Methods for monitoring steel processing lines, related electronic devices, and steel processing lines

Abstract

The present invention relates to a control module (11) for determining a line control signal (MP) for controlling a steel processing line, wherein the line control signal is determined according to the chemical composition CC of a steel semi-finished product (6) to be processed and the target characteristics P of the semi-finished product; and an abnormality detector (12) for determining an abnormality indicator (ind) that specifies whether the line control signal (MP) is normal or abnormal, and then an abnormality cause (AC) selected from a predetermined list of abnormality causes. k The present invention relates to a method for monitoring a steel processing line (1), comprising an anomaly detector (12) in which an anomaly indicator is determined using a trained classifier that includes at least the chemical composition CC, the target characteristic P, and a line control signal (MP), wherein an anomaly indicator is specified and the anomaly indicator is determined using a trained classifier that includes at least the chemical composition CC, the target characteristic P, and a line control signal (MP).

Description

Technical Field

[0001] The present invention relates to a method for monitoring a steel processing line for processing semi-finished steel products in order to detect possible process parameter drifts, or possible product parameter drifts, or possible malfunctions of process actuators. The present invention also relates to associated anomaly detectors and a method for training such detectors.

[0002] During the production of steel sheets, for example, the steel sheets are subjected to several heat treatments in order to obtain a target microstructure and properties for a specific application. Such heat treatments can be, for example, continuous heat treatments or quenching and tempering treatments before the deposition of a metal coating.

[0003] During heat treatment in a continuous heat treatment line, the process parameters are selected to obtain the target microstructure and properties at the end of the process. These process parameters include the continuous temperature, heating and cooling rates, and the heat path corresponding to the time spent in each section of the heat treatment process.

[0004] During heat treatment, some of the processing steps can be impaired by a malfunction of a particular actuator of the continuous heat treatment line, or by a drift in particular operating conditions (such as the temperature of the inert gas in the furnace of the heat treatment line, or a drift in the sheet speed, etc.), and thus may require correction of other process parameters and / or may cause a deviation from the target properties and microstructure. In addition, some differences in chemical composition may occur, which may change the target final microstructure and properties.

Background Art

[0005] International Publication No. 2013023903 relates to a method for operating a continuous heat treatment line for processing rolled steel strip, comprising: using a measuring device to measure the properties of the rolled steel strip; supplying the properties of the rolled steel strip as input variables to a model of a control device; using a model and properties of the rolled steel strip to predict the material properties of the rolled steel strip after the continuous heat treatment line and to produce the predicted material properties; and comparing the predicted material properties with predetermined target values, wherein if the simulated material properties deviate from the target values, at least one process variable of the continuous heat treatment process is adjusted.

[0006] In this method, a model is used during production for each production campaign in order to implement adjustments during this production campaign.

[0007] More generally, the operation of steel processing lines, such as hot rolling lines, cold rolling lines, pickling lines, or annealing lines, is regulated and controlled, for example, using feedback control, to obtain a final product with desired characteristics despite possible deviations in operating conditions or the possibility of actuator saturation or malfunction. These regulatings allow for at least partial compensation of such deviations or malfunctions. However, they do not allow for identification of which process or product parameters are drifting or which actuators are malfunctioning.

[0008] Identifying that a specific actuator is malfunctioning is beneficial because it allows for planning appropriate maintenance or repairs, particularly to prevent the malfunction from escalating to the point where it necessitates an unintended abrupt shutdown of the processing line. Similarly, identifying that process or product parameters are deviating from values ​​corresponding to normal operating conditions is beneficial because it allows for addressing the drift before it becomes significant. Such detection is also beneficial because it allows for adjusting process parameters to compensate for the drift or malfunction in order to maintain product quality. [Prior art documents] [Patent Documents]

[0009] [Patent Document 1] International Publication No. 2013 / 023903 [Overview of the project] [Problems that the invention aims to solve]

[0010] In this regard, an object of the present invention is to provide a method for monitoring a steel processing line, enabling detection of abnormal operating conditions, and enabling identification of the cause of the abnormality from a predetermined list of abnormality causes. [Means for solving the problem]

[0011] The object of the present invention is achieved by providing the method according to claim 1.

[0012] In the event of a drift in operating conditions, such as a drift in the temperature of the inert gas present in the furnace, one of the monitoring signals changes (in this case, the signal output by the temperature sensor installed in the furnace changes). The control module then adjusts one or more of the line control signals (e.g., signals controlling the radiating tubes in the furnace) in response to this fluctuation in the monitoring signal, in an attempt to obtain the desired target characteristics at the end of the process despite this fluctuation in operating conditions. Therefore, in such a situation, the line control signals will have values ​​different from those under normal reference operating conditions. Thus, the line control signals are "abnormal" in such a situation and differ from those under normal reference operating conditions. Therefore, analysis of the line control signals to detect whether they are abnormal can, in principle, make it possible to detect that there is a drift or malfunction. Similarly, if one of the cooling jets of a cooling system is inoperable, the control module compensates for this by adjusting the force of the other cooling jets, which is reflected by an abnormal "unusual" line control signal and can, in principle, be detectable.

[0013] While such detection may be possible in principle, it is not immediately clear whether it is actually achievable and how it can be achieved. In fact, in such a steel processing line, the characteristics of the successive semi-finished products being processed, such as their size or chemical composition, usually vary from semi-finished product to semi-finished product, and the target properties (e.g., desired final tensile strength) also vary. Therefore, monitoring signals and line control signals fluctuate over time and from semi-finished product to semi-finished product (sometimes significantly). Consequently, even in defect-free operation, monitoring signals fluctuate.

[0014] To address such inherent variations, the detection method is: - The chemical composition and target properties (or estimated final properties) of the steel semi-finished product, along with the line control signals output by the control module, - Using a trained classifier and Combine.

[0015] The inventors observed that this approach actually makes it possible to detect anomalies and identify their causes. In other words, it makes it possible to identify anomaly signatures within line control signals, despite the complexity of the situation and their natural variability in normal conditions.

[0016] Test results demonstrating the effectiveness of this method will be presented in a detailed explanation.

[0017] It should be noted that the monitoring methods described above do not necessarily involve training a trained classifier; generally, the trained classifier is trained in advance before the monitoring method begins (during the step-up phase).

[0018] Some of the labeled training data used to train the classifier may be simulated training data instead of pre-recorded actual steelworking data. This is useful because drift and malfunctions do not occur frequently on steelworking lines (and are therefore tried to avoid). Thus, by simulating abnormal steelworking data, it becomes possible to have processing data at hand that corresponds to drift or failure scenarios that may occur in the future but have not occurred on the line beforehand. Thus, by using such simulated training data, a broad search for the cause of anomalies becomes possible. Once trained, the classifier leverages the substantial simulation and training work achieved, and when it comes to using the trained classifier, it enables real-time and accurate detection of anomalies in line control signals, despite the a priori complexity of such detection / analysis.

[0019] The simulated training data described above may be computed using (or a numerical copy of) the same control module used to control the processing line. In other words, the simulated training data may be computed using the same steel prediction model and the same calculation rules used by the control module to compute the actual line control signals.

[0020] The method may also include features described in any one of claims 2 to 16, either individually or in combination.

[0021] The present invention also relates to an electronic device for controlling a steel processing line as described in claim 17. This electronic device may include one or more of the additional features defined with respect to the methods of claims 2 to 16.

[0022] The present invention also relates to the steel processing line described in claim 18.

[0023] The present invention also relates to a training method according to claim 19 for training an anomaly detector of such an electronic device. This training method may include one or more of the additional features defined in claims 2 to 13.

[0024] The present invention also relates to the computer program described in claim 20.

[0025] Next, the present invention will be described in detail without introducing any limitations, with reference to the attached figures, and will be illustrated by examples. [Brief explanation of the drawing]

[0026] [Figure 1] This is a schematic diagram of a steel processing line according to the present invention. [Figure 2] This diagram provides a more detailed representation of the steel processing line shown in Figure 1, specifically when the steel processing line is a hot-dip galvanizing line equipped with heat treatment facilities. [Figure 3] This diagram schematically illustrates the steps involved in determining the line control signal for controlling the steel processing line shown in Figure 1. [Figure 4] Figure 1 is a flowchart illustrating the steps taken to obtain simulated labeled training data for training anomaly detectors in a steel processing line. [Modes for carrying out the invention]

[0027] First, some general embodiments of the present invention are presented. Next, a more detailed description of the implementation of anomaly detectors in steel processing lines when the processing includes heat treatment is provided. Finally, numerical examples are described.

[0028] Steel processing line Figure 1 shows a steel processing line 1 for processing steel semi-finished products 6. Steel semi-finished products can be steel sheets such as hot-rolled or cold-rolled steel sheets, slabs, billets, blooms, ingots, bars, beams, tubes, or wires. More generally, steel semi-finished products are intermediate supply products that are subsequently intended to be used to manufacture parts, articles, or structures.

[0029] Steel processing line 1 may be a continuous processing line suitable for processing steel semi-finished products without interruption. The steel processing line may include a furnace such as a heating furnace or annealing furnace, a roughing mill or a finishing mill, a runout table, or other types of heat treatment equipment (including a controlled cooling system, or a furnace, a soaking system, and a controlled cooling system). Steel processing line 1 may be a hot rolling line, a cold rolling line, a pickling line, a hot-dip galvanizing line, or another steel processing line. It may also include two or more of the processing lines described above.

[0030] The steel processing line 1 includes an actuator 3 for operating the steel processing line to process steel semi-finished products 6. The actuator 3 includes, for example, a furnace heater, a motor for rotating rollers, a colling device (e.g., gas or water jet), and / or roller jacks.

[0031] Steel processing line 1 also, - Operating conditions such as the temperature or composition of the gas filled in the furnace, the measured moving speed of the steel plate, or the force applied by the mill roller, -and / or intermediate characteristics of the steel semi-finished product 6 being processed, such as temperature, dimensions, measured microstructure or emissivity of the steel semi-finished product at a predetermined intermediate position (or at the start thereof) in the steel processing line. It is equipped with a sensor 2 suitable for outputting a monitoring signal MS representing [the specified value].

[0032] Sensor 2 may comprise one or more of the following: a temperature probe, a pyrometer, a scanner, a laser sensor for measuring sheet travel speed or sheet thickness, a barometer, a gas composition analyzer, a camera or set of cameras, an ultrasonic sensor, or a voltmeter or ammeter for measuring the voltage or current supplied to one of the actuators.

[0033] The steel processing line 1 also includes an electronic device 10, which includes a control module 11 and an anomaly detector 12. The electronic device 10 includes at least one processor and one memory. It has the structure of a computer device or system. It is configured and programmed, for example, to perform the monitoring method according to the present invention. The control module 11 and the anomaly detector 12 may take the form of two separate electronic units. They may also correspond to two different instruction groups (two separate programs, or subprograms). The operation of the control module 11 and the anomaly detector 12 will be described in more detail later.

[0034] The steel processing line 1 may, in particular, include heat treatment equipment for heat-treating steel sheets (for example, if the line is a hot-dip galvanizing line). During the heat treatment, the sheets are subjected to at least one cooling step and possibly one heating step according to a heat path. Typically, the heat treatment can be carried out in an oxidizing atmosphere, i.e., an atmosphere containing an oxidizing gas such as O2, CH4, CO2, or CO. They can also be carried out in a neutral atmosphere, i.e., an atmosphere containing a neutral gas such as N2, Ar, or He. Finally, they can also be carried out in a reducing atmosphere, i.e., an atmosphere containing a reducing gas such as H2 or HNx. The heat path may also include at least one isothermal holding step, which typically consists of a heating step followed by a cooling step. The cooling step may include isothermal holding called an overaging substep, followed by a subsequent cooling step. The molten plating coating step in a high-temperature metal bath (e.g., high-temperature bath 23 in Figure 2) can also be used in such a thermal path and is another type of isothermal holding, as the metal sheet immersed in such a high-temperature metal bath is maintained at the bath temperature during the holding time in such a bath.

[0035] The aforementioned heat treatment includes, for example, the following processes: - Recrystallization annealing, - Tempering, -recovery, - Rapid cooling and tempering, - Rapid cooling and partitioning They can be selected from among them.

[0036] Figure 2 shows the steel processing line in Figure 1 in more detail when the steel processing line is a hot-dip galvanizing line 4. The hot-dip galvanizing line 4 includes a decoiler 15, an annealing furnace 16, a coating device 17, and a coiler 18.

[0037] The annealing furnace 16 is equipped with, for example, a first sensor 19 for measuring the temperature Ts of a steel semi-finished product 6 (in this case, a steel sheet), a second sensor 20 for measuring the temperature of the gas filled in the annealing furnace, and a third sensor 21 for determining the composition of the gas. The annealing apparatus 16 also includes rollers 22 for guiding the steel sheet and a heating device 32 (such as a ceramic radiating tube).

[0038] The coating apparatus 17 comprises a snout 24 housing a cooling device 26, a bath 23 of molten metal, such as molten zinc, a wiping device 33, and rollers 27 for guiding the steel sheet. The cooling device 26 comprises cooling jets, which here comprises 11 consecutive cooling jets J1 to J11, each spraying HNx. The coating apparatus 17 is also equipped with a fourth sensor 28 for measuring the coating thickness on the sheet after wiping and a fifth sensor 29 for measuring the sheet temperature Ts' after cooling. The coating apparatus 17 also comprises at least one electric motor 31 for driving one of the rollers 27.

[0039] Control module The control module 11 performs the following steps: -below: o Chemical composition of steel semi-finished product 6 CC, o Target characteristics P of the steel semi-finished product obtained at the end of the process, The steps include acquiring a monitoring signal MS output by sensor 2 of the steel processing line, - A step of determining a line control signal MP for controlling an actuator 3 of a steel processing line 1, wherein the line control signal MP is determined as a function of chemical composition CC, target characteristic P, and monitoring signal MS, and is determined using a steel property prediction model. - A step of controlling actuator 3 based on line control signal MP and It is configured to execute.

[0040] Monitoring signals and line control signals are not necessarily time-varying signals; each can be a constant setpoint for configuring one of the process parameters.

[0041] The target characteristic P may be a mechanical property such as tensile strength, yield strength, total elongation, uniform elongation, hole expansion ratio, or impact toughness. It may also be a dimension of a steel semi-finished product, such as the thickness of a steel sheet, or the length of a coil made from such a steel sheet. It may also be a thermal property, such as the coil temperature, or an radiative property, such as the emissivity of a steel semi-finished product. It may also be a target microstructure or a target surface finish (e.g., target surface roughness). The control module 11 may acquire not just one but more of these target characteristics and then determine the line control signal MP as a function of these multiple target characteristics.

[0042] As shown in Figure 1, the control module 11 may also acquire one or more process parameters of the prior processes that the steel semi-finished product 6 underwent before being processed in the steel processing line 1. Such prior process parameters PPP are, for example, the rolling ratio of the prior hot rolling or cold rolling operation. In addition to the chemical composition CC, the control module 11 may also acquire one or more additional product parameters APP for the steel semi-finished product 6, such as the initial dimensions of the steel semi-finished product and / or the initial microstructure that the steel semi-finished product has before entering the steel processing line, and this microstructure is determined using sensors and / or models. In such cases, the control module 11 takes into account the prior process parameters PPP and / or additional product parameters APP when determining the line control signal MP.

[0043] At least some of the monitoring signals MS are taken into consideration by the control module 11 to adjust the corresponding processing conditions (such as steel sheet speed) in a closed-loop control scheme, so that these processing conditions match the desired processing conditions despite possible process variations or other disturbances. The desired processing conditions may be calculated using a steel property prediction model, depending on the chemical composition CC, and to yield one or more of the above-mentioned target properties at the end of the process.

[0044] The steel property prediction model is, -One or more target characteristics, -Chemical composition CC, and, if applicable, prior process parameters PPP and / or additional product parameters APP, - The operating conditions of the process (temperature at a predetermined point in the process, and / or steel plate speed, etc.), or, in some cases, the direct relationship with the control signal MS used to control the actuator 3. It is a computational tool that provides relationships (in other words, links) between things.

[0045] This relationship may take the form of one or more mathematical formulas (either explicit or implicit) and / or lookup tables and / or iterative calculations and / or other computing algorithms. The steel property prediction model may also be based on a so-called "black box model" that is based on a physical modeling of the process and / or data analysis (e.g., achieved using machine learning algorithms).

[0046] If the steel processing line 1 is a hot-dip galvanizing line as shown in Figure 2, or more generally, includes a heat treatment element, the control module 11 may be based on a computer program or method described, for example, in European Patent No. 3559284 or European Patent No. 3559287. If the steel processing line 1 is a runout table or includes one, the control module 11 may be a control device described in European Patent No. 3645182.

[0047] The control module 11 may implement dynamic control of the steel processing line 1, where the operating conditions of the second part of the line are adjusted according to monitoring signals representing operating conditions or product parameters measured in the first preceding part of the line. Then, in order to obtain one or more target characteristics despite this deviation, the operating conditions of the second subsequent part of the line are adjusted to compensate for the deviation observed in the first part of the line (the deviation between the desired operating conditions and the actually measured operating conditions, or product parameters). Such dynamic control is implemented, for example, as described in European Patent No. 3559287.

[0048] The control module 11 also considers the chemical composition CC, the acquired monitoring signal MS, and / or the line control signal MP determined by the control module to determine one or more estimated final characteristics P of the steel semi-finished product 6 expected at the end of the process. EST It may be configured to calculate one or more estimated final characteristics P. EST These are of the same type as the one or more target properties described above, but they are more accurate predictions of what is expected at the end of the process. In fact, the steel property prediction model may be configured to iteratively determine operating conditions (e.g., thermal paths) that allow for obtaining final properties that are close to (within compliance range) the target properties but not necessarily exactly equal to the target properties. Thus, once this thermal path is selected, the final properties expected at the end of the process may differ slightly from the target properties. Similarly, the control module may calculate different operating conditions and then select the one that yields the properties closest to (but not exactly equal to) the target properties.

[0049] Figure 3 schematically illustrates how the determination of the line control signal MP can be configured. As shown in Figure 3, during the first step 11.1, the desired operating condition OC is determined taking into account one or more target properties (including target property P), chemical composition CC, and optionally, prior process parameters PPP and / or additional product parameters APP. This step is achieved using a steel property prediction model. The desired operating condition OC can also be determined based on one or more of the monitoring signals MS, especially if the steel property prediction model implements dynamic adjustment of operating conditions as described above.

[0050] If the process involves heat treatment, the desired operating conditions typically include the thermal path TP followed by the steel semi-finished product.

[0051] In the next step, 11.2, the line control signal MP is determined based on the desired operating conditions OC and monitoring signal MS obtained. The control signal MP may be determined based on a pre-established calculation rule representing the operating characteristics of the actuator or line, and / or using adjustment techniques (e.g., closed-loop adjustment).

[0052] During step 11.2, the control module 11 also estimates one or more final characteristics P of the steel semi-finished product 6. EST Determine these characteristics. Alternatively, these characteristics can be determined directly during step 11.1 (before step 11.2).

[0053] In principle, different sets of line control signals are often possible to obtain specific desired operating conditions or final characteristics. Therefore, the control module may be configured to determine two or more sets of line control signals. In such a case, one of these possible sets of control signals is selected and then used to control the steel processing line 1. The selected set of control signals might be, for example, one or more estimated final characteristics P that are closest to one or more target characteristics P. EST It brings about this.

[0054] If the process involves heat treatment, the line control signal MP (or, in other words, the manufacturing parameter MP) determined in step 11.1 may include process settings for controlling the furnace heating device, controlling the force of the cooling jets, and / or controlling the speed of the semi-finished products in different sections of the line. The speed may be determined taking into account the maximum speed that is theoretically achievable on the line, taking into account the dimensions or weight of the steel semi-finished products. More generally, the line control signal MP is a signal, or more simply a setpoint, for controlling one or more of the line actuators.

[0055] Regarding the force of the cooling jets, different force combinations are possible. For example, the force of the first cooling jet in a series of consecutive cooling jets can be higher than that of subsequent jets (commonly called "early cooling"). Conversely, the force of the last cooling jet in a series can be higher ("late cooling"). Alternatively, the cooling force can be evenly distributed among the different jets. There are also many other possible force distributions. The control module 11 may determine the force of the cooling jets according to one of these cooling strategies ("e.g., early cooling") and according to the desired heat path determined in step 11.1. Alternatively, the control module 11 may determine different sets of candidate line control signals according to these different cooling strategies, and then, taking into account the target characteristic P, select one of the more appropriate sets of candidate line control signals. The selected cooling strategy is, for example, one or more estimated final characteristics P that are closest to one or more target characteristics P. EST It brings about this.

[0056] As described above, there are often different possible sets of line control signals that are suitable for obtaining one or more final characteristics close to one or more target characteristics of a steel semi-finished product. In fact, a steel processing line 1 typically has a number of actuators whose effects are partially redundant or complementary. Therefore, if one of the actuators 3 is partially or completely inoperable, it is possible to compensate for this failure at least partially by adjusting the line control signals that consequently control the other actuators. In this regard, the control module 11 (if operating according to Figure 3, or in a different manner) is programmed here to be able to determine the most suitable set of line control signals possible, taking into account that one of the actuators is stopped or partially stopped. In other words, the control module 11 can determine the set of line control signals while assuming that one or more of the actuators are stopped or partially stopped. For example, in the case of a cooling jet, the control module 11 can determine the cooling jet force as preferably as possible, assuming that one or more jets are stopped, taking into account the acquisition of one or more target characteristics. In such a case, the other jet forces are determined to compensate for the non-operating jet.

[0057] Anomaly detector The abnormality detector 12 is configured to determine and output an abnormality indicator ind (Figure 1) that specifies whether the line control signal MP is normal or abnormal, and then determines a predetermined abnormality cause AC. kAn abnormality cause selected from a list (where k is an integer from 1 to z) is identified. The abnormality indicator ind may be output by a user interface such as a display screen. It may also be output by sending the abnormality indicator to the control module 11 or another control element, taking into account automated processing and, in some cases, automated implementation of countermeasures when an abnormality is detected. The abnormality detector 12 can be considered a "metasensor" that determines process or product characteristics (here, defects or drift) from data derived from measured values ​​(e.g., steel plate temperature measurements). In other words, it implements a kind of indirect measurement method.

[0058] The anomaly detector 12 uses a trained classifier to determine the anomaly indicator ind, and its input is, -The chemical composition CC of the steel semi-finished product, - One or more target characteristics P of the steel semi-finished product 6 determined by the control module 11, or one or more estimated final characteristics P EST (For example, estimated final tensile strength and microstructure), - Line control signal MP and Includes.

[0059] The input to the trained classifier may also include one or more of the monitoring signals MS.

[0060] Before being used in the steel processing line 1, the trained classifier 12 is trained using several labeled training data, each labeled training data being: - The input to the trained classifier (thus including one or more product characteristics such as chemical composition, target tensile strength and target microstructure, as well as associated line control signals) and the same type of steel processing data, - A label that specifies whether steel processing data is normal or abnormal, and then one of the abnormality causes in the list is specified, and Includes.

[0061] The label of the labeled training data may be specified in the form of a dedicated variable whose value is an integer from 0 to z, for example (0 corresponds to normal conditions, and 1 to z correspond to the possible abnormal causes AC k in the list).

[0062] In the embodiments described herein, the list of abnormal causes includes at least the following abnormal causes: - One of the actuators 3 on the steel processing line 1 is partially or completely inoperable; for example, a cooling jet or a radiant tube is partially or completely inoperable; different actuators, for example different cooling jets, can be respectively associated with different abnormal causes (for example, if there are 11 different cooling jets in the line, there are 11 different abnormal causes), so that when an abnormality is detected, the potentially malfunctioning actuator can be identified. - One of the monitoring signals MS deviates from the corresponding reference monitoring signal MS REF by one of the operating conditions or a drift of one of the intermediate characteristics of the semi-finished steel product; for example, a drift of the furnace gas temperature Tg or its gas composition (for example, its HNx content), a drift of one of the skin pass operating conditions, or a drift of the semi-finished product temperature at a predetermined point in the line (for example, the holding temperature or the temperature after cooling) compared to the desired reference temperature at this point. - Or a drift of one of the semi-finished product characteristics at the inlet of the processing line (a drift of the chemical composition or dimensions of the product, or a drift of past process parameters such as the reduction ratio of the previous cold rolling operation) is included.

[0063] Furthermore, in the embodiments described herein, the labeled training data with abnormal steel processing data is simulated training data generated by simulating what the line control signal output by the control module is when one of the abnormal causes exists.

[0064] For each of these simulated training datasets, the steel machining data is simulated steel machining data, and each of them is, - Chemical composition of steel semi-finished products CC i and, - One or more target characteristics P of the steel semi-finished product m , or one or more estimated final properties P of a steel semi-finished product FAIL,k,i,l,m and, -Abnormal cause AC k The virtual line control signal MP calculated assuming its existence FAIL,l Set and Includes.

[0065] The labels associated with these simulated processing data are those that indicate the cause of the anomaly AC. k Specify that it corresponds to.

[0066] Virtual line control signal MP FAIL,l Each set is similar to this one, -Chemical composition CC i and one or more target characteristics P m As a function of, -The calculation may be performed using the same steel prediction model and calculation rules used by the control module 11 to calculate the actual line control signal MP. oa) Virtual line control signal MP FAIL,l One of these values ​​is forced to be determined to a value corresponding to a partial or complete shutdown of one of the actuators 3 of the steel processing line 1 (i.e., a value that, when used to control the actuator, leads to a partial or complete shutdown of the actuator), or ob) Based on virtual monitoring signals, where at least one of the virtual monitoring signals is the corresponding reference monitoring signal MS REF It deviates from the norm.

[0067] As described above when explaining the control module 11, the control module is programmed to determine a line control signal, taking into account the constraint that one of the actuators 3 is stopped or partially stopped. This feature of the control module corresponds to a virtual line control signal MP in case a) above, which is a response to the whole or partial failure of one of the actuators. FAIL,l It is used to make a decision.

[0068] In case b), the reference monitoring signal MS REF This could be a signal output by a corresponding sensor, recorded during the operation of a steel processing line considered normal. In practice, the operation of the steel processing line may be considered normal if the operator in charge of the line determines, based on this experience and monitoring signals or other observations, that no abnormalities have occurred, or if the characteristics actually obtained at the end of the process match the target characteristics P.

[0069] Reference monitoring signal MS REF Furthermore, using the steel prediction model, the chemical composition CC i and one or more target characteristics P m The signal may be calculated based on the following: For example, if a possible monitoring signal is the temperature of the steel semi-finished product 6 measured by a pyrometer at the end of the cooling stage, the corresponding reference monitoring signal MS REF is, chemical composition CC i and one or more target characteristics P m Taking this into consideration, the target temperature at the end of cooling may be determined according to the heat path determined by the steel prediction model.

[0070] In the embodiments described herein, the trained classifier is trained and tested based on such simulated training data. The determination and collection of labeled training data, as well as the training and testing of the classifier, can be achieved automatically by a computer programmed for this purpose. Once trained, the trained classifier 12 is used in the steel processing line 1 to detect possible anomalies and identify the corresponding causes.

[0071] Figure 4 illustrates the steps performed in an exemplary manner to obtain simulated and labeled training data when the steel semi-finished product process includes heat treatment as previously presented.

[0072] Step 1 As shown in Figure 4, this method uses chemical composition CC i Step s1 includes collecting (i is an integer from 1 to x). Step s1 also includes collecting the format (dimensions) and / or other initial characteristics of the steel semi-finished product that can be processed by the steel processing line 1.

[0073] Chemical composition CC i The format and other initial characteristics are preferably selected within the raw product database, which collects data on products that have been pre-processed on the steel processing line 1.

[0074] An industrial production database, such as a raw product database, typically contains a vast amount of data on many different steel grades and formats.

[0075] Chemical composition CC i These data are selected to correspond to the same given steel grade. In practice, different steel semi-finished products of the same steel grade (e.g., different coils) will have individual chemical compositions that vary slightly from one semi-finished product to another, while remaining within the permissible chemical composition range (tolerance) corresponding to the possible steel grade. Chemical composition CC i The set is selected from the raw product database to represent this industrial variability in the chemical composition of steel semi-finished products having the same steel grade. Furthermore, by limiting the selection to a single predetermined steel grade, it is possible to significantly reduce the amount of data subsequently used to train the classifier.

[0076] The chemical composition CCi is, for example, i) In the raw product database, select the chemical composition of steel semi-finished products that all have the same steel grade and were processed on steel processing line 1 during one or more production campaigns, ii) From these chemical compositions, identify representative clusters, and if necessary, identify outlier chemical compositions and exclude them. iii) Select several representative clusters so as to homogeneously cover the range of chemical compositions of the chemical composition selected in step i), iv) From the representative clusters selected in step iii), chemical composition CC i Choosing and Selected by

[0077] Steps ii) to iv) may be parameterized to reduce the number of selected chemical compositions (between step i) and step iv) by, for example, five times or more.

[0078] Chemical composition CC i From the selected chemical compositions, the ratio of the content of the two alloying elements can be kept the same (in fact, such a ratio is usually constant for steel produced using the same production process).

[0079] For example, using the procedure described above to select only a few chemical compositions from the raw product database is highly beneficial because the raw product database typically contains a vast amount of data related to many different steel grades and formats. Therefore, calculating many different corresponding training data and then training the classifier while taking all of this into account would make the training process extremely slow.

[0080] In contrast, training data is collected for one of a given steel grades. Therefore, the trained classifier derived from this data is suitable for detecting anomalies when a processed steel semi-finished product 6 is of that same steel grade (in other words, the classifier is specific to one type of steel grade).

[0081] Selected chemical composition CC i The number x is, for example, 2 to 1000, more preferably 10 to 1000, and even more preferably 10 to 500.

[0082] The steel semi-finished product format may be selected using the same or similar methods as those presented in the chemical composition selection above.

[0083] In step s1, one or more target characteristics P m For example, they are selected from the same raw product database.

[0084] Step s2 Step s2 is to achieve one or more target characteristics P at the end of the heat treatment. m (For example, target tensile strength and target microstructure), or properties close to them, each chemical composition CC i and one or more target characteristics P collected in step s1 m Regarding the thermal path TP REF,i,m This includes calculating the thermal path TP. REF,i,m This is calculated using the same steel prediction model as that used in control module 11.

[0085] Thermal path TP REF,i,m This includes, for example, the continuous temperature, heating and cooling rates, and the time spent in each section of the heat treatment process.

[0086] The additional reference thermal path is also the thermal path TP calculated as described above. REF,i,m Therefore, it may be determined by slightly altering the heat path within a typical industrial variation range (taking into account typical variations in the heat path that actually occur, even in the absence of anomalies).

[0087] Step s3 Each heat path TP REF,i,m Regarding, and in some cases for each additional reference thermal path, the line control signal MP is used for the steel semi-finished product to follow such thermal paths. REF,jOne or more sets of (where j is an integer from 1 to y) are calculated. One or more line control signals MP REF,j The set is calculated using the same calculation rules as those used by the control module 11. y is the thermal path TP, as described above when describing the control module 11. REF,i,m To control the actuator to obtain the desired heat path TP, there may be different possible combinations, and it may be greater than 1. In practice, with respect to the cooling jet, the desired heat path TP REF,i,m There are numerous possible combinations for controlling the jets to obtain (in other words, many different possibilities for distributing the total cooling force required among the jets). For example, J1=J2=J3=90%, but it could also be J1=85%, J2=90%, J3=95%, or J1=85%, J2=92%, J3=93%, etc. Typically, with 11 cooling jets, the desired heat path TP REF,i,m There are hundreds of slightly different combinations that allow tracing to be performed, resulting in extremely close final characteristics, for example, yielding the same tensile strength value, with variation from one jet combination to another of less than 0.05%. In other words, for a given target thermal path, there are many different jet configurations corresponding to normal operation. Therefore, line control signal MP REF,j And many such possible sets of corresponding training data are very beneficial in terms of classification accuracy. In fact, it helps the classifier distinguish jet force fluctuations corresponding to drift from fluctuations corresponding to one of many possibilities of normal operation. Thus, in practice, y may be selected to be very high (higher than 10, or even higher than 20 or 100). For example, in the case of an ensemble of 11 cooling jets, y may be in the range of 20 to 2000, more preferably 100 to 1000.

[0088] Step 4 Next, in step s4, each chemical composition CC i And, manufacturing parameter MP REF,jFor each combination, one or more estimated final characteristics PREF,i,j,m (typically including the obtained microstructure MREFi,j,m) expected at the end of the heat treatment are calculated (using the same calculation rules as those used by the control module), thereby defining the reference state.

[0089] Step 5 Step s5 is the above abnormal cause AC k This involves obtaining a predetermined list of (where k is an integer from 1 to z). In a preferred embodiment, z is from 2 to 100, more preferably from 10 to 50, and even more preferably from 20 to 50. This list of possible abnormal causes, or in other words, possible failure scenarios, is intended to be as exhaustive as possible. However, to account for drifts or failures that have been forgotten or not observed, abnormal causes called “unknown causes” may be added to the list.

[0090] Step 6 Each abnormality cause AC k Regarding this, step s6 simulates the anomaly by issuing a corresponding abnormal virtual line control signal MP FAIL,l This involves calculating at least one set of (where l is an integer from 1 to a) anomalous virtual line control signals MP FAIL,l This is calculated and described above in the general presentation of the anomaly detector 12. a is equal to or of a similar magnitude to y. In particular, a may be greater than 10, or even greater than 20 or 100.

[0091] Step 7 Step s7 is for each abnormal cause AC k , composition CCi, and abnormal virtual line control signal MP FAIL,l For each set, this involves calculating one or more estimated final properties PFAILk,i,l,m (typically including the microstructure MFAILk,i,l,m resulting from this processing). The calculation rules used for this purpose may be the same as those used in step s4.

[0092] Steps s8 and s9 In step s8, the results obtained in step s4 are collected in the form of labeled training data, each of which, among other things, has a chemical composition CC. i One of them, estimated final characteristic P REF,i,j,m One of the following, and line control signal MP REF,j One of the corresponding sets is collected, and the training data is labeled as normal. Each of these labeled training data is also the aforementioned reference monitoring signal MS REF This may include one or more of the following (which actually represent signals measured on the line): Here, these reference monitoring signals MS REF This is the reference thermal path TP determined in step s2. REF,i,m This includes at least some of the consecutive temperature points. These labeled training data are collected in a database. This database may be completed based on training data which are actual production data recorded during the operation of steel processing line 1, as well as here.

[0093] In step s9, the results obtained in step s7 are collected in the form of labeled training data, each of which, among other things, has a chemical composition CC. i One of them, estimated final characteristic P FAIL,k,i,l,m One of the following, and an abnormal virtual line control signal MP FAIL,lj The corresponding sets are collected, labeled training data are labeled as anomalies, and the cause of the anomaly is AC k Each of these labeled training data may also include one or more of the virtual monitoring signals described above (which actually represent signals measured on the line in the event of a malfunction or drift). These labeled training data are also collected in the database described above. This database may also be completed with actual production data, if available, recorded during the operation of steel processing line 1 that is suspected to be abnormal.

[0094] Steps 10 In step s10, the classifier is trained and tested using pre-obtained labeled training data.

[0095] The classifier may be a soft classifier, in which the probabilities associated with each class (i.e., each abnormal cause and the normal case) are estimated, and the selected class is based on these estimated probabilities.

[0096] A portion of the database that collects labeled training data is used as a training database to train a classifier to classify information corresponding to normal or abnormal processing conditions. A class associated with the normal operation of the criterion may be class 0. Preferably, 70% to 80% of the database is used as the training database.

[0097] Next, the remainder of the database is used as a test database to determine the accuracy of the trained classifier. Preferably, 20% to 30% of the database is used for such testing. Preferably, the validation used is stratified cross-validation, which means that this validation ensures that the class distribution is the same in the training and test databases used.

[0098] To evaluate the accuracy of the classifiers in classifying information into the correct class, the alpha and beta risks of each classification tool are calculated for each class: - Risk alpha (α) represents the probability of classifying an abnormal control signal, which could be considered a false negative, as normal. Therefore, a value of (1-α) = 1 or close to 1 corresponds to a good classification. -Risk beta (β) represents the probability of classifying a normal control signal as an abnormal control signal, which could be considered a false positive. Therefore, a value of (1-β) = 1 or close to 1 corresponds to good classification.

[0099] The precision, expressed as a percentage, is calculated as the average of the (1-α) or (1-β) values ​​across different classes.

[0100] The classifier type is, Decision trees (dt), especially the C4.5 algorithm, random forests (rf), extra trees (et), adaptive boosting (ab), Gaussian Knives Bayes (gnb), or support vector machines (SVM) A variety of classifiers, including [specific types of classifiers], can be selected.

[0101] The inventors have observed that decision tree-based classifiers (and in some cases multi-tree classifiers) perform well for the applications related here, and in particular, they perform better than support vector machines.

[0102] The inventors also observed that a “pre-induced” multi-tree classifier provides good classification accuracy. “Pre-induced” means a classifier based on multiple basic decision trees, where each basic decision tree has the same anomaly cause AC. k It is trained using a set of labeled training data that has this characteristic. Therefore, this basic decision tree has this characteristic AC k It works very well to detect (though it doesn't do so well for other anomalies, the overall performance of the classifier is achieved by combining decisions made by different base trees). This pre-guided training method can be implemented, for example, by appropriately selecting the training data used for each tree during the training of a forest-like classifier.

[0103] Here, the same classifier is used to detect different causes of anomalies. Alternatively, separate, trained classifiers, each dedicated to one (or a few) causes of anomalies, can be used to detect different causes.

[0104] After this initial setup phase is completed, the classifier can then be used on steel processing line 1 to detect any potential anomalies.

[0105] countermeasure The electronic device 10 of the steel processing line 1 may be programmed, for example, to determine the severity indicator when the abnormality indicator ind indicates that the line control signal MP is abnormal. The severity indicator is determined by the abnormality cause AC k Determined based on, and in some cases, one or more estimated final properties P of the steel semi-finished product 6. EST It is also based on the difference between and one or more target characteristics P. The severity indicator specifies whether the anomaly is not critical, critical, or urgent.

[0106] Non-critical anomalies include, for example, malfunctions or drifts detected, but affecting the final characteristics of the steel semi-finished product, particularly P. EST This addresses situations where deviation from the acceptable compliance range is not expected.

[0107] A serious abnormality could be, for example, the final properties of a steel semi-finished product, particularly P EST This addresses situations where a deviation from the acceptable compliance range is anticipated, but subsequent repair or online compensation of the steel semi-finished product is deemed possible.

[0108] An urgent anomaly would be addressed, for example, when a steel semi-finished product is deemed unsuitable and irreparable.

[0109] The electronic device may be further configured to perform the following steps: -In the case of a non-critical anomaly, a warning message will be issued indicating the cause of the anomaly and prompting the user to correct the malfunction or drift cause during the next planned line stop. -In the event of a serious anomaly, a serious warning message is issued prompting the user to adjust the parameters of subsequent machining (or even the parameters of the machining in progress) to compensate for the anomaly, or the parameters of the machining are automatically adjusted to compensate for the anomaly, and -In the event of an emergency or malfunction, the steel processing line 6 will be stopped.

[0110] Next, the present invention will be further explained by the following embodiments, but these are by no means limiting.

[0111] Examples To obtain a steel sheet having a chemical composition CC containing 0.15% by weight of C, 1.90% by weight of Mn, 0.20% by weight of Si, 0.20% by weight of Cr, 0.003% by weight of Mo, 0.017% by weight of P, 0.005% by weight of N, 0.015% by weight of Cu, and 0.023% by weight of Ti, with the remainder of the composition being iron and unavoidable impurities resulting from refining, the following sequential steps are taken: a. Casting step of semi-finished product, b. Reheating step, c. Hot rolling step, d.Temperature T coil The cooling step, e. Reduction Speed ​​CR rate The cold rolling step and f. A preheating step in which the steel plate is heated to a temperature T1, g. A heating step in which the steel plate is heated from temperature T1 to temperature T2, h. A holding step in which a steel plate is held at a temperature T2 for a holding time t, and the holding ends at a temperature T3. i. A cooling step in the snout, in which the steel plate is cooled from T3 to snout temperature T4 using 11 cooling jets that spray HNx, j. The rate v in a zinc bath having a temperature T5 corresponding to isothermal holding at such a temperature GAL The hot-dip plating coating step, k. Cooling step to top roll temperature T6, l. Cooling step at room temperature and It must be produced in accordance with [the specified method / standard].

[0112] The heat path consists of steps f to l, with the first step being implemented in advance on a separate production line.

[0113] Step 1: Collection of chemical composition data Table 1 shows 15 chemical compositions CC selected for the steel grades considered here, according to the method described above with reference to Figure 4. i The following are being collected. The elemental content of these chemical compositions is expressed in weight percent (W%). In this example, only one set of target properties (TS=820MPa; YS=500MPa; and the microstructure defined above) is considered.

[0114] [Table 1]

[0115] Step s2: Defining the reference heat path Chemical composition CC collected in Step 1 i For each sheet, a reference thermal path is determined according to the method presented above with reference to Figure 4, which enables obtaining a microstructure and properties as close as possible to the target at the end of the heat treatment. These thermal paths include temperature, time, and heating and cooling rates. Table 2 shows each CC i The end temperature of each stage is collected. The heating and cooling rates vary (slightly) depending on the different sheet formats possible for each chemical composition, so for brevity, they are not explicitly shown in Table 2.

[0116] [Table 2]

[0117] Step s3: Line control signal MP that enables obtaining the heat path REF,i,j decision Step s3 is achieved as described above with reference to Figure 4, and the value of y falls within the range of 500 to 700.

[0118] Step 4: Calculation of base MREFi,j and PREFi,j Each CC i For combinations of jet forces, the reference properties PREFi,j (including reference microstructure MREFi,j) obtained at the end of the heat treatment process, free from jet failures or other anomalies, are then calculated as described above. These properties are the tensile strength TS, yield strength YS, and microstructure (defined as the respective surface fractions of ferrite, bainite, martensite, and possibly pearlite).

[0119] Next, steps s5 to s10 were performed as described above, referring to Figure 4. Here, 11 possible causes of abnormality were considered, each associated with a failure (complete shutdown) of one of the 11 cooling jets of the cooling system.

[0120] Each of the "normal" labeled training data collected in this manner is classified as having a chemical composition CC. i One of them, related line control signal MP REF,i,j One of the following, and the corresponding estimated final characteristic PREFi,j, as well as the reference thermal path TP REF,i This also includes temperature setting points T1 to T6 (these setting points serve as line monitoring signals measured by sensors on the actual processing line).

[0121] The training data labeled "abnormal" has the same structure as the training data labeled "normal".

[0122] The classifier is configured to classify data into 12 classes. Class n=0 corresponds to normal line control signals (and therefore normal operation). Classes n=1 through n=12 each correspond to a failure of cooling jet number n.

[0123] The accuracy obtained after training is collected in Table 3 for both the Random Forest Classifier (RF) and the Extra Tree Classifier (ET). Training was performed using 70% of the database, and accuracy testing was performed using 30% of the database. The database contains approximately 45,000 different training data for each class (different classes are represented evenly within the database). For class n=0, the data in the database includes actual production data recorded during line operation and simulated data. In Table 3, a quantity of "1" indicates a quantity of 0.999 or greater.

[0124] [Table 3]

[0125] These classifiers exhibit excellent classification results. Tests conducted with dt(C4.5), ab, and gnb also show excellent classification results.

Claims

1. A method for monitoring a steel processing line (1) during the processing of a steel semi-finished product (6), wherein the steel processing line (1) comprises a sensor (2), an actuator (3), and an electronic device (10) which includes a control module (11) and an anomaly detector (12), - In the method, the control module, o The following: - Chemical composition CC of steel semi-finished products, - The target characteristics P of the steel semi-finished product obtained at the end of the above processing, - A monitoring signal (MS) output by the sensor (2) of the steel processing line, wherein the operating conditions (T ｇ ) or intermediate properties of steel semi-finished products (T ｓ , T ｓ’ The monitoring signal (MS) representing ) is obtained, The line control signal (MP) for controlling the actuator (3) of the steel processing line is determined, and the line control signal is determined as a function of the chemical composition CC, target characteristic P, and monitoring signal (MS), and is determined using a steel property prediction model. Based on the line control signal (MP), the actuator (3) is controlled. - The abnormality detector (12) determines and outputs an abnormality indicator (ind) that specifies whether the line control signal (MP) is normal or abnormal, and then identifies the abnormality cause (AC) selected from a predetermined list of abnormality causes. ｋ ) is specified, and an anomaly indicator (ind) is determined using a trained classifier, and the input to the trained classifier is, The chemical composition CC of the semi-finished steel product, o Target characteristic P, or estimated final characteristic (P) of the steel semi-finished product determined by the control module (11) ＥＳＴ )and, o-line control signal (MP) and The trained classifier includes several labeled training data, and each labeled training data is, o Steel processing data of the same type as the input to the trained classifier, A label that specifies whether steel processing data is normal or abnormal, and then one of the abnormality causes from the list is specified, and Methods that include...

2. The specified list of abnormal causes is as follows: - One of the actuators (3) of the steel processing line is partially or completely inoperable. - A drift occurs in one of the operating conditions or one of the intermediate characteristics of the steel semi-finished product, resulting in one of the monitoring signals deviating from the reference monitoring signal. The method according to claim 1, comprising one or more of the following.

3. For some of the labeled training data, - The steel machining data is simulated steel machining data calculated while one of the abnormality causes is present. - The label associated with the steel processing data indicates that the cause of the abnormality exists. The method according to claim 1 or 2.

4. Each simulated steel machining data, - Chemical composition CC of steel semi-finished products ｉ and, - Target characteristic P of the steel semi-finished product ｍ , or estimated final properties P of the steel semi-finished product ＦＡＩＬ，ｋ，ｉ，ｌ，ｍ and, - A set of virtual line control signals (MP ＦＡＩＬ，ｌ ) calculated while taking into account the presence of the abnormal cause and The method according to claim 3, including the method described in claim 3.

5. Virtual line control signal (MP) ＦＡＩＬ，ｌ Each set of ) o Chemical composition CCi and target characteristic P ｍ As a function of, The line control signal (MP) is calculated using the same steel prediction model and calculation rules used by the control module (11) to calculate the line control signal (MP), and; • Virtual line control signal (MP) ＦＡＩＬ，ｌ ) is determined by forcing one of them to a value corresponding to the partial or complete shutdown of one of the actuators in the steel processing line, or Based on the virtual monitoring signals, at least one of the virtual monitoring signals is out of sync with the corresponding reference monitoring signal. The method according to claim 4 and claim 2.

6. The reference monitoring signal uses the steel prediction model to determine the chemical composition CC ｉ and target characteristic P ｍ The method according to claim 5, calculated based on the following:

7. Chemical composition CC ｉ However, by selecting products listed in the raw products database that correspond to the same steel grade, data on products pre-processed on the steel processing line is collected within the raw products database (s1), and the selected products have a chemical composition that is homogeneously distributed within the range of chemical compositions permissible for the steel grade. The method according to any one of claims 4 to 6.

8. The steel semi-finished product undergoes heat treatment, and the determination of simulated steel processing data is performed in the following steps: -s2: Each chemical composition CC ｉ and target characteristic P ｍ Regarding the thermal path TP ＲＥＦ，ｉ，ｍ Calculate the target characteristic P at the end of the heat treatment. ｍ A step to obtain, wherein the calculation is based on the steel prediction model, -s3: The heat path TP ＲＥＦ，ｉ，ｍ A step of determining one or more sets of line control signals MPj so as to follow a path, wherein j is an integer from 1 to y, -s4: Each chemical composition CC ｉ and line control signal MP ＲＥＦ For the set of ,j, the step is to calculate the estimated final characteristic PREF,i,j,m obtained at the end of the heat treatment, -s5: Abnormality cause AC ｋ A step of obtaining the predetermined list, wherein k is an integer from 1 to z, -s6: Each abnormality cause AC ｋ Regarding the corresponding abnormal virtual line control signal MP ＦＡＩＬ，ｌ A step of simulating the anomaly by calculating at least one set of, where l is an integer from 1 to a, -s7: Each abnormality cause AC ｋ , composition CC ｉ , and abnormal virtual line control signal MP ＦＡＩＬ，ｌ For the set, the steps include calculating the estimated final characteristics PFAIL, k, i, l, m expected at the end of the heat treatment, - s8: A step of collecting the results obtained in step s4 in the form of labeled training data, each having at least the chemical composition CC ｉ One of the following: one of the estimated final characteristics PREF, i, j, m, and line control signal MP ＲＥＦ， The steps include collecting the corresponding set of j and labeling it as normal steel machining data, - s9: A step of collecting the results obtained in step s7 in the form of labeled training data, each of which has at least the chemical composition CC ｉ One of the following: one of the estimated final characteristics PFAIL, k, i, l, m, and an abnormal virtual line control signal MP ＦＡＩＬ，ｌ The corresponding set of j is collected, labeled as abnormal, and the cause of the abnormality is AC ｋ The steps and The method according to any one of claims 4 to 7, including the method described in any one of claims 4 to 7.

9. The method according to claim 8, wherein y and a are each greater than 10, or further greater than 20.

10. The method according to any one of claims 1 to 9, wherein the input to the trained classifier further includes one or more monitoring signals (MS).

11. The method according to any one of claims 1 to 10, wherein the trained classifier is a decision-tree based classifier.

12. The method according to claim 11, wherein the trained classifier is based on a plurality of basic decision trees, each of which is trained using a set of labeled training data having the same anomaly cause.

13. The method according to any one of claims 1 to 12, wherein the control module (11) determines a line control signal (MP) also taking into account one or more process parameters (PPP) of a prior process that the steel semi-finished product (6) underwent before being processed in the steel processing line (1).

14. When the abnormality indicator (ind) indicates that the line control signal (MP) is abnormal, the severity indicator then indicates the cause of the abnormality (AC). ｋ ) and estimated final properties (P ＥＳＴ The severity indicator is determined based on the following criteria, and it indicates whether the anomaly is not critical, critical, or urgent. - In the case of a minor anomaly, a warning message will be issued indicating the cause of the anomaly and prompting repairs during the next scheduled line stop. - In the event of a serious anomaly, a critical warning message is issued prompting the user to adjust the parameters of the subsequent machining process to compensate for the anomaly, or the parameters of the subsequent machining process are automatically adjusted to compensate for the anomaly. - In the event of an emergency or malfunction, the steel processing line (6) will be stopped. The method according to any one of claims 1 to 13.

15. The method according to any one of claims 1 to 14, wherein the steel processing line includes one of the following, or a combination thereof, a furnace, a mill, a runout table, a hot rolling line, a cold rolling line, a pickling line, a heat treatment facility, a hot-dip galvanizing line.

16. The method according to any one of claims 1 to 15, wherein the steel semi-finished product (6) is a steel plate.

17. A steel processing line (1) comprising a sensor (2) and an actuator (3), and an electronic device (10) for controlling the steel processing line (1) suitable for processing steel semi-finished products (6), wherein the electronic device (10) comprises a control module (11) and an anomaly detector (12), - The control module performs the following steps: o is a step to obtain: • Chemical composition of steel semi-finished products CC, - Target characteristics P of the steel semi-finished product obtained at the end of the above processing, - A monitoring signal (MS) output by a sensor in a steel processing line, wherein the operating conditions (T ｇ ) or intermediate properties of steel semi-finished products (T ｓ The steps include obtaining a monitoring signal (MS) that represents ), The steps include determining a line control signal (MP) for controlling an actuator in a steel processing line, wherein the line control signal is determined as a function of chemical composition CC, target characteristic P, and monitoring signal (MS), and is determined using a steel property prediction model. It is configured to perform the steps of controlling an actuator based on a line control signal, - The abnormality detector (12) is configured to determine and output an abnormality indicator (ind) that specifies whether the line control signal (MP) is normal or abnormal, and then a predetermined abnormality cause (AC) is determined. ｋ The selected anomaly cause is specified in the list of ), the anomaly indicator (ind) is determined by the trained classifier, and the input to the trained classifier is o Chemical composition of steel semi-finished products CC, o Target characteristic P, or estimated final characteristic (P) of the steel semi-finished product determined by the control module ＥＳＴ ), and o Line control signal (MP), - A trained classifier is a classifier that has been pre-trained using several labeled training data, where each labeled training data is o Steel processing data of the same type as the input to the trained classifier, and A label that specifies whether steel processing data is normal or abnormal, and then one of the abnormality causes in the list is specified. Electronic device (10), including

18. A steel processing line (1), comprising a sensor (2), an actuator (3), and an electronic device (10) as described in claim 17.

19. A method for training a classifier of an anomaly detector (12) of an electronic device (10), wherein the electronic device is a steel processing line (1) comprising a sensor (2) and an actuator (3), and comprising a control module (11) for controlling the steel processing line (1) suitable for processing steel semi-finished products (6), - The control module performs the following steps: o is a step to obtain: • Chemical composition of steel semi-finished products CC, - Target characteristics P of the steel semi-finished product obtained at the end of the above processing, - A monitoring signal (MS) output by a sensor in a steel processing line, wherein the operating conditions (T ｇ ) or intermediate properties of steel semi-finished products (T ｓ The steps include obtaining a monitoring signal (MS) that represents ), The steps include determining a line control signal (MP) for controlling an actuator in a steel processing line, wherein the line control signal is determined as a function of chemical composition CC, target characteristic P, and monitoring signal (MS), and is determined using a steel property prediction model. It is configured to perform the steps of controlling an actuator based on a line control signal, - The classifier is trained using several labeled training data, and each labeled training data is, o Steel processing data of the same type as the input to the trained classifier, and A label that specifies whether steel processing data is normal or abnormal, and then a label that specifies one of the abnormality causes in the list, A method wherein at least some of the labeled training data are simulated training data calculated using the same steel prediction model and the same calculation rules as those used by the control module to calculate the line control signal, while one of the anomaly causes is present.

20. A computer program including instructions, wherein when the instructions are executed on a computer device connected to sensors and actuators of a steel processing line, the computer program causes the computer device to perform the method according to claim 1.

Images