Method for determining the installation heights of horizontally oriented optical fiber sensors in a continuous casting mold and a correspondent data model, method for continuous casting and continuous casting mold

Abstract

The invention relates to the determination of installation heights (h1,...,hN) for horizontally oriented optical fiber sensors (11) for detecting temperatures in a continuous casting mold (10) together with a corresponding data model (100). Using a reference mold (20) which has a plurality of second sensor points (22) arranged in several rows (Rj) for detecting temperatures, instantaneous production parameters and second temperature values ​​(Ti,j) are detected as input variables (E) during continuous casting at a plurality of time intervals (Δt), and instantaneous meniscus heights (Mi) are determined from these as output variables (A). For each row (Rj), a correlation coefficient (Kj) averaged over all time intervals (Δt) between the second temperature values ​​(Ti,j) and the meniscus heights (Mi) is determined. The installation heights of those series (Rj1,..., RjN) with the largest averaged correlation coefficients (Kj) are referred to as installation heights (h1,...,hN) for the continuous casting mold (10) and the data model (200) is trained on the basis of the recorded input variables (E) and output variables (A).

Description

[0001] Method for determining the installation heights of horizontally oriented optical fiber sensors in a continuous casting mold and a corresponding data model, method for continuous casting and continuous casting mold with corresponding data model.

[0002] The invention relates to a method for determining installation heights for horizontally oriented optical waveguide sensors in a continuous casting mold and a corresponding data model according to claim 1, a continuous casting mold with a corresponding data model according to claim 10 and a method for continuous casting according to claim 12.

[0003] In continuous casting with continuous casting molds, both downward and upward flow patterns (e.g., in the form of a so-called 'double roll') typically form in the introduced metallic melt, which have a significant influence on the quality of the cast product. In the interest of increasing quality requirements, efforts are therefore made to capture these flows as precisely as possible in order to implement appropriate control measures in the continuous casting process if necessary.

[0004] For example, upward currents cause an exaggeration of the meniscus profile in the mold in certain areas, with this exaggeration depending on the respective flow velocity. Therefore, such an exaggeration (or the corresponding meniscus profile) can indirectly be used to infer the instantaneous flow velocity, which in turn allows for significant conclusions regarding the quality of the cast product.

[0005] The integration of horizontally or vertically oriented fiber optic sensors into the side walls of continuous casting molds is known, for example, from WO 2019 / 180229 A2 and EP 3 639 949 A1. These fiber optic sensors each have multiple sensor points (e.g., in the form of so-called fiber Bragg gratings) for measuring temperatures. This allows temperature values ​​to be measured simultaneously at several locations within the mold plates, enabling, for example, a two-dimensional measurement of the temperature distribution of the solidifying strand shell within the mold and the detection of strand shell defects.

[0006] Furthermore, the determination of a meniscus profile in a continuous casting mold using such a sensor arrangement is known, for example, from the article "Development and Application of Fiber Bragg Gratings for Slab Casting" by Spierings, Kamperman, Hengeveld, Kromhout and Dekker, pages 1655-1663 in 'AISTech 2017 Proceedings' or from the article "Online Flow Control with Mold Flow Measurements and Simultaneous EM Braking and Stirring" by M. Sedén and N. Jacobson in '2018 IOP Conf. Ser.: Mater. Sci. Eng. 424 012015', DOI 10.1088 / 1757-899X / 424 / 1 / 012015.

[0007] A disadvantage of such installations is the relatively high installation effort for the fiber optic sensors, because on the one hand, corresponding recesses must be created in the relevant mold plates, which involves corresponding manufacturing costs. On the other hand, the fiber optic cables require correspondingly elaborate protective measures against the harsh production conditions of a continuous casting plant, as well as expensive optical connections to the respective evaluation units (so-called interrogators).

[0008] It is therefore an object of the invention to reduce the costs of operating such installations on fiber optic sensors in continuous casting molds, by means of which a meniscus profile can be determined during continuous casting, while at the same time maintaining at least approximately the accuracy with which the meniscus profile can be determined.

[0009] The task is essentially solved by limiting the number of horizontally oriented fiber optic sensors to a few, optimized installation heights, and incorporating a trained data model. Due to its (trained) predictive capabilities, the data model is able to determine the meniscus profile with comparable accuracy to that achieved with a significantly larger number of fiber optic sensors, based on the limited number of available sensor points.

[0010] Specifically, the problem is solved by the method according to independent claim 1 for determining installation heights for horizontally oriented optical fiber sensors in a continuous casting mold and a corresponding data model. Claims 2 to 9 describe advantageous embodiments of the method according to the invention.

[0011] Furthermore, the problem is solved by a continuous casting mold with a corresponding data model according to independent claim 10 and a method for continuous casting according to independent claim 12. Claims 11 and 13 to 15 again constitute advantageous embodiments.

[0012] According to the inventive method, one or more installation heights h1,...,hN for horizontally oriented fiber optic sensors in a continuous casting mold, as well as a corresponding data model, are determined. The fiber optic sensors can be arranged on at least one side plate of the continuous casting mold. The installation heights h1,...,hN are defined with respect to a reference level of the continuous casting mold (e.g., an upper edge of the mold plates). Multiple fiber optic sensors can also be arranged at the same installation height, for example, one fiber optic sensor at the same installation height in each of several plates of the continuous casting mold.

[0013] The fiber optic sensors each have multiple sensor points for measuring temperatures within the continuous casting mold. Fiber optic sensors offer the advantage over conventional thermocouples that the spacing between individual sensor points can be freely selected and, for example, can be as small as a few millimeters (e.g., 3-5 mm). This allows for very precise temperature monitoring of the continuous casting mold. In contrast, thermocouples are generally limited to fixed positions—for example, the positions of existing retaining bolts in the mold—which consequently prevents very fine spatial temperature monitoring.

[0014] The installation heights h1,...,hN of the optical fiber sensors are determined using a reference mold that has a plurality of secondary sensor points. These secondary sensor points are arranged in a grid pattern at several – pairwise distinct – horizontal installation positions xi in multiple horizontal rows Rj at a respective installation height hj within the reference mold and are configured for temperature measurement. Specifically, each horizontal row Rj, comprising several secondary sensor points, is arranged at a respective installation height hj relative to a reference level of the reference mold. A row Rj does not necessarily have to be configured as a physically continuous, horizontal optical fiber with multiple optical fiber sensors, but is simply defined by the arrangement of a plurality of optical fiber sensors at the same installation height hj.

[0015] Since the second sensor points can be designed, for example, as identical Fiber Bragg Gratings in vertically arranged optical waveguides, a grid of measuring points with horizontal spacings dh and various vertical spacings h1 to hN is obtained. This grid, created by the measuring points of the vertical optical waveguides, can also be described as a grid of fictitious horizontal optical waveguides, where the horizontal optical waveguides are then arranged in horizontal rows Rj with a respective installation height hj, and the measuring points of these fictitious optical waveguides are located at the horizontal positions xi.

[0016] For example, in the reference mold, in the area of ​​the pouring level (e.g., on a vertical section 100 mm high in the area of ​​a nominally assumed pouring level), the second sensor points are arranged in several horizontal rows, with the horizontal rows spaced 5 mm apart vertically. In an area further away from the pouring level, the vertical spacing of the horizontal rows can be greater. For example, the second sensor points are spaced 30–100 mm apart horizontally.

[0017] The second sensor points are preferably arranged in at least five horizontal rows Rj, with each row Rj comprising at least ten, preferably at least fifteen, differently paired horizontal installation positions xi for the second sensor points. This advantageously reduces the number of first sensor points in the continuous casting mold compared to the multitude of second sensor points in the reference mold, resulting in reduced procurement and installation costs for the fiber optic sensors and the corresponding evaluation device.

[0018] The second sensor points can also be optical fiber sensors. For example, the rows Rj in the reference mold can each be configured as an optical fiber sensor with multiple sensor points. Alternatively, the second sensor points can also be configured as individual thermocouples arranged in a grid pattern within the reference mold.

[0019] During continuous casting with the reference mold, instantaneous production parameters are recorded at numerous time intervals Δt (i.e., the recorded production parameters may vary over time and are always assigned to the respective time interval Δt). Furthermore, during continuous casting with the reference mold, instantaneous second temperature values ​​Ti,j are recorded at the horizontal installation position xi and the installation height hj of the series Rj of the respective sensor point in the reference mold at the aforementioned numerous time intervals Δt using the second sensor points. This results in a temperature distribution with high spatial resolution being recorded for each time interval due to the numerous second sensor points arranged in a grid pattern within the reference mold.

[0020] From the recorded instantaneous second temperature values ​​Ti,j, instantaneous meniscus heights Mi (i.e., assigned to the respective time interval Δt) are determined at the horizontal installation positions xi of the respective second sensor points. Specifically, a meniscus height Mi at a particular horizontal installation position xi is determined based on the set of temperature values ​​Ti,j that were recorded at the same installation position xi at different installation heights hj. The meniscus heights are each referenced to the level of the continuous casting mold, e.g., to the upper edge of a side plate of the continuous casting mold.

[0021] The determination of meniscus heights is based, for example, on the vertical temperature gradients of the recorded secondary temperature values ​​Ti,j and is known, for example, from CN 118204469 A. The determination of the instantaneous meniscus heights Mi is advantageously possible with high accuracy because the aforementioned grid-like arrangement of the secondary sensor points in the reference mold allows for high spatial resolution in the vertical direction. Within the scope of the invention, the instantaneous meniscus heights Mi represent the best approximation of the actual meniscus profile in the mold.

[0022] Furthermore, for each time interval Δt and for each series Rj, an instantaneous correlation coefficient Kj – i.e., assigned to the respective time interval Δt – is determined between the (instantaneous) second temperature values ​​T i,j of the respective series Rj and the instantaneous meniscus heights M i. In other words, the recorded second temperature values ​​T i,j of the sensor points of the respective (horizontally oriented) series Rj are correlated with the set of instantaneous meniscus heights M i in the reference mold.

[0023] The correlation coefficient Kj can preferably be calculated using the known formula K j = ∑ i = 1 N j T i , j − T j ¯ ⋅ M i − M ¯ / ∑ i = 1 N j T i , j − T j ¯ 2 ⋅ ∑ i = 1 N j M i − M ¯ 2 to be determined with T j ¯ = 1 N j ⋅ ∑ i = 1 N j T i , j M ¯ = 1 N j ⋅ ∑ i = 1 N j M i where Nj denotes the number of second sensor points of the respective row Rj of the reference mold. The expression T j In equation (2), denotes an average value of the instantaneous second temperature values ​​T i,j of the respective series, calculated over a given series. The expression Min equation (3) denotes an average of the instantaneous meniscus heights M i .

[0024] Furthermore, in the inventive method, for each series R j, a correlation coefficient averaged over all time intervals Δt is calculated from all instantaneous correlation coefficients K j. K j determined. This temporal averaging highlights those series R j or gives greater weight to their contributions whose correlation with the actual meniscus shape is particularly representative for the different production parameters.

[0025] Subsequently, from the series R j of the reference mold, those N series R j1 ,..., R jN are selected that have the largest averaged correlation coefficients. K j exhibit. The (to be determined) installation heights h 1 ,...,h N subsequently correspond to the installation heights of the selected series R j1 ,..., R jN and are equated to them.

[0026] Furthermore, for the selected series B j1 ,..., R jN, a data set D, comprising input variables E and output variables A, is created for each time interval Δt. The input variables E include the instantaneous second temperature values ​​T i,j of the respective time interval Δt of the respective series R j1 ,...,R jN, as well as the instantaneous production parameters in the respective time interval Δt. The output variables A include the instantaneous meniscus heights M i of the respective time interval Δt.

[0027] Finally, the data model is trained using the data sets D. Subsequently, to predict the meniscus shape, the trained data model no longer needs to be fed the temperature values ​​of all second sensor points of the reference mold, but only the temperature values ​​of those sensor points located in the reference mold at the determined N installation heights h1,...,hN. The trained data model is thus adapted to casting processes with the respective reference mold.

[0028] Training the data model in the aforementioned sense is known from the prior art and is not part of the invention. During this training, internal model parameters of the data model are adjusted. Particularly in the case of data models based on so-called 'machine learning' (so-called ML models), training and validation datasets that are disjoint from each other are selected from the available datasets D. The adjustment of the internal model parameters is carried out using the training datasets, followed by validation using the validation datasets, the latter serving to verify the 'trained' predictive capability of the model. Such a procedure is known, for example, from EP 4 124 398 B1.

[0029] According to a preferred embodiment of the method according to the invention, only a single installation height h1 is determined, in which optical fiber sensors can be arranged in a continuous casting mold (not identical to the reference mold). The installation height h1 corresponds to the position of that row Rjmax of second sensor points which exhibit the highest averaged correlation coefficient. K jmax all rows are present. This allows the installation effort for the optical fibers in the mold to be kept particularly low.

[0030] In a further preferred embodiment of the method according to the invention, the number of installation heights is determined such that the averaged correlation coefficients K j of the selected series R j each lie within a given quantile q of the series R jmax that has the largest averaged correlation coefficient K jmax all series Rj exhibit a particularly favorable quantile q of 30%. The quantile q of the series Rjmax defines the range of values ​​around the average correlation coefficient. K jmax , within which the percentage of instantaneous correlation coefficients K j of the respective series R j, determined by the quantile q, lie.

[0031] By specifying a quantile q, several installation heights for horizontally oriented fiber optic sensors are typically determined. The correlation between these heights and the current meniscus heights Mi is stronger the lower the value of the quantile q is chosen. In the case of multiple installation heights, the accuracy of determining the actual meniscus path is advantageously increased compared to selecting only a single installation height (corresponding to N = one). Furthermore, the installation effort is reduced compared to a reference mold because the fiber optic sensors are only installed at those heights for which the most relevant correlations with the actual meniscus path are expected.

[0032] According to a further preferred embodiment of the inventive method, the production parameters include a casting speed vc of the reference mold and / or an immersion depth L c of a casting tube into a melt introduced into the reference mold and / or an argon flow f Ar of the casting tube (measured, for example, in liters per minute), wherein the argon is blown into the molten metal in the distributor or directly into the casting tube to increase the steel quality or as a protective agent to prevent caking. The immersion depth L c is defined as the distance between the outlet openings of the casting tube and a (actual or predetermined) casting level at a specific point in the mold. The casting level is, for example, referenced to an upper edge of the mold plates.

[0033] For example, the casting level - viewed in the width direction of the mold - is continuously determined at the position of the casting tube using one or more additional sensors, and the amount of melt introduced during continuous casting is regulated in such a way that the immersion depth L c remains constant.

[0034] According to a further, particularly preferred embodiment, the continuous casting mold has an electromagnetic brake for generating a specific flow pattern of a melt introduced into the reference mold, and the production parameters include - alternatively or additionally - setting values ​​bc for the electromagnetic brake (for example, a current or several currents if the electromagnetic brake includes several independent electrical circuits).

[0035] The aforementioned production parameters – casting speed vc, immersion depth L c, and settings bc for the electromagnetic brake – are generally variable over time and are controlled or regulated, for example, by plant automation. Because these production parameters have the strongest influence on the flow pattern of the introduced melt in continuous casting – and thus on the meniscus profile itself – recording these parameters allows the data model to be trained with particular reliability with respect to the key influencing factors for the meniscus profile.

[0036] According to a further preferred embodiment of the inventive method, the production parameters comprise – in addition to or as an alternative to those already mentioned – at least one of the following parameters: a casting width wc or a casting thickness dc of the reference mold, an oscillation frequency f osc or an oscillation stroke h osc of the reference mold, a metallurgical composition mc of a melt introduced into the continuous casting mold, or a casting powder type pc applied to the melt. These production parameters – in particular an instantaneous oscillation frequency f osc or an instantaneous oscillation stroke h osc – can also be time-dependent and accordingly have a different value for each time interval Δt.Furthermore, these production parameters each have a different effect on the friction conditions during continuous casting, so that by recording these additional production parameters, a particularly accurate determination of a menuskus profile is advantageously enabled with the help of the trained data model with regard to different product classes.

[0037] According to a further preferred embodiment of the method according to the invention, the reference mold has an electromagnetic stirrer. Electromagnetic stirrers can influence the flow of the melt introduced into the mold—especially flowing in a horizontal plane—which in turn affects the meniscus profile. According to this embodiment, the production parameters—alternatively or additionally to those already mentioned—include setting values ​​sc for the electromagnetic stirrer (such as current or stirring frequency), which can be time-dependent. This allows the data model to be advantageously equipped with even better predictive accuracy for the instantaneous meniscus profile in the continuous casting mold (which is not identical to the reference mold) by additionally recording these setting values, provided that the latter has an electromagnetic stirrer.

[0038] According to a preferred embodiment of the method according to the invention, the data model is designed as a machine learning-based data model (ML model). Examples of ML models are those based on Random Forest, Ensemble Learning, Support Vector Machines, and artificial neural networks.

[0039] Such machine learning (ML) models, as well as their training and validation, are known from the prior art and are not themselves part of the invention. An advantage of using ML models is that, with a sufficiently large number of training and test datasets for validation (corresponding to a large number of recorded production parameters P, secondary temperature values ​​Ti,j, and determined instantaneous meniscus heights Mi), an instantaneous meniscus profile can be reliably predicted even with a complex relationship to the production data P.

[0040] A continuous casting mold according to the invention, with a corresponding data model, can be provided by arranging horizontally oriented optical fiber sensors in at least one side plate of the continuous casting mold at one or more installation heights h₁,...,hₙ. The installation heights h₁,...,hₙ correspond to the installation heights h₁,...,hₙ in a reference mold, based on which the data model was created. The optical fiber sensors each have a plurality of first sensor points for detecting temperatures in the continuous casting mold. The installation heights h₁,...,hₙ and the data model are determined according to the method described above according to the invention.

[0041] Preferably, the first sensor points in the respective fiber optic sensors are spaced horizontally from each other by a distance of 30 mm to 100 mm. This distance between the first sensor points does not necessarily have to be constant, but can vary within the aforementioned range for a single fiber optic sensor. This allows for a significantly finer spatial resolution of the temperature distribution in the continuous casting mold compared to temperature measurement with conventional thermocouples, which are typically spaced approximately 200 mm apart horizontally.

[0042] In a further embodiment of the invention, the optical fiber sensors can preferably be arranged at only a single installation height h1 in the continuous casting mold. For example, several optical fiber sensors can also be arranged in several plates (e.g., in the two side plates) of the continuous casting mold, each at the same installation height h1.

[0043] In a continuous casting process according to the invention, a continuous casting mold as described above is used with a corresponding data model. The data model is determined according to the method described above. During continuous casting with the continuous casting mold, initial temperature values ​​Ti,j' are recorded at one or more time intervals Δt using the first sensor points and fed to the data model as input variables. Optionally, preferably, an instantaneous casting velocity vc of the continuous casting mold and / or an immersion depth Lc of a casting tube into a melt introduced into the continuous casting mold and / or an argon flow fAr of the casting tube are also recorded and fed to the data model as input variables. Based on the (trained) data model (and the supplied input variables), an instantaneous meniscus profile M' in the continuous casting mold is determined at each time interval Δt.

[0044] The instantaneous meniscus profile M' is first determined for those horizontal positions that correspond to the installation positions of the first sensor points of the optical fiber sensors of the continuous casting mold according to the invention. For intermediate positions, the instantaneous meniscus profile M' can be approximated, for example, by means of spline interpolation.

[0045] In a preferred embodiment of the continuous casting process according to the invention, the continuous casting mold has an electromagnetic brake. According to this embodiment, setting values ​​bc for the electromagnetic brake are also acquired in each time interval Δt and supplied to the data model as input variables. Additionally or alternatively, according to this embodiment, the continuous casting mold has an electromagnetic stirrer; in this case, setting values ​​sc for the electromagnetic stirrer are also acquired in each time interval Δt and supplied to the data model as input variables.

[0046] In a further preferred embodiment of the continuous casting method according to the invention, in each time interval Δt, a casting width wc and / or a casting thickness dc and / or an oscillation frequency f osc and / or an oscillation stroke h osc of the continuous casting mold and / or a metallurgical composition mc of a melt introduced into the continuous casting mold and / or a type of casting powder pc applied to the melt are recorded and supplied to the data model as input variables.

[0047] The advantageous effects of the continuous casting method according to the invention and its preferred embodiments correspond to those of the previously described method according to the invention for determining installation heights for horizontally oriented optical fiber sensors in a continuous casting mold and a corresponding data model.

[0048] The properties, features, and advantages of this invention described above, as well as the manner in which they are achieved, will become clearer and more readily understandable in connection with the following description of exemplary embodiments, which are explained in more detail in conjunction with the drawings. Identical details are marked with the same designations in all figures. The figures show: Figure 1 (FIG 1) a schematic representation of a reference mold of an embodiment according to the invention, Figure 2 (FIG 2) two diagrams of instantaneous temperatures and instantaneous meniscus heights, Figure 3 (FIG 3) instantaneous correlation coefficients for series of optical fiber sensors in a reference mold, Figure 4 (FIG 4) time-averaged correlation coefficients of a reference mold, Figure 5 (FIG 5) an embodiment of the method according to the invention for determining installation heights and a reference mold according to a first embodiment, Figure 6 (FIG 6) a schematic representation of a continuous casting mold according to a first embodiment according to the invention, Figure 7 (FIG 7) a continuous casting mold or reference mold according to a second embodiment according to the invention, and Figure 8 (FIG 8) a continuous casting mold or reference mold according to a third embodiment according to the invention.

[0049] Identical parts in each figure are labelled with the same identifiers.

[0050] Figur 1 (FIG 1 Figure 1 shows a schematic representation of a mold plate 13 of a reference mold 20 according to an embodiment of the invention for determining installation heights for horizontally oriented optical fiber sensors in a continuous casting mold 10 that is not identical to the reference mold 20 (in FIG 1 (not shown). The reference mold 20 and the continuous casting mold 10 are both designed in the form of plate molds for casting slab formats and each have a pair of broad side plates and a pair of narrow side plates.

[0051] In the FIG 1 The broadside plate 13 of the reference mold 20, as illustrated, shows 30 horizontally oriented, second optical fiber sensors 21 arranged at horizontal intervals dh to each other at xi. Each second optical fiber sensor 21 comprises nine second sensor points 22 arranged at constant intervals along the same optical fiber sensor. The second sensor points 22 are configured, for example, as a fiber Bragg grating and are set up for temperature detection.

[0052] This design of the second optical fiber sensors 21 and their arrangement in the side plate 13 results in a grid-like arrangement of the second sensor points 22 in nine horizontal rows R1 to R9. This allows for the area-wide detection of second temperatures Ti,j in the side plate 13. The index i denotes a horizontal position of the respective sensor point 22, while the index j denotes a vertical position.

[0053] Furthermore, in FIG 1 Recesses 27 for retaining bolts of the side plate 13 are shown, which can also serve as mounting locations for conventional thermocouples for temperature measurement. However, these retaining bolts 27 have a significantly lower areal density than the second optical fiber sensors 21 directly integrated into the side plate 13.

[0054] In the right half of FIG 1 A thick dashed line schematically depicts the pouring surface profile inside the reference mold 20 during continuous casting. The indicated pouring surface has a raised edge, corresponding to a flow pattern of the melt introduced during continuous casting.

[0055] Furthermore, in FIG 1 A pouring level sensor 29 is shown, by means of which a pouring level 14 in the reference mold 20 can be determined independently of the second sensor points 22. The pouring level 14 corresponds to the pouring level profile at a specific horizontal position in the reference mold 20.

[0056] Finally, in FIG 1 Current meniscus heights M i are shown, which correspond to the horizontal installation positions xi and are determined at a respective installation position xi.

[0057] In FIG 2 The upper graph shows examples of instantaneous recorded second temperature values ​​Ti,jmax of a series Rjmax, and the lower graph shows the instantaneous meniscus heights Mi in the reference mold 20 at the respective horizontal installation positions xi. The series Rjmax is that series of second sensor points 22 whose temperature values ​​– after time averaging over a large number of recorded casting situations in the reference mold 20 – exhibit a maximum correlation with the instantaneous meniscus heights Mi.

[0058] FIG 3 The figure shows the numerical value (horizontal axis) of instantaneous correlation coefficients K1 to K21 in a casting situation of a reference mold 20, which has 21 rows of second sensor points 22. Each correlation coefficient Kj is determined for the corresponding row Rj. Furthermore, in FIG 3 A separately determined pouring level 14 in the reference mold 20 is indicated by a dashed line, which is located a little below the row R 10 in the vertical direction.

[0059] Furthermore, it is from FIG 3 It is evident that in the depicted casting situation, those rows exhibiting the highest correlation coefficients are those arranged at a comparable height to the casting level 14. In particular, relatively speaking, more rows above the casting level 14 show a comparable correlation coefficient than those below it; from row R 12 onwards, the correlation decreases sharply with increasing downward distance.

[0060] In FIG 4 These are the numerical values ​​(horizontal axis) of averaged correlation coefficients. K 1 until K 9 ,The instantaneous correlation coefficients K1 to K9, averaged over a multitude of instantaneous correlation coefficients, are represented as filled dot symbols on a reference mold with nine rows of second sensor points 22. Furthermore, a quantile range is indicated by two dashed lines, within which a percentage (e.g., 30%) of instantaneous correlation coefficients Kj of the respective series Rj lies, defined by the quantile q.

[0061] The series R 5 (corresponding to R jmax ) has the largest averaged correlation coefficient. K jmax all series Rj. The quantile q of the series Rjmax determines that range of values ​​(in FIG 4 (represented by two vertical lines) around the average correlation coefficient K jmax , within which the percentage of instantaneous correlation coefficients K jmax of the series R jmax, determined by the quantile q, lie.

[0062] Within this range of values ​​(relative to the horizontal axis) lie the averaged correlation coefficients of the series R3 to R6. Consequently, according to the in FIG 4 In the example shown, the series R 3 to R 6 (corresponding to R j1 to R jN with N = 4) were used to determine installation heights; as shown in FIG 4 As can be seen, the series R3 to R6 are also those with the largest averaged correlation coefficients. K j all series. For example, in the specific case shown, the installation heights of series R 3 to R 6 of the reference mold are transferred to the continuous casting mold with the reduced number of up to four horizontally oriented fiber optic sensors, whereby those installation heights that are not feasible in the continuous casting mold, e.g. for structural reasons, are omitted.

[0063] FIG 5 Figure 1 shows an embodiment of the inventive method for determining installation heights and a corresponding data model using a mold plate 13 of a reference mold 20, which is analogous to that of FIG 1 is trained. In addition to FIG 1 A casting tube 24 is shown, through which liquid metallic melt 23 and an argon flow f AR are introduced into the interior of the reference mold 20.

[0064] The molten metal 23 exits the casting tube 24 at corresponding openings in the region of its lower end, generally forming both upward and downward currents (in FIG 5 (represented by corresponding curved arrows). The casting level 14 is determined by means of the casting level sensor 29. The immersion depth Lc of the casting tube 24, defined as the distance from the outlet level of the melt 23 to the casting level 14, can be determined from its geometric dimensions. Furthermore, in FIG 5 by means of a downward-pointing arrow, a casting rate vc is symbolized, which corresponds to the rate at which the partially solidified melt is drawn off from the lower end of the reference mold 20.

[0065] The casting speed vc and the casting level 14 are available as real-time values, so that during continuous casting with the reference mold 20, the instantaneous values ​​of the casting speed vc and / or the immersion depth L c and / or the argon flow f Ar are recorded as instantaneous production parameters at a large number of time intervals Δt. Furthermore, second temperature values ​​T i,j (in FIG 5 (represented by a corresponding connecting arrow) as real-time values.

[0066] The time intervals Δt have a duration preferably in the range of 0.2 s to 1 s and can, but do not necessarily have to, follow each other immediately. Furthermore, the time intervals Δt taken together do not necessarily cover only one specific casting process, but rather encompass different casting situations (e.g., melts with different compositions or different casting speeds) that are representative of the product range produced with the reference mold 20. The essential feature is the multiple repetition of the steps carried out within a time interval Δt, which is described in FIG 5 is symbolized by a symbol of Δt encircled by an arrow.

[0067] The second temperature values ​​T i,j correspond to the currently measured temperatures in the side plate 13 at the horizontal installation position xi and the installation height hj of the series R j of the respective sensor point 22. From the second temperature values ​​T i,j, instantaneous meniscus heights M i are determined (as described above): this is in FIG 5 symbolized by a corresponding connecting arrow.

[0068] Furthermore, in each time interval Δt, for each series R j at second sensor points 22, an instantaneous correlation coefficient K j is determined between the second temperature values ​​T i,j of the respective series R j and the set of determined instantaneous meniscus heights M i: this is in FIG 5 This is symbolized by the two arrows pointing to Kj. The correlation coefficients Kj are calculated, for example, using the preceding formula (1).

[0069] Subsequently, an average correlation coefficient is calculated for each series R j ( K j ) determined by averaging the values ​​of the instantaneous correlation coefficients K j of the respective series over all time intervals Δt: this is only done after a large number of instantaneous correlation coefficients K j have been determined in a corresponding number of time intervals Δt, which preferably cover different casting situations.

[0070] From the horizontally oriented series R j at second sensor points 22 of the reference mold 20, those N series R j1 ,..., R jN with the largest averaged correlation coefficients are then selected. K j selected, for example, based on a given quartile q as above in connection with FIG 4 described. The installation heights of the series R j1 ,..., R jN selected in this way are equated to the installation heights h 1 ,...,h N for horizontally oriented optical fiber sensors 11 in a continuous casting mold 10 (not identical to the reference mold) (in FIG 5 (not shown).

[0071] Furthermore, for the selected series R j1 ,...,R jN, a data set D with input variables E and output variables A is created for each time interval Δt: the input variables E comprise the second temperature values ​​T i,j recorded in the respective time interval Δt as well as the instantaneous production parameters, the latter being in the FIG 5 In the illustrated embodiment, the immersion depth Lc, the pouring rate vc, and the argon flow rate fAr correspond to the respective time interval Δt. The instantaneous meniscus heights Mi correspond to the initial parameters A of the respective time interval Δt.

[0072] The data model 100 is trained using the large number of data sets D that have been recorded for each time interval Δt. This training process – e.g., splitting the available data sets D into independent training and validation data sets – is known from the prior art and is not the subject of the invention.

[0073] FIG 6 Figure 1 shows a continuous casting mold 10 with a side plate 13 in which three rows of horizontally oriented optical fiber sensors 11 are arranged. The installation heights h1, h2 and h3 of the optical fiber sensors 11 and the corresponding data model 100 of the continuous casting mold 10 are determined according to the figure in Figure 10. FIG 5 The described procedure was used. The optical fiber sensors 11 each have a plurality of first sensor points 12 for detecting temperatures in the continuous casting mold 10 (specifically: directly in the side plate 13) and can, for example, be designed as FiberBragg gratings.

[0074] During continuous casting with the continuous casting mold 10, initial temperature values ​​T i,1 ', T i,2 ' and T i,3 ' are recorded at a given time interval Δt using the first sensor points 12. The index i here denotes a horizontal position of the corresponding sensor point 12 in the side plate 13, while the second index denotes the row or installation height of the sensor point 12 with which the respective temperature value is recorded.

[0075] Liquid metallic melt 23 and an argon flow fAr are introduced into the continuous casting mold via a pouring tube 24. A pouring level sensor 29 continuously determines the instantaneous pouring level 14 in the continuous casting mold 10, which is preferably used to control the flow rate of the melt 23 through the pouring tube 24. Furthermore, the instantaneous pouring velocity vc of the continuous casting mold 10 and – alternatively or additionally – the immersion depth Lc of the pouring tube 24 as well as the instantaneous argon flow fAr into the melt 23 are recorded.

[0076] The first temperature values ​​T i,1 ', T i,2 ' and T i,3 ', the instantaneous casting velocity vc, the measured immersion depth L c and the measured instantaneous argon flow f Ar are fed into the trained data model 100 as input variables. Finally, for the relevant time interval Δt, an instantaneous meniscus profile M' in the continuous casting mold 10 is determined using data model 100.

[0077] The described process – recording initial temperature values ​​T i,1 ', T i,2 ', T i,3 ', the instantaneous pouring rate vc, the immersion depth L c and the argon flow f Ar, feeding these quantities to the data model 100 and determining the instantaneous meniscus profile M' – is preferably repeated in a multitude of successive time intervals Δt, which in FIG 6 as indicated by a correspondingly circled symbol of Δt.

[0078] FIG 7 Figure 1 shows a continuous casting mold 10 and a reference mold 20, respectively, according to a second embodiment of the invention. FIG 7 are - in addition to those in FIG 5 or FIG 6 The features shown – each further features of the continuous casting mold 10 or the reference mold 20 are shown, which will be discussed below.

[0079] The mold 10 or 20 has an electromagnetic stirrer 28 in its upper region and an electromagnetic brake 26 in its lower region, each arranged on the outside of the side plate 13. Preferably, a further electromagnetic stirrer or a further electromagnetic brake can be arranged on a further side plate 13' opposite the side plate 13 (in FIG 7 (not shown).

[0080] During continuous casting with the reference mold 20, the setpoint values ​​bc for the electromagnetic brake and sc for the electromagnetic stirrer 28 are recorded as instantaneous production parameters in each time interval Δt and added to the data set D as further input variables E. The data model 100 is thus ultimately trained with respect to the setpoint values ​​bc and sc.

[0081] Similarly, when a data model 100 trained in this way is available, during continuous casting with the continuous casting mold 10, the setpoint values ​​bc for the electromagnetic brake 26 and the setpoint values ​​sc for the electromagnetic stirrer 28 are recorded in each time interval Δt and fed to the data model 100 as further input variables E. Based on the data model 100, an instantaneous meniscus profile M' is thus determined in this configuration in each time interval Δt, also taking into account the setpoint values ​​bc for the electromagnetic brake 26 and the setpoint values ​​sc for the electromagnetic stirrer 28.

[0082] FIG 8 Figure 1 shows a continuous casting mold 10 and a reference mold 20, respectively, according to a third embodiment of the invention. FIG 8 are - in addition to those in FIG 5 or FIG 6 The features shown are further features of the continuous casting mold 10 and the reference mold 20, respectively, which are described below.

[0083] During continuous casting, the mold 10 or 20 is moved up and down by one oscillation stroke h osc at an oscillation frequency f osc (in FIG 8 (symbolized by a double arrow). Furthermore, the mold 10 or 20 has a specific casting thickness dc corresponding to the dimensions of the respective narrow side plates 17, 17'. A respective casting width wc of the produced continuous casting product can be set by shifting the narrow side plates 17, 17'. Finally, the metallurgical composition mc of the melt 23 and a type of casting powder pc added to it are recorded.

[0084] During continuous casting with the reference mold 20, the oscillation frequency f osc, the oscillation stroke h osc, the casting thickness dc, the casting width wc, the metallurgical composition mc and the casting powder type pc are recorded as instantaneous production parameters in each time interval Δt (or are stored as predefined values ​​in a plant control system) and are added as further input variables E to the data set D, so that the data model 100 is subsequently trained with respect to these production parameters.

[0085] Similarly, when such a trained data model 100 is present, during continuous casting with the continuous casting mold 10, an oscillation frequency f osc, an oscillation stroke h osc, a casting thickness dc, a casting width wc, a metallurgical composition mc and a casting powder type pc of the melt 23 are recorded in each time interval Δt and supplied as further input variables E to the data model 100.

[0086] Accordingly, using data model 100 in this configuration, an instantaneous meniscus profile M' is determined in each time interval Δt, also taking into account the quantities f osc , h osc , dc , wc , mc and pc. Reference symbol list

[0087] 10 Continuous casting mold 11 Optical fiber sensor 12 First sensor point 13, 13' Side plate 14 Pouring level 17, 17' Narrow side plate 20 Reference mold 21 Second optical fiber sensor 22 Second sensor point 23 Melt 24 Pouring tube 26 Electromagnetic brake 27 Recess for retaining bolt 28 Electromagnetic stirrer 29 Pouring level sensor 100 Data model A Output variables bc Brake setting value dc Casting thickness dh Horizontal distance D Data set E Input variables f Ar Argon flow f osc Oscillation frequency h 1 ,...,h N Installation height hj Installation height h osc Oscillation stroke K j Instantaneous correlation coefficient K jmax maximum average correlation coefficient K javerage correlation coefficient L c immersion depth M' instantaneous meniscus profile M i instantaneous local meniscus height mc metallurgical composition pc casting powder type q quantile R j series R j1 ,...,R jN series R jmax series sc stirrer setting value T i,j 'first temperature values ​​T i,j second temperature values ​​vc pouring speed wc pouring width xi , xi 'horizontal installation position Δt time interval

Claims

1. Method for determining one or more installation heights (h1,...,h N ) for horizontally oriented optical fiber sensors (11) in a continuous casting mold (10) and a corresponding data model (100), - wherein the optical fiber sensors (11) can be arranged in at least one side plate (13) of the continuous casting mold (10), - wherein the optical fiber sensors (11) each have a plurality of first sensor points (12) for detecting temperatures in the continuous casting mold (10), - wherein the installation heights (h1,...,h N ) are determined using a reference mold (20) which has a plurality of second sensor points (22), - wherein the second sensor points (22) are arranged in a grid pattern at several horizontal installation positions (x i ) in several horizontal rows (R j ) at a respective installation height (h j) are arranged in the reference mold (20) and configured to detect temperatures, - wherein, during continuous casting, instantaneous production parameters are recorded at a multitude of time intervals (Δt) using the reference mold (20) and second temperature values ​​(T) are recorded by means of the second sensor points (22). i,j ) at the horizontal installation position (x i )and the installation height (h j ) of the series (R j ) of the respective sensor point (22) in the reference mold (20) are recorded, - whereby the second temperature values ​​(T i,j ) current meniscus heights (M i ) at the horizontal installation positions (x i ) are determined, - where for each time interval (Δt) and for each series (R j ) an instantaneous correlation coefficient (K j ) between the second temperature values ​​(T i,j ) of the respective series (R j ) and the current meniscus heights (M i ) is determined, - where for each series (Rj ) from all instantaneous correlation coefficients (K j ) a correlation coefficient averaged over all time intervals (Δt) K j ) is determined, - where from the series (R j ) those (N) series (R j1 ,..., R jN ) with the largest average correlation coefficients ( K j ) are selected, - where the installation heights (h1,...,h N ) the installation heights of the selected series (R j1 ,..., R jN ) are equated, - where the selected series (R) continue to be used. j1 ,...,R jN ) for each time interval (Δt) a data set (D), comprising input variables (E) and output variables (A), is created, - where the input variables (E) are the second temperature values ​​(T i,j ) of the respective time interval (Δt) of the respective series (R j1 ,..., R jN) as well as the instantaneous production parameters in the respective time interval (Δt) and the output variables (A) the instantaneous meniscus heights (M) i ) of the respective time interval (Δt), and - where the data model (100) is trained on the data sets (D).

2. The method of claim 1, wherein the second sensor points (22) are arranged in at least five horizontal rows (R j ), comprising at least ten, preferably at least fifteen, horizontal installation positions (x i ), are arranged.

3. Method according to claim 1 or 2, wherein the instantaneous correlation coefficients (K) j ) according to K j = ∑ i = 1 N j T i , j − T j ¯ ⋅ M i − M ¯ / ∑ i = 1 N j T i , j − T j ¯ 2 ⋅ ∑ i = 1 N j M i − M ¯ 2 with T j ¯ = 1 N j ∑ i = 1 N j T i , j und M ¯ = 1 N j ⋅ ∑ i = 1 N j M i are determined, where (N j ) the number of second sensor points (22) of the respective row (R j ) the reference mold (20).

4. Method according to one of the preceding claims, wherein the number of installation heights (h1,...,h N) is determined in such a way that the averaged correlation coefficients ( K j ) of the selected series (R j ) each within a given quantile (q) of a series (R jmax ) with the largest average correlation coefficient K jmax lie, where the quantile (q) is preferably 30%.

5. Method according to any of the preceding claims, wherein the production parameters include a casting speed (v c ) the reference mold (20) and / or an immersion depth (L c ) of a casting tube (21) into a melt (23) introduced into the reference mold (20) and / or an argon flow (f Ar ) of the pouring tube (21).

6. Method according to any of the preceding claims, wherein the reference mold (20) has an electromagnetic brake (26) and wherein the production parameters further include setting values ​​(b c ) for the electromagnetic brake (26) include.

7. Method according to any of the preceding claims, wherein the production parameters further include a casting width (w). c ) and / or a casting thickness (d c ) and / or an oscillation frequency (f osc ) and / or an oscillation amplitude (h osc ) the reference mold (20) and / or a metallurgical composition (m c ) a melt (23) introduced into the reference mold (20) and / or an applied casting powder type (p c ) include.

8. Method according to any of the preceding claims, wherein the reference mold (20) has an electromagnetic stirrer (26) and wherein the production parameters further include setting values ​​(s c ) for the electromagnetic stirrer (28).

9. Method according to any of the preceding claims, wherein the data model (100) is based on machine learning.

10. Continuous casting mold (10) with a corresponding data model (100), - wherein in at least one side plate (13) of the continuous casting mold (10) in one or more installation heights (h1,...,h N ) horizontally oriented optical fiber sensors (11) are arranged, - wherein the optical fiber sensors (11) each have a plurality of first sensor points (12) for detecting temperatures in the continuous casting mold (10), and - wherein the installation heights (h1,...,h N ) and the data model (100) according to one of claims 1 to 9 are determined.

11. Continuous casting mold (10) according to claim 10, wherein the first sensor points (12) in the respective optical fiber sensors (11) are spaced horizontally apart from each other (d h ) are spaced 30mm to 100mm apart.

12. Method for continuous casting with a continuous casting mold (10) with a corresponding data model (100), - wherein the continuous casting mold (10) and the data model (100) are configured according to claim 10 or 11, - wherein during continuous casting with the continuous casting mold (10) first temperature values ​​(T) are recorded at one or more time intervals (Δt) by means of the first sensor points (12). i,j ') are recorded and fed into the data model (100) as input variables, and -- based on the data model (100) an instantaneous meniscus profile (M') in the continuous casting mold (10) is determined.

13. Method Claim 12, wherein during continuous casting with the continuous casting mold (10) an instantaneous casting speed (v) is further maintained at one or more time intervals (Δt). c ) the continuous casting mold (10) and / or an immersion depth (L c) of a casting tube (24) into a melt (23) introduced into the continuous casting mold (10) and / or an argon flow (f Ar ) of the casting pipe (21) are recorded and fed into the data model (100) as input variables.

14. Method of claim 12 or 13, wherein - the continuous casting mold (10) has an electromagnetic brake (26) and further setting values ​​(b) are available in each time interval (Δt). c ) for the electromagnetic brake (26) are recorded and supplied to the data model (100) as input variables, and / or - the continuous casting mold (20) has an electromagnetic stirrer (26) and continues to have setpoint values ​​(s) in each time interval (Δt). c ) for the electromagnetic stirrer (26) are recorded and supplied to the data model (100) as input variables.

15. Method according to any one of claims 12 to 14, wherein in each time interval (Δt) a casting width (w) is further c ) and / or a casting thickness (d c) and / or an oscillation frequency (f osc ) and / or an oscillation amplitude (h osc ) the continuous casting mold (10) and / or a metallurgical composition (m c ) a melt (23) introduced into the continuous casting mold (10) and / or a type of casting powder used (p c ) are captured and fed into the data model (100) as input variables.

Images