelation is below a predefined threshold). Since the autocorrelation is applied to a single time series, correlations for multiple time series can also be computed (see Figure 7B for {λ1} and {λ2} in ).

[0378] ● You can also use similarity metrics (such as dynamic time warping), use pre-trained

[0379] A neural network, or rules (i.e., additional auxiliary networks) is used to detect time series patterns.

[0380] • Furthermore, the correlation between one radiation signal and another signal (for a moving window) can be recorded.

[0381] ● Comparison with specific historical data can be based on specific historical casting (such as the best performance

[0382] The castings are classified as “successful castings”, or even castings with poor performance).

[0383] ● To limit the computational workload (in terms of memory, CPU, bandwidth, and other resources),

[0384] Feature recognition can be applied through dimensionality reduction (such as principal component analysis), or autoencoder machine learning.

[0385] analogy

[0386] Technicians can find some analogies to enhanced radiometric data by talking to astronomers. Stars may periodically change their color, may change their brightness, and so on. This and other changes may point to properties of the star. But the analogy stops there, because astronomers don't take physical samples (the moon and comet examples don't apply).

[0387] Time interval for data

[0388] Figure 4 The time interval T_cast is shown, and as already explained, T_cast may start when the drilling machine starts to drill the opening hole. In other words, T_cast may mark the start of each individual casting process. For the computer 203, receiving radiation data for data from the start of each casting (step 413, Figure 5A , Figure 5B and process them in step 453). The same principle applies to training: separating the historical data according to T_cast (optional) has been interpreted as advantageous.

[0389] However, this approach need not be followed. Figure 4 The time interval T_back from the point in time when casting is already in progress until t_current is shown. Thus, T_back can mark the time length of the time series to be processed. In many cases, T_back will correspond to a window (refer to Fig. 7A ).

[0390] T_back (ie, a value representing its duration) may also be input to the processing module 253 of the computer 203. (The same principle also applies to training.) Although T_back does not come from a sensor, it may be processed like sensor data.

[0391] General purpose computer

[0392] Fig.11 An example of a general purpose computer device that may be used with the techniques described herein is shown. Fig.11It is a diagram showing an example of a general-purpose computer device 900 and a general-purpose mobile computer device 950 that can be used with the technology described herein. The computing device 900 is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The general-purpose computer device 900 may correspond to the computers already shown in other figures. The computing device 950 is intended to represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, on-board computers of driving assistance systems or vehicles, and other similar computing devices. For example, the computing device 950 can be used by a user (e.g., an operator of a blast furnace) as a front end for interacting with the computing device 900. The components shown here, their connections and relationships, and their functions are only meant to be exemplary and are not meant to limit the implementation of the present invention described and / or required in this document.

[0393] The computing device 900 includes a processor 902, a memory 904, a storage device 906, a high-speed interface 908 connected to the memory 904 and a high-speed expansion port 910, and a low-speed interface 912 connected to a low-speed bus 914 and the storage device 906. Each of the components 902, 904, 906, 908, 910, and 912 is interconnected using various buses and can be installed on a common motherboard or in other appropriate ways. The processor 902 can process instructions for execution within the computing device 900, including instructions stored on the memory 904 or the storage device 906, to display graphical information of a GUI on an external input / output device (such as a display 916 coupled to the high-speed interface 908). In other implementations, multiple processors and / or multiple buses, as well as multiple memories and multiple types of memories, can be used appropriately. In addition, multiple computing devices 900 can be connected, where each device provides a portion of the necessary operations (for example, as a server bank, a group of blade servers, or a multi-processor system).

[0394] The memory 904 stores information within the computing device 900. In one implementation, the memory 904 is one or more volatile memory units. In another implementation, the memory 904 is one or more non-volatile memory units. The memory 904 may also be another form of computer-readable medium, such as a magnetic disk or optical disk.

[0395] The storage device 906 can provide mass storage for the computing device 900. In one implementation, the storage device 906 can be or include a computer-readable medium (such as a floppy disk device, a hard disk device, an optical disk device, or a magnetic tape device, a flash memory or other similar solid-state storage device, or an array of devices), including devices in a storage area network or other configurations. A computer program product can be embodied in an information carrier in a tangible manner. The computer program product can also include instructions that, when executed, will perform one or more methods, such as those described above. The information carrier is a computer-readable medium or a machine-readable medium, such as a memory 904, a storage device 906, or a memory on a processor 902.

[0396] The high-speed controller 908 manages bandwidth-intensive operations for the computing device 900, while the low-speed controller 912 manages relatively low bandwidth-intensive operations. Such functional allocation is merely exemplary. In one implementation, the high-speed controller 908 is coupled to the memory 904, the display 916, and the high-speed expansion port 910 (e.g., by a graphics processor or an accelerator), and the high-speed expansion port 910 can accept various expansion cards (not shown). In this implementation, the low-speed controller 912 is coupled to the storage device 906 and the low-speed expansion port 914. The low-speed expansion port that can include various communication ports (e.g., USB, Bluetooth, Ethernet, wireless Ethernet) can be coupled to one or more input / output devices, such as keyboards, pointing devices, scanners, or network devices such as switches or routers, for example, through a network adapter.

[0397] The computing device 900 can be implemented in a variety of different forms, as shown in the figure. For example, it can be implemented as a standard server 920, or multiple times in a group of such servers. It can also be implemented as part of a rack server system 924. In addition, it can be implemented in a personal computer such as a laptop computer 922. Alternatively, components from the computing device 900 can be combined with other components in a mobile device (not shown) such as device 950. Each of such devices can include one or more of the computing devices 900, 950, and the entire system can be composed of multiple computing devices 900, 950 that communicate with each other.

[0398] The computing device 950 includes a processor 952, a memory 964, an input / output device such as a display 954, a communication interface 966, and a transceiver 968, among other components. The device 950 may also be provided with a storage device such as a micro drive or other device to provide additional storage. Each of the components 950, 952, 964, 954, 966, and 968 are interconnected using various buses, and multiple components may be mounted on a common motherboard or in other suitable ways.

[0399] The processor 952 can execute instructions within the computing device 950, including instructions stored in the memory 964. The processor can be implemented as a chipset in a chip that includes separate and multiple analog and digital processors. For example, the processor can provide coordination of other components in the device 950, such as control of a user interface, applications run by the device 950, and wireless communications through the device 950.

[0400] The processor 952 can communicate with the user through a control interface 958 and a display interface 956 coupled to the display 954. The display 954 can be, for example, a TFT LCD (thin film transistor liquid crystal display) or an OLED (organic light emitting diode) display, or other appropriate display technology. The display interface 956 may include appropriate circuits for driving the display 954 to present graphics and other information to the user. The control interface 958 can receive commands from the user and convert them to be submitted to the processor 952. In addition, an external interface 962 in communication with the processor 952 can be provided to enable near-area communication of the device 950 with other devices. The external interface 962 can be provided, for example, for wired communication in some implementations, or for wireless communication in other implementations, and multiple interfaces can also be used.

[0401] The memory 964 stores information in the computing device 950. The memory 964 can be implemented as one or more computer-readable media, one or more volatile memory units, or one or more non-volatile memory units. An expansion memory 984 can also be provided, and the expansion memory 984 is connected to the device 950 through an expansion interface 982, which can include, for example, a SIMM (single-line memory module) card interface. Such an expansion memory 984 can provide additional storage space for the device 950, or can also store applications or other information for the device 950. Specifically, the expansion memory 984 can include instructions to perform or supplement the process described above, and can also include security information. Therefore, for example, the expansion memory 984 can act as a security module for the device 950, and can be programmed with instructions that allow the safe use of the device 950. In addition, a security application can be provided via a SIMM card, as well as additional information, such as placing identification information on the SIMM card in an unbreakable manner.

[0402] For example, the memory may include flash memory and / or NVRAM memory, as discussed below. In one implementation, the computer program product is tangibly embodied in an information carrier. The computer program product includes instructions that, when executed, perform one or more methods, such as those described above. The information carrier is a computer-readable medium or machine-readable medium, such as the memory 964, the expansion memory 984, or a memory on the processor 952, which may be received, for example, via the transceiver 968 or the external interface 962.

[0403] The device 950 can communicate wirelessly via a communication interface 966, which may include digital signal processing circuitry if necessary. The communication interface 966 may provide for communication in various modes or protocols, such as GSM voice calls, SMS, EMS or MMS messages, CDMA, TDMA, PDC, WCDMA, CDMA2000 or GPRS, etc. Such communication may, for example, be generated via a radio frequency transceiver 968. In addition, short-range communications may be generated, such as using Bluetooth, WiFi or other such transceivers (not shown). In addition, a GPS (global positioning system) receiver module 980 may provide additional navigation-related and location-related wireless data to the device 950, which may be used appropriately by applications running on the device 950.

[0404] Device 950 may also communicate audibly using audio codec 960, which may receive voice information from a user and convert the voice information into usable digital information. Audio codec 960 may also generate audible sounds to the user, such as through a speaker, for example, in a handset of device 950. Such sounds may include sounds from voice phone calls, may include recorded sounds (e.g., voice messages, music files, etc.), and may also include sounds generated by applications operating on device 950.

[0405] As shown in the figure, computing device 950 can be implemented in many different forms. For example, it can be implemented as a cellular phone 980. It can also be implemented as a part in a smart phone 982, a personal digital assistant or other similar mobile device.

[0406] Various implementations of the systems and techniques described herein can be realized in digital electronic circuitry, integrated circuits, specially designed ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementations in one or more computer programs that are executable and / or interpretable on a programmable system that includes at least one programmable processor that can be special or general purpose, at least one input device, and at least one output device that is coupled to receive data and instructions from a storage system and transmit data and instructions to the storage system.

[0407] These computer programs (also referred to as programs, software, software applications, or code) include machine instructions for a programmable processor and may be implemented in high-level procedural and / or object-oriented programming languages, and / or user assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and / or device (e.g., a disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0408] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and any form of input from the user can be received, including acoustic input, voice input, or tactile input.

[0409] The systems and techniques described herein may be implemented in a computing device that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a client computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or that includes any combination of such back-end, middleware, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), and the Internet.

[0410] Computing devices may include clients and servers. Clients and servers are generally remote from each other and typically interact through a communication network. The relationship of client and server arises from computer programs running on the respective computers and having a client-server relationship to each other.

[0411] A number of embodiments have been described, however, it will be appreciated that various modifications can be made without departing from the spirit and scope of the invention.

[0412] Furthermore, the logic flows depicted in the figures do not require the particular order shown or sequential order to achieve the desired results. In addition, other steps can be provided from the described flows, or these steps can be eliminated from the described flows, and other components can be added to the described systems, or other components can be removed from the described systems. Therefore, other implementations are within the scope of the appended claims.

[0413] Further conclusion

[0414] As explained in detail, a pyrometer is mechanically arranged at the container within a monitoring distance. The pyrometer monitors the opening, and a computer receives radiation data from the pyrometer. The radiation data represents thermal radiation at at least two wavelengths from the material at the opening. So far, the description has explained that the computer operates a processing module that processes the radiation data into feature-enhanced data. However, feature enhancement is not necessary in all cases. Therefore, the description ends with a summary of the method that only optionally uses feature enhancement.

[0415] The computer operates a processing module, which processes the radiation data. The processing module has a regression model to estimate the content of a specific chemical element (ie, "content data"), and the processing module has been pre-trained.

[0416] The pyrometers can measure the radiation from the tapholes of the blast furnace. To increase accuracy, additional sensor data can optionally be processed.

[0417] A computer-implemented method is provided for estimating the content of a specific chemical element in a molten material (available at an opening of a metallurgical vessel). A computer receives radiation data from a pyrometer configured to monitor the opening, the radiation data representing thermal radiation for at least two wavelengths from the molten material at the opening. The computer operates a processing module that processes the radiation data, wherein the processing module has a regression model to estimate the content of the specific chemical element, and wherein the processing module has been pre-trained.

[0418] Optionally, the step of operating the processing module comprises operating a module that has been trained using training data collected during the data collection phase. The collected training data is a combination of historical radiation data and historical content data obtained by measuring samples taken from the molten material at the opening.

[0419] Optionally, in a further receiving step, the computer receives sensor data from a sensor configured to monitor the container; and in the step of operating the processing module, the computer processes the radiation data in combination with the sensor data.

[0420] Optionally, after receiving, the computer performs adaptation of the sensor data to obtain intermediate data. Therefore, the computer performs the step of operating the processing module using the intermediate data.

[0421] Optionally, the computer performs the adaptation with the aid of a machine learning tool. The tool may be an autoencoder.

[0422] Optionally, the computer performs the additional step of receiving one or more images showing the opening or showing the path between the opening and the pyrometer from a camera device, and the computer classifying the one or more images to identify phenomena. These phenomena selectively include at least one of the following: accumulations, smoke, or reflections located in the path; the degree of alignment of the pyrometer with the opening; and the proportion of metal and slag in the image. Based on the identified phenomena, the computer classifies the radiation data from the pyrometer as reliable radiation data or unreliable radiation data. In other words, reliability and unreliability are quality categories, and the computer distinguishes these categories by processing the images. As mentioned, the computer processes the radiation data to estimate the content of a specific chemical element, but investigating reliability can be regarded as a "quality assurance measure" that ultimately causes the computer to automatically ignore certain estimates.

[0423] Optionally, the molten material is a composition comprising a molten metal or a metal alloy, and wherein the specific chemical element is selected from the group consisting of carbon, silicon, iron and sulfur.

[0424] Optionally, the method steps are applied to a metallurgical vessel, the metallurgical vessel being a blast furnace, for which the opening comprises a tap hole and a runner to carry material away from the blast furnace during casting.

[0425] Optionally, the method may be performed using feature-enhanced radiation data, which is data representing local characteristics for radiation data: a computer may receive the feature-enhanced radiation data, and in an operation step, the computer may process the feature-enhanced radiation data.

[0426] Optionally, the computer performs the further step of presenting said estimate together with the content data to an operator of the blast furnace.

[0427] In an iterative manner, this method can be used to control the operation of the container.

[0428] A computer program product which, when loaded into a memory of a computer system and executed by at least one processor of the computer system, causes the computer system to perform the steps of a computer-implemented method.

[0429] The computer system includes a plurality of modules that perform the steps of the computer-implemented method.

[0430] A system is provided for estimating the content of a specific chemical element in a molten material available at an opening of a metallurgical vessel.

[0431] A pyrometer is suitable for monitoring an opening and for providing radiation data, which represents thermal radiation for at least two wavelengths from a molten material at the opening. A camera device is suitable for monitoring the opening or a path between the opening and the pyrometer to provide one or more images showing the opening or the path. A computer has an estimation module and an image classification module. The estimation module is suitable for processing the radiation data using a regression model to provide an estimate of preliminary content data for a specific chemical element in the molten material. The estimation module has been trained by training data collected during a data collection phase. The collected data is a combination of historical radiation data and historical content data (obtained by measuring samples collected from the molten material at the opening). The image classification module is suitable for classifying one or more images to identify phenomena at the opening or at the path, and then providing a reliability classification RC for the radiation data based on the identified phenomena, so that the computer outputs the preliminary content data as content data only when the reliability classification RC meets predefined rules.

[0432] Optionally, the phenomenon selectively includes at least one of: accumulation, smoke, or reflections located in the path; the degree of alignment of the pyrometer with the opening; and the proportion of metal and slag on the image.

[0433] Reference numerals

[0434] The following overview considers the various phases and reference numerals are given to the items shown.

[0435] 100 / 101 / 103 Metallurgical Containers

[0436] 110 / 111 / 113 Opening

[0437] 120 holes or holes on the image

[0438] 130 Guide groove

[0439] 141-P / 143-P Pyrometer

[0440] 140-C Camera Device

[0441] 150-P / 151-P / 153-P Radiation Data

[0442] 150-X Sensor Data

[0443] 150-C Image (for example sensor data)

[0444] 150-F Signature Enhanced Radiation Data

[0445] 150-Q, 153-Q characteristic data

[0446] 160 Phenomenon

[0447] 171 / 173 Sample Operator

[0448] 181 Laboratory

[0449] 190 / 193 Container Operator

[0450] 200 / 202 / 203 Computer

[0451] 241 Training Data

[0452] 252 / 253 Processing Module

[0453] 260 Adapter

[0454] 270, 273 Content data estimated by computer

[0455] 270' Preliminary content data

[0456] 280 / 282 / 283 Feature Enhancer Module

[0457] 300 / 301 / 303 Molten Material

[0458] 311 / 313 Samples of molten material

[0459] 350 Metal

[0460] 360 Slag

[0461] 370,371 Content data measured by laboratory

[0462] 4xx method, method steps 501, 503 controller

[0463] 6xx blast furnace and system components in the example

[0464] 9xx general purpose computers and their components

[0465] 1000 Systems

[0466] δT_COM The available interval of content data at the output of the processing module δT_LAB The available interval for content data from the laboratory ΔT Sampling interval at the input of the processing module M Training repetitions

[0467] RC reliability classification.

[0468] As mentioned, the radiation data λ1 , λ2 are given without phase indication.

Claims

1. A computer-implemented method (403) for estimating the content (273) of a specific chemical element in a molten material (303) accessible at an opening (113) of a metallurgical vessel (103), the method (403) comprising the following steps: The computer (203) receives (413) radiation data (153-P) from a pyrometer (143-P) configured to monitor the opening (113), the radiation data (153-P) representing thermal radiation for at least two wavelengths (λ1, λ2) from the molten material (303) at the opening (113); The computer (203) identifies features from the radiation data (153-P) according to a plurality of feature enhancement rules and provides feature data (153-Q), wherein the radiation data (153-P) and the feature data (153-Q) together become feature-enhanced radiation data (150-F); and The computer (203) operates (453) a processing module (253) that processes the feature-enhanced radiation data (153-F), wherein the processing module (253) has a regression model to estimate the content (273) of the specific chemical element, wherein the processing module (253) has been pre-trained.

2. The method (403) of claim 1, wherein: The step of operating (453) the processing module (253) includes operating a module (253 / 252) that has been trained by training data (241) collected during a data collection phase (**1), wherein the training data (241) is a combination of historical radiation data (151-P) and historical content data (371) obtained by measuring a sample (311) collected from the molten material (301) at the opening (111), and wherein the training data (241) also includes feature data (151-Q), and the feature data (151-Q) is identified according to a plurality of feature enhancement rules.

3. The method (403) of claim 2, wherein: When the data collection phase (**1) has been completed, the feature data (151-Q) has been identified according to a plurality of the feature enhancement rules.

4. The method (400) according to any one of claims 1 to 3, wherein: In a further receiving step (423), the computer (203) receives sensor data (150-X, 153-X) from a sensor configured to monitor the container (103); as well as In the step of operating (453) the processing module (253), the computer (203) processes the feature-enhanced radiation data (153-F) in combination with the sensor data (150-X, 153-X).

5. The method (400) of claim 4, wherein: After receiving (423) the sensor data (150-X, 153-X), the computer adapts (433) the sensor data to obtain intermediate data, and wherein the computer performs the step of operating (453) the processing module (253) using the intermediate data.

6. The method (400) of claim 5, wherein: The computer performs the adaptation (433) by assisting a machine learning tool (260).

7. The method (400) of claim 5, wherein: The computer performs the adaptation (433) by means of an auxiliary machine learning tool (260), which is an autoencoder.

8. The method (400) according to any one of claims 1 to 7, comprising the further steps of: receiving one or more images (150-C) showing the opening (110) or showing a path (110 / 140) between the opening (110) and the pyrometer (140-P) from a camera (140-C); classifying one or more of the images (150-C) to identify a phenomenon (160), wherein: The phenomenon (160) selectively includes at least one of: accumulation, smoke, or reflections in the path (110 / 140); the degree of alignment of the pyrometer with the opening (110); and the proportion of metal and slag on the image (150-C); as well as Based on the identified phenomenon (160), radiation data (153-P) from the pyrometer is classified as reliable radiation data or unreliable radiation data.

9. The method (403) according to any one of claims 1 to 8, wherein: The molten material (123) is a composition ({C, Fe, S, Si}) including molten metal or metal alloy, and wherein the specific chemical element is selected from carbon (C), silicon (Si), iron (Fe) and sulfur (S).

10. The method (400) of claim 9, wherein: The method steps (413, 423, 453) are applied to a metallurgical vessel (101 / 103), which is a blast furnace, wherein the opening (123) comprises a tap hole and a flow channel to transport the material away from the blast furnace during casting.

11. A method (400) as claimed in claim 10, comprising the further step of presenting the estimate together with the content data (273) to an operator (193) of the blast furnace (100).

12. Repeated use of the method (403) according to claims 1 to 11 for controlling the operation of the container (103).

13. A computer program product which, when loaded into a memory of a computer system and executed by at least one processor of the computer system, causes the computer system to perform the steps of the computer-implemented method according to any one of claims 1 to 11.

14. A computer system comprising a plurality of modules for performing the steps of the computer-implemented method according to any one of claims 1 to 11.

15. A system (1000) for estimating the content of a specific chemical element in a molten material accessible at an opening (110) of a metallurgical vessel (100), the system comprising: a pyrometer (140-P) adapted to monitor the opening (110) and provide radiation data (150-P) representing thermal radiation from the molten material (300) at the opening (110) for at least two wavelengths (λ1, λ2); a camera device (140-C) adapted to monitor the opening (110) or the path (110 / 140) between the opening (110) and the pyrometer (140-P) to provide one or more images (150-C) showing the opening or showing the path; A computer (200) having a feature enhancer module (280), an estimation module (250), and an image classification module (260), wherein the feature enhancer module (280) is adapted to process the radiation data (150-P) into feature data (150-Q) according to a plurality of feature enhancement rules, wherein the radiation data (153-P) together with the feature data (150-Q) become feature enhanced radiation data (150-F); wherein the estimation module (250) is adapted to process the feature enhanced radiation data (150-F) using a regression model to provide an estimate of preliminary content data (270') for a specific chemical element in the molten material, wherein the estimation module (250) has been trained by training data (241) collected during a data collection phase (**1), wherein the collected data is a combination of historical radiation data (151-P) and historical content data (371) obtained by measuring samples (311) collected from the molten material (301) at the opening (111), wherein the training data (241) also includes feature data, the feature data being identified according to a plurality of the feature enhancement rules; and The image classification module (260) is adapted to classify the one or more images (150-C) to identify a phenomenon (160) at the opening (110) or at the path (110 / 140), and subsequently provide a reliability classification (RC) of the radiation data based on the identified phenomenon (160), so that the computer outputs preliminary content data (270') as content data (270) only when the reliability classification (RC) meets predefined rules.

16. The system (1000) of claim 15, wherein: The phenomenon (160) selectively includes at least one of: accumulation, smoke, or reflections in the path (110 / 140); the degree of alignment of the pyrometer with the opening (110); and the proportion of metal and slag on the image.

Images