Abnormality detection device, abnormality detection method, and program

Abstract

To accurately predict a risk of occurrence of breakout before a cast piece is drawn from a cast.SOLUTION: An abnormality detection device includes: a temperature change speed calculation section which calculates a temperature change speed at a position of each of thermometers from the temperature measurement value of the thermometer; a casting length calculation section which calculates a casting length from the start of casting from a casting speed of a continuous casting machine; a tracking surface management section which virtually generates a tracking surface, which is a surface parallel to a molten metal surface, at prescribed time intervals within a cast and storing data in an explanatory variable record corresponding to each short side of the tracking surface; a probability value calculation section which calculates a probability value indicating a possibility that an abnormality is occurring for each short side of the tracking surface on the basis of the data stored in the explanatory variable record; and an abnormality determination section which determines whether a breakout in which a broken hole is created in part of a cast piece discharged from the cast occurs, on the basis of the probability value calculated by the probability value calculation section.SELECTED DRAWING: Figure 8

Description

[Technical field]

[0001] The present invention relates to an anomaly detection device, an anomaly detection method, and a program, and more particularly to an anomaly detection device, an anomaly detection method, and a program that provide technology that enables the risk of a breakout to be accurately predicted before the slab is pulled out of the mold. [Background technology]

[0002] A continuous casting machine produces cast pieces from molten steel and is equipped with a mold consisting of long and short sides, and a submerged nozzle that supplies molten steel from a tundish to the mold. After the cast piece leaves the mold, it is supported by rolls. The surface of the mold perpendicular to the axis of the rolls is called the short side (or short side), and the discharge hole of the submerged nozzle opens toward this short side.

[0003] It is desirable for the solidified shell formed from the molten steel in the mold to grow uniformly around the entire circumference of the mold.

[0004] However, the solidified shell may grow unevenly, such as by forming an air gap between the solidified shell and the copper plate due to deformation of the solidified shell caused by sticking (restraining) the solidified shell to the copper plate due to breakage of the powder flux film, etc.

[0005] In addition, when casting hypoperitectic steel or alloy steel, which have a large solidification shrinkage, poor crystallization of the flux or uneven inflow from the molten metal surface can cause uneven heat transfer around the circumference of the mold, which can lead to deformation of the solidified shell and the formation of an air gap between the solidified shell and the copper plate.

[0006] When a poorly solidified portion where the area of ​​the unevenly grown solidified shell has expanded is pulled out from the bottom of the mold, the solidified shell at the poorly solidified portion may break because its tensile strength cannot withstand the static pressure of the molten steel inside the slab. In such a case, a breakout occurs, in which the molten steel flows out from the broken portion. The occurrence of a breakout requires the stopping of casting, the withdrawal of the slab, the removal of the molten steel ingot that has flowed out, and segment replacement, and therefore it takes a long time to return to normal operation, which is a factor in reducing productivity. Therefore, it is important to detect the occurrence of poor solidification before it leads to a breakout.

[0007] Furthermore, for example, when inclusions such as alumina adhere to the vicinity of the discharge hole on one side inside the submerged nozzle, the opening area of ​​the discharge hole is reduced, and a so-called drift occurs in which the discharge flow rate is biased to the discharge hole on the opposite side. The molten steel flow generated just below the surface of the molten steel on the drift side has a higher flow velocity than the molten steel flow generated on the opposite side. Such a drift of the molten steel flow generated just below the surface of the molten steel also disrupts the heat balance in the early stage of solidification of the molten steel, and can be a factor in destabilizing the formation of a solidified shell.

[0008] Patent Document 1 discloses a method for detecting a breakout in continuous casting, which determines the heat flux that enters the solidification interface between the molten steel in a mold during continuous casting from the molten steel surface to the mold outlet, determines a total heat flux Q1 obtained by integrating the heat flux over the time during which it is decreasing, and a total heat flux Q2 obtained by integrating the heat flux over the time during which it is increasing, and determines the presence or absence of a risk of a breakout occurring based on the relationship between these total heat fluxes Q1 and Q2.

[0009] Furthermore, Non-Patent Document 1 discloses a calculation formula for estimating the flow velocity of molten steel in a mold from the inclination angle of the dendritic structure of a cast slab. [Prior art documents] [Patent documents]

[0010] [Patent Document 1] JP2009-061469A [Non-patent literature]

[0011] [Non-Patent Document 1] Iron and Steel, Vol. 61, p2982-2990, 1975 Relationship between large inclusions in continuously cast slabs and the growth direction of columnar grains Summary of the Invention [Problem to be solved by the invention]

[0012] In the method of Patent Document 1, the risk of breakout is judged using the overall heat flux measurement results from the molten metal surface to the mold outlet. However, if the degree of risk is calculated using the ratio of the total heat flux as in Patent Document 1, there is a possibility that the risk will be overestimated even when the overall heat flux is small and the possibility of breakout is small. In such cases, so-called false alarms may occur. If a large number of false alarms occur, there is a risk of reducing the efficiency of casting.

[0013] One aspect of the present invention is to provide a technique that allows the risk of a breakout to be accurately predicted before the strand is withdrawn from the mold. [Means for solving the problem]

[0014] In order to solve the above problems, an anomaly detection device according to one aspect of the present invention is an anomaly detection device that detects an anomaly in casting by a continuous casting machine that casts a rectangular slab, the continuous casting machine having a rectangular mold having long sides and short sides, an immersion nozzle that supplies molten steel into the mold from an outlet opening toward the short side of the mold, and thermometers that are provided at a plurality of positions on each of the two short sides of the mold including positions higher and lower than a target level of the molten steel surface, and that measure the temperature inside the short sides of the mold, the anomaly detection device having a temperature change rate calculation unit that calculates a temperature change rate at each thermometer position from the temperature values ​​measured by the thermometers, a casting length calculation unit that calculates a casting length from the start of casting based on the casting speed of the continuous casting machine, and a tracking plane that is a plane parallel to the molten steel surface within the mold, the tracking plane being ... the tracking surface management unit virtually generates a tracking surface at time intervals and moves the tracking surface in the casting direction at the casting speed to obtain data on the temperature and temperature change rate inside the short side when the tracking surface passes the position of the thermometer, and stores the obtained temperature and temperature change rate data in explanatory variable records corresponding to each short side of the tracking surface together with casting control data related to control of the continuous casting machine; a probability value calculation unit calculates a probability value representing the possibility that an abnormality has occurred for each short side of the tracking surface based on the data stored in the explanatory variable records; and an abnormality determination unit determines whether a breakout will occur, in which a hole is broken in part of the cast piece released from the mold, based on the probability value calculated by the probability value calculation unit.

[0015] In order to solve the above problems, an anomaly detection method according to one aspect of the present invention is an anomaly detection method for an anomaly detection device that detects an anomaly in casting by a continuous casting machine that casts a rectangular slab, the continuous casting machine having a rectangular mold having long sides and short sides, an immersion nozzle that supplies molten steel into the mold from an outlet opening toward the short side of the mold, and thermometers that are provided at a plurality of positions on each of the two short sides of the mold including positions higher and lower than a target level of the molten steel surface, and measure the temperature inside the short sides of the mold, the method comprising the steps of: a temperature change rate calculation step of calculating a temperature change rate at each position of the thermometer from the temperature values ​​measured by the thermometer; a casting length calculation step of calculating a casting length from the start of casting based on the casting speed of the continuous casting machine; and a tracking surface that is a surface parallel to the molten steel surface within the mold, the tracking surface being a surface parallel to the molten steel surface within a predetermined the tracking surface being virtually generated at time intervals of 100° C. and the tracking surface being moved in the casting direction at the casting speed to obtain data on the temperature and temperature change rate inside the short sides when the tracking surface passes the position of the thermometer, and the obtained temperature and temperature change rate data are stored in explanatory variable records corresponding to each short side of the tracking surface together with casting control data related to control of the continuous casting machine; a probability value calculation step of calculating a probability value representing the possibility that an abnormality has occurred for each short side of the tracking surface based on the data stored in the explanatory variable records; and an abnormality determination step of determining whether or not a breakout has occurred, in which a hole has been broken in a part of the slab released from the mold, based on the probability value calculated in the probability value calculation step. Effect of the Invention

[0016] According to one aspect of the present invention, the risk of a breakout occurring can be predicted more accurately. [Brief description of the drawings]

[0017] [Figure 1] FIG. 2 is a cross-sectional view of a mold included in a continuous casting machine. [Diagram 2] FIG. 1 is a diagram showing an example in which inclusions such as alumina adhere to the vicinity of a discharge hole on one side inside a submerged nozzle. [Diagram 3] 1 is a graph showing an estimated flow velocity of molten steel in the mold along the vertical direction of both narrow sides of the mold immediately before the molten steel reaches the outflow hole. [Figure 4] FIG. 2 is a diagram for explaining the collision position of a molten steel flow in a mold directed from an immersion nozzle toward a narrow side. [Diagram 5] FIG. 1 is a diagram illustrating the temperature rise within the copper plate constituting the short side of the mold and the remelting of the solidified shell SH. [Figure 6] 1 is a graph showing the temperature distribution inside the copper plate of the mold when a breakout occurs and the temperature distribution inside the copper plate under normal conditions. [Figure 7] 1 is a graph plotting estimated values ​​of the rate of temperature change at each thermometer position at a point where the cast piece breaks. [Figure 8] 1 is a block diagram showing a configuration example of an anomaly detection device; [Figure 9] FIG. 13 is a diagram showing an example of a thermometer placed on a short side of a mold. [Figure 10] 4 is a diagram for explaining the movement of the tracking surface, the temperature corresponding to each position, and the rate of temperature change. FIG. [Figure 11] FIG. 13 is a diagram illustrating an example of an explanatory variable record. [Figure 12] FIG. 13 is a diagram for explaining data of items in an explanatory variable record used when calculating a probability value. [Figure 13] 11 is a flowchart illustrating an example of the flow of an abnormality detection process. [Figure 14] 10 is a flowchart illustrating an example of the flow of a process for detecting a collision flow along a narrow side just below the molten metal surface. [Figure 15] 13 is a flowchart illustrating an example of the flow of a solidified shell re-melting detection process. [Figure 16] 10 is a flowchart illustrating an example of the flow of an abnormality determination process. [Figure 17] FIG. 13 is a diagram showing the verification results of a machine learning model. [Figure 18] FIG. 13 is a diagram showing a simulation result. [Figure 19] FIG. 1 is a block diagram illustrating a physical configuration of a computer used as an anomaly detection device. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0018] Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the drawings.

[0019] (molten steel in the mold) Fig. 1 is a cross-sectional view of a mold included in a continuous casting machine. The mold shown in Fig. 1 is used in continuous casting by the continuous casting machine to produce cast pieces from molten steel.

[0020] The mold 2 shown in FIG. 1 is a rectangular mold having long sides and short sides, and molten steel MS is supplied to the mold 2 from a tundish (not shown) through a submerged nozzle 3.

[0021] The tip of the submerged nozzle 3 is immersed in the molten steel in the mold 2, and a discharge hole 4 from which the molten steel flows out is provided near the tip of the submerged nozzle 3. The discharge holes 4 are provided, for example, on both the left and right sides of the submerged nozzle 3.

[0022] The molten steel solidified in the mold 2 is exposed from the bottom of the mold 2 as a slab and is clamped by rolls (not shown). As the rolls rotate, the slab moves toward a cutter downstream. For this reason, the vertical direction from top to bottom in Figure 1 is called the casting direction.

[0023] The surface of the mold 2 perpendicular to the axis of the roll is called the short side surface or mold short side because its horizontal length is shorter than that of the surface parallel to the roll. Hereinafter, it may be simply called the short side. In Fig. 1, the short sides are shown on the left and right of the mold 2 as rectangles whose longitudinal direction is the vertical direction in the figure. The discharge hole 4 of the submerged nozzle 3 opens toward this short side.

[0024] In order to maintain good surface and internal quality of the slab, an electromagnetic stirrer 5 is provided above the mold 2 to stir the molten steel inside the mold 2. In order to prevent defects inside the slab, an electromagnetic brake device 6 is provided below the mold 2 to apply a static magnetic field and suppress the downward flow of the molten steel.

[0025] In Fig. 1, an arrow F1 is shown indicating the flow of molten steel flowing out from the discharge hole 4. When the electromagnetic brake device 6 strongly suppresses the arrow F1 (referred to as discharge flow F1) of the molten steel, the downward flow is restricted, and a counterflow in the opposite direction to the mainstream is generated around the discharge flow F1 as shown in Fig. 1. In Fig. 1, the counterflow is indicated by an arrow F2. The counterflow F2 (arrow F2) rises along the submerged entry nozzle 3 and diffuses just below the surface of the molten steel, but a short-side collision flow F3 (arrow F3) is formed on the center line of the mold 2 in the thickness direction, which collides from the submerged entry nozzle 3 toward the short side.

[0026] For example, as shown in FIG. 2, when inclusions such as alumina adhere to the vicinity of the discharge hole 42 on one side inside the submerged entry nozzle 3, the opening area of ​​the discharge hole 42 decreases, causing a so-called drift in which the discharge flow rate of molten steel is biased toward the discharge hole 41 on the opposite side.

[0027] In Fig. 2, the discharge flow F11 (arrow F11) on the drift side (left side in the figure) and the discharge flow F12 (arrow F12) on the opposite side (right side in the figure) are shown with the submerged entry nozzle 3 as the axis of symmetry when drift occurs. Similarly, the counter flows F21, F22 (arrows F21, F22) and the short side impingement flows F31, F32 (arrows F31, F32) are shown. It has been confirmed by a numerical simulation that the short side impingement flow F31 generated immediately below the surface of the molten steel MS on the drift side has a higher flow velocity than the short side impingement flow F32 generated on the opposite side.

[0028] It is desirable for the solidified shell SH generated from the molten steel MS in the mold 2 to grow uniformly around the entire circumference of the mold 2. However, due to deformation of the solidified shell SH caused by the adhesion (restraint) of the solidified shell SH to the copper plate due to the breakage of the powder flux film, etc., an air gap may occur between the solidified shell SH and the copper plate. In such cases, the solidified shell SH may grow unevenly. The powder flux is a molten powder that is scattered on the surface of the molten steel MS, and is a lubricant between the solidified shell SH and the copper plate of the mold.

[0029] In addition, when casting hypoperitectic steel or alloy steel, which have a large solidification shrinkage, poor crystallization of the flux or uneven inflow from the molten metal surface causes uneven heat transfer in the circumferential direction of the mold 2, which can lead to deformation of the solidified shell SH and uneven growth of the solidified shell SH.

[0030] In addition, the bias of the narrow side impingement flow F3 described above can also be a factor in destabilizing the formation of the solidified shell SH by disturbing the heat balance in the early stage of solidification of the molten steel MS. In particular, in hypoperitectic steel with a [C] weight percentage of 0.08 to 0.10% in the molten steel, the solidification shrinkage is large, so an air gap is likely to occur between the solidified shell SH and the copper plate.

[0031] When a poorly solidified portion where the area of ​​the unevenly grown solidified shell SH has expanded is pulled out from the bottom end of the mold 2, the solidified shell at the poorly solidified portion breaks because its tensile strength cannot withstand the static pressure of the molten steel inside the slab. In such a case, a breakout may occur, in which the molten steel flows out from the broken portion. The occurrence of a breakout requires the stopping of casting, the withdrawal of the slab, the removal of the molten steel ingot that has flowed out, and segment replacement, and therefore it takes a long time to return to normal operation, which reduces productivity. Therefore, it is important to detect the occurrence of poor solidification before it leads to a breakout.

[0032] When a breakout occurs, the cause of the solidification failure can be identified after the fact by observing and analyzing the solidified shell near the opening through which the molten steel flowed out of the slab recovered after the breakout. Furthermore, the flow velocity of the molten steel in the mold along the vertical direction of both short sides in the mold just before the breakout can be estimated from the inclination angle of the dendritic structure of the slab using the calculation formula shown in Non-Patent Document 1.

[0033] Figure 3 is a graph showing the estimated flow velocity of molten steel in the mold along the vertical direction of both short sides in the mold for an actual breakout case and a normal casting case. As shown in Figure 3, in the normal solidified shell, the flow velocity direction reverses at a depth of 35 mm from the molten steel surface, whereas in the solidified shell where a breakout occurred, the flow velocity direction reverses at a depth of 8 mm.

[0034] Fig. 4 is a diagram explaining the collision position of the molten steel flow in the mold from the submerged entry nozzle toward the short side under normal conditions and when a breakout has occurred. For convenience, in Fig. 4, the longitudinal direction of the short side of the mold 2 is drawn horizontally in the figure, and in this case, the discharge flow from the submerged entry nozzle 3 toward the short side is generated in the direction from top to bottom in the figure.

[0035] The diagram at the top of Fig. 4 corresponds to the normal state, and the diagram at the bottom of Fig. 4 corresponds to the state immediately before breakout, and each diagram includes a graph showing the molten steel flow velocity in the mold along the vertical direction of the short side corresponding to the distance from the molten metal surface, and a diagram showing the flow of molten steel MS predicted from this graph. In the diagram showing the flow of molten steel MS, the short side of the mold is shown by a rectangle, and the area corresponding to the solidified shell SH is shown by oblique lines. The flow of molten steel MS along the vertical direction of the short side is shown by thin black arrows, and the collision position of the molten steel flow from the submerged entry nozzle 3 toward the short side is shown by a thick white arrow.

[0036] As shown in the upper diagram of Figure 4, in a normal state, the collision position of the flow from the submerged entry nozzle 3 toward the narrow side is estimated to be approximately 35 mm from the molten steel surface. On the other hand, as shown in the lower diagram of Figure 4, when a breakout occurs, the collision position is estimated to be approximately 8 mm from the molten steel surface. In other words, when a breakout occurs, the collision position of the flow from the submerged entry nozzle 3 toward the narrow side moves closer to the molten steel surface.

[0037] In both cases, the estimated flow velocity was negative (downward flow) at 50 mm from the molten steel surface, but the absolute value of the estimated flow velocity at this position just before the breakout was greater than in the normal case. From this, it is estimated that the flow velocity of the molten steel colliding with the short side at a position about 50 mm from the molten steel surface when a breakout occurs is greater than in the normal case. This is thought to have caused a spot of poor solidification, which was pulled down to the bottom of the mold and broke.

[0038] In other words, when drift occurs, the growth of the thickness of the solidified shell SH slows down at the position where the molten steel flow collides with the short side, and the thickness of the solidified shell in the casting direction becomes smaller than when there is no drift.

[0039] In addition, the downward flow rate of the molten steel increases below the position where the molten steel flow collides with the short side, and the heat transfer from the molten steel increases, so that the thin part of the solidified shell SH may re-melt when the slab is pulled out of the mold 2. Since the strength of the solidified shell SH at this re-melted part is lower than that of the surrounding area, it cannot withstand the static pressure of the molten steel after being pulled out of the mold 2 and breaks, causing a breakout in which the molten steel flows out of the slab.

[0040] When a breakout occurs, a characteristic phenomenon appears in the change in temperature of the copper plate that makes up the short side.

[0041] That is, since heat transfer from the molten steel increases at the position where the drift of the molten steel collides with the short side, the measured temperature near the molten steel surface becomes high in the temperature distribution inside the copper plate that constitutes the short side of the mold 2.

[0042] In addition, in the area where the breakout occurs, the thermal resistance of the solidified shell SH decreases continuously as remelting progresses, so the temperature of the area in contact with the area where the breakout occurs rises rapidly in the temperature distribution inside the copper plate that constitutes the short side of the mold 2. In other words, the rate of temperature change is high in the area in contact with the area where the breakout occurs.

[0043] FIG. 5 is a diagram for explaining the temperature rise in the copper plates constituting the short sides of the mold 2 and the remelting of the solidified shell SH.

[0044] In Fig. 5, of the left and right short sides of the mold 2, the short side 2A located on the side where drift occurs is shown, and the solidified shell SH growing along the short side 2A is shown. Also, the circles and "X" symbols arranged vertically on the short side 2A indicate the positions of thermometers.

[0045] In Fig. 5, only the short side 2A located on the side where drift occurs is shown among the left and right short sides of the mold 2, but the solidified shell SH also grows on the short side 2B opposite to the short side 2A. The solidified shell SH also moves from the top to the bottom of the mold 2 in Fig. 5 along with the slab moving due to the rotation of the rolls.

[0046] During casting, the level of the molten steel is constantly controlled to keep the level of the molten steel surface at a target level, but because a flow from the submerged entry nozzle 3 toward the short side occurs directly below the molten steel surface, the shape of the molten steel surface in contact with the short side rises above the target level of the molten steel surface control. In Figure 5, the target level L of the molten steel surface is shown, and as described above, a rise in the shape of the molten steel surface in contact with the short side 2A (also called the molten steel meniscus) occurs.

[0047] The flow of molten steel MS along the short side 2A in the mold 2 is indicated by thin black arrows, and the short side collision flow (drift flow) F31 that collides from the submerged entry nozzle 3 toward the short side 2A is indicated by thick white arrows. In addition, the heat flux vectors in the copper plate that constitutes the short side 2A are indicated by thick black arrows.

[0048] When no drift occurs, the flow velocity of the short side impinging flow is low and it impinges on the short side at a position away from the molten metal surface, so the thermal energy of the short side impinging flow does not easily reach a position higher than the molten metal surface. In contrast, when drift occurs as shown in Figure 5, the flow velocity of the short side impinging flow F31 relative to the short side 2A is high and it impinges on the short side 2A at a position close to the molten metal surface (directly below the molten metal surface), so the thermal energy of the short side impinging flow tends to reach a position higher than the target level L of the molten metal surface as a heat flux inside the copper plate. For this reason, it is thought that in the distribution of temperature measurements when a breakout occurs, the high temperature areas are biased toward positions closer to the molten metal surface (shallower) than under normal conditions.

[0049] As described above, the downward molten steel flow rate increases below the position where the molten steel flow collides with the short side, and the heat transfer from the molten steel increases, so the solidified shell SH may remelt. In the example of Fig. 5, the downward molten steel flow rate increases near the position indicated by the dotted ellipse, and the solidified shell SH remelts. In this case, the temperature measured by the thermometer near the position indicated by the dotted ellipse rises rapidly, and the temperature change rate becomes high.

[0050] (Temperature distribution and temperature change rate in the mold copper plate) Figure 6 is a graph showing the temperature distribution inside the copper plate of the mold 2 when a breakout occurs and the temperature distribution inside the copper plate under normal conditions. The horizontal axis of Figure 6 represents the depth from the molten metal surface, and the vertical axis represents the temperature inside the copper plate. As shown in Figure 6, in the temperature distribution when a breakout occurs, there is a tendency for the high temperature areas to be concentrated at positions shallower from the molten metal surface compared to normal conditions, and the temperature readings of the mold copper plate thermometers located at positions 15 mm and 35 mm higher than the target level of the molten metal surface tended to be higher than the average value under normal conditions.

[0051] Figure 7 is a graph plotting the estimated temperature change rate at each thermometer position where the slab breaks when a breakout occurs. The horizontal axis of Figure 7 represents the depth from the molten metal surface, and the vertical axis represents the temperature change rate inside the copper plate.

[0052] For comparison, FIG. 7 shows the estimated temperature change rates of the short side on the side where a breakout occurred and the opposite short side (normal side) at the same time.

[0053] On the short side where the breakout occurred, the temperature change rate continues to increase from above the molten metal surface to near the molten metal surface, and between positions about 45 mm and 125 mm from the molten metal surface. On the other hand, on the short side on the normal side, an increase in the temperature change rate can be seen between positions about 10 mm and 40 mm from the molten metal surface, but no notable information on the temperature change rate is available in other locations.

[0054] The increase in the temperature change rate near the molten metal surface on both the short side where the breakout occurred and the short side where the mold was normal is believed to be due to the short side impingement flow. On the other hand, the increase in the temperature change rate between the position of about 45 mm and the position of about 125 mm from the molten metal surface on the short side where the breakout occurred is believed to be due to the decrease in the thermal resistance of the solidified shell due to remelting.

[0055] In this embodiment, the possibility of the occurrence of a narrow side collision flow just below the molten metal surface (occurrence of a drift flow) is estimated from the characteristics of the temperature distribution described with reference to FIG. 6, and the possibility of the occurrence of re-melting of the solidified shell is estimated from the characteristics of the temperature change rate described with reference to FIG. 7.

[0056] (Configuration of anomaly detection device) Fig. 8 is a block diagram showing an example of the configuration of the abnormality detection device 100 according to this embodiment. The abnormality detection device 100 detects a casting abnormality in a continuous casting machine 1 that casts a rectangular slab. In the example of Fig. 8, the abnormality detection device 100 has a data acquisition unit 121, a temperature change rate calculation unit 122, a casting length calculation unit 123, a tracking surface management unit 124, a probability value calculation unit 125, a parameter setting unit 126, an abnormality determination unit 127, and an alarm output unit 128.

[0057] The continuous casting machine 1 has, for example, a rectangular mold 2 having long and short sides as described with reference to Fig. 1, and an immersion nozzle 3 that supplies molten steel into the mold from a discharge port that opens toward the short side of the mold. In addition, a thermometer is disposed on the short side of the mold 2 to measure the temperature inside the short side of the mold.

[0058] (Thermometer placement) FIG. 9 is a diagram showing an example of a thermometer placed on a short side of the mold 2. On the left side of FIG. 9, a front view of a copper plate constituting the short side of the mold 2 is shown, and on the right side, a side view of the same copper plate is shown. In the figure, a symbol shown as an "x" surrounded by a circle represents a thermometer. Here, multiple thermometers are collectively shown as thermometer 11.

[0059] 9 shows one of the short sides of the mold 2. That is, as described above with reference to FIG 1, the mold 2 has left and right short sides. Therefore, the thermometers are placed at the same positions on each of the left and right short sides.

[0060] The thermometer is a device for measuring the temperature distribution in the copper plate constituting the short side of the mold 2, and is, for example, composed of a thermocouple or an FBG (Fiber Bragg Grating) thermometer. A plurality of thermometers are arranged along the casting direction, including positions higher and lower than the target level L of the molten steel surface. As shown in the side view on the right side of FIG. 9, it is desirable that the distance from the surface of the copper plate in contact with the molten steel to each thermometer is constant. Also, it is desirable that the interval between the thermometers in the casting direction (up and down direction of the short side) is 20 mm or less.

[0061] (Data Acquisition Section) The data acquiring unit 121 acquires temperature measurements from the thermometers arranged on the copper plates constituting the short sides of the mold 2. The data acquiring unit 121 also acquires casting control information from the continuous casting machine. The casting control information is information that indicates, for example, the magnetic field strength of the electromagnetic stirrer 5, the degree of superheat of the molten steel, the height of the tundish, the width of the mold, the casting speed, and the like.

[0062] (Temperature change rate calculation section) The temperature change rate calculation unit 122 calculates the temperature change rate at the position of each thermometer based on the temperature measurement value from each thermometer acquired by the data acquisition unit 121. For example, the temperature change rate calculation unit 122 acquires the temperature of each thermometer at a predetermined time interval, and calculates the temperature change rate each time the temperature is acquired. The temperature change rate may be calculated by differential approximation of the time derivative of each temperature measurement value, or may be calculated by estimating using a Kalman filter for a state equation model that represents the relationship between temperature and change rate in each temperature measuring device.

[0063] As an example, a state equation model showing the relationship between temperature and rate of change is expressed by equations (1) and (2).

[0064]

number

[0065]

number

[0066] (Casting length calculation section) The casting length calculation unit 123 calculates the casting length from the start of casting based on the casting speed included in the control information acquired by the data acquisition unit 121. The casting speed corresponds to the moving speed of the slab that moves with the rotation of the rolls. The casting length calculation unit 123 calculates the casting length at the molten metal surface inside the mold 2, for example, by integrating the casting speed with respect to time from the casting start time.

[0067] Moreover, the casting length calculation unit 123 calculates the distance between the position of each thermometer and the position of the target level of the molten metal surface, and specifies the casting length corresponding to the position of each thermometer at each time by subtracting the calculated distance from the casting length at the position of the target level of the molten metal surface.

[0068] (Tracking surface management section) The tracking surface management unit 124 virtually generates tracking surfaces at predetermined intervals in the mold 2, which are surfaces orthogonal to the casting direction and perpendicular to the short side, i.e., surfaces parallel to the molten metal surface. Then, the tracking surface management unit 124 moves the tracking surfaces in the casting direction at the casting speed.

[0069] More specifically, when the casting length increases by a predetermined length Δz from the start of casting, the tracking surface management unit 124 generates one tracking surface at the position of the uppermost thermometer, above the target level L of the molten metal surface. Then, the tracking surface management unit 124 moves the position of the tracking surface downward by a distance proportional to the casting speed vc every time time Δt elapses.

[0070] Furthermore, the tracking surface management unit 124 identifies and acquires the temperature and the rate of temperature change within the copper plate of the mold 2 when the tracking surface passes the position of the thermometer.

[0071] Now, the position of each thermometer in the casting direction is z p (p is the thermometer number), and the position z of the tracking surface that occurred at the kth time point at time t-1 t-1 k and the position z of the same tracking surface at time t t k between z t-1 k <z p ≦z t k The above conditions shall be satisfied.

[0072] At this time, the tracking surface management unit 124 detects that the tracking surface is at position z p The variable Y represents the temperature or the rate of temperature change at the time point p k at least at time t-1 and at position z p The value of the variable Y t-1 p and Y tp It is estimated by interpolation using

[0073] For example, the variable value Y p k can be obtained by linear interpolation of equation (3).

[0074]

number

[0075] In this way, the tracking surface management unit 124 identifies the temperature and temperature change rate within the copper plate at the time when the kth generated tracking surface passes the position of each thermometer, and stores them in the explanatory variable record together with the casting control data related to the control of the continuous casting machine.

[0076] (Tracking surface movement) FIG. 10 is a diagram for explaining the movement of the tracking surface, and the temperature and temperature change rate corresponding to each position on the tracking surface.

[0077] FIG. 10 shows the short side of the mold 2 at time t-1 and the short side of the mold 2 at time t. The k-th tracking surface TR k The positions of are indicated by diamonds on the solidified shell SH. In addition, in Fig. 10, a graph showing the temperature values ​​measured by each thermometer and the temperature change rate at time t-1 is displayed on the left side of the short side of the mold 2 at time t-1. Similarly, a graph showing the temperature values ​​measured by each thermometer and the temperature change rate at time t is displayed on the left side of the short side of the mold 2 at time t.

[0078] The graph showing the temperature readings from each thermometer has one axis showing the temperature readings and the other axis (Z) showing the position in the casting direction on the short side. Similarly, the graph showing the temperature change rate has one axis showing the temperature change rate and the other axis (Z) showing the position in the casting direction on the short side.

[0079] In FIG. 10, the fifth thermometer from the top is numbered P-1, and the sixth thermometer from the top is numbered P. In this case, from the formula (3), the kth tracking surface TR k The rate of temperature change U when the thermometer passes the position of the number P-1 k P-1 can be calculated using equation (4).

[0080]

number

[0081]

number

[0082]

number

[0083]

number

[0084] In this way, the tracking surface management unit 124 manages the k-th tracking surface TR kThe temperature change rate and temperature at the time when the thermometer passes the position of each thermometer are calculated, and these values ​​are stored as data in the explanatory variable record corresponding to the kth tracking surface. As shown in FIG. 1, the mold 2 has short sides on the left and right. Therefore, the tracking surface management unit 124 calculates the temperature change rate and temperature at the time when the thermometer passes the position of each thermometer, and stores these values ​​as data in the explanatory variable record corresponding to the kth tracking surface. k The tracking surface management unit 124 further calculates the temperature change rate and temperature at the time when the thermometer passes the position of each thermometer. k Calculate the temperature change rate and temperature at the time when the temperature passes each thermometer position.

[0085] (Explanatory variable record) 11 is a diagram showing an example of explanatory variable records. The explanatory variable records are generated corresponding to each of the tracking surfaces. That is, explanatory variable records are generated such that an explanatory variable record corresponds to the first tracking surface, an explanatory variable record corresponds to the second tracking surface, and so on, corresponding to the kth tracking surface.

[0086] In this example, the numbers of the data items stored in the explanatory variable records are shown as "1, 2, 3, 4,...", and the items corresponding to each number are shown.

[0087] Item numbered 1 is the "temperature at the position of thermometer No. 1," item numbered 2 is the "rate of temperature change at the position of thermometer No. 11," item numbered 3 is the "temperature at the position of thermometer No. 2," and item numbered 4 is the "rate of temperature change at the position of thermometer No. 2." The numbers of the thermometers may be given in order, for example, from the highest position on the short side of the mold 2 to the lowest position.

[0088] For example, "temperature at thermometer No. 1 position" indicates the temperature reading of the thermometer at the highest position, and "rate of temperature change at thermometer No. 1 position" indicates the rate of temperature change calculated based on the temperature reading of the thermometer at the highest position.

[0089] The explanatory variable record stores the temperature and temperature change rate at all thermometer positions. For example, if eight thermometers are placed on the short side of mold 2, the explanatory variable record will further store "temperature at thermometer No. 3," "rate of temperature change at thermometer No. 3," "temperature at thermometer No. 4," "rate of temperature change at thermometer No. 4," ... "temperature at thermometer No. 8," and "rate of temperature change at thermometer No. 8."

[0090] The explanatory variable record may further store casting control data. The casting control data is data indicating the magnetic field strength of the electromagnetic stirrer 5, the superheat of molten steel, the height of the tundish, the width of the mold, the casting speed, etc., included in the casting control information. That is, items such as "magnetic field strength" and "superheat of molten steel" may be further added to the explanatory variable record in FIG. 11.

[0091] The casting control data stored in the explanatory variable record corresponding to the k-th tracking surface is generated, for example, based on the casting control information acquired at the time when the k-th tracking surface passes the target level L of the molten metal surface. Alternatively, the casting control data may be generated based on the casting control information acquired at the time when the k-th tracking surface occurs.

[0092] Two such explanatory variable records are generated for one tracking surface, that is, an explanatory variable record related to the left short side and an explanatory variable record related to the right short side are generated for one tracking surface.

[0093] (Probability value calculation part) The probability value calculation unit 125 calculates a probability value representing the possibility that an abnormality has occurred on the tracking surface based on the data stored in the explanatory variable record. The probability value calculation unit 125 uses a machine learning model that learns model parameters using the data of the explanatory variable record acquired in past castings as training data.

[0094] In the example of FIG. 8, the probability value calculation unit 125 has a below-surface narrow side collision flow detection unit 141 and a solidified shell re-melting detection unit 142.

[0095] (Detection of collision flow on the short side directly below the molten metal surface) The narrow side collision flow detection unit 141 directly below the molten metal surface calculates a probability value representing the possibility that a narrow side collision flow directly below the molten metal surface has occurred, in which the molten steel flow from the molten steel MS supplied from the submerged entry nozzle 3 collides with the narrow side directly below the molten metal surface, for a tracking surface that has passed through one of the thermometers 11 on the narrow side of the mold 2 at a position higher than a preset first position.

[0096] More specifically, the short side impingement flow detection unit 141 directly below the molten metal surface mainly detects characteristics similar to the temperature distribution on the breakout occurrence side in Fig. 6 based on the temperature values ​​measured by thermometers placed at relatively high positions (for example, the first to fourth thermometers from the top), and calculates the probability that the tracking surface has passed the position where the short side impingement flow directly below the molten metal surface is occurring. Note that the short side impingement flow detection unit 141 directly below the molten metal surface predicts the occurrence of the short side impingement flow directly below the molten metal surface on each of the left and right short sides for each of the tracking surfaces that are generated sequentially.

[0097] As an example, the unit 141 for detecting a narrow side impinging flow just below the molten metal surface is configured by a machine learning model that detects the occurrence of a narrow side impinging flow just below the molten metal surface by a k-nearest neighbor method.

[0098] The model parameters used in the machine learning model of the below-surface narrow side impingement flow detection unit 141 are learned using explanatory variable records created from data acquired in past castings and label data for each explanatory variable record. The model parameters include a neighborhood size parameter k (k is an integer).

[0099] The label data indicates whether or not a narrow side impingement flow just below the molten metal surface occurs for each explanatory variable record. That is, among explanatory variable records created from data acquired in a casting in which a breakout occurred among previously performed castings, a label indicating the occurrence of a narrow side impingement flow just below the molten metal surface is attached to the explanatory variable record of the tracking surface related to the occurrence of a narrow side impingement flow just below the molten metal surface.

[0100] That is, the distance between the data of the explanatory variable record and the reference data is calculated, the labels of k pieces of reference data that are close in distance are detected, and the abnormality probability pT is calculated using equation (8) based on the number y of abnormal labels.

[0101]

number

[0102] The calculation of the probability value by the short side collision flow detection unit 141 just below the molten metal surface may be performed without waiting until each tracking surface passes the last thermometer. For example, if the temperatures used to predict the occurrence of the short side collision flow just below the molten metal surface correspond to the positions of the first to fourth thermometers from the top, the calculation of the probability value may be performed immediately after the tracking surface passes the position of the fourth thermometer from the top.

[0103] (Solidified shell remelting detection part) The solidified shell remelting detection unit 142 calculates a probability value representing the possibility that remelting of the solidified shell has occurred for a tracking surface that has passed through the lowest thermometer among the thermometers 11 on the short sides of the mold 2.

[0104] More specifically, the solidified shell remelt detection unit 142 mainly detects features similar to the distribution of the temperature change rate on the breakout occurrence side in Fig. 7 based on the temperature change rate calculated from the temperature measurement value of a thermometer placed at a relatively low position, and thereby calculates the probability that the tracking surface has passed through the position where remelt of the solidified shell SH has occurred. Note that, for each of the tracking surfaces generated sequentially, the solidified shell remelt detection unit 142 predicts the occurrence of remelt of the solidified shell SH on each of the left and right short sides.

[0105] As an example, the solidified shell remelting detection unit 142 is configured with a machine learning model that detects the occurrence of remelting of the solidified shell SH by logistic regression. The machine learning model of the solidified shell remelting detection unit 142 is, for example, a machine learning model that sets the output of a logistic function for the value of a linear expression for an item x of an explanatory variable record as an abnormality probability.

[0106] The model parameters of the machine learning model of the solidified shell remelting detection unit 142 are learned using explanatory variable records created from data acquired in castings performed in the past and label data for each explanatory variable record as training data. The coefficients c0, c1,...cn (c0 is the intercept value, and n is the number of explanatory variable data items) in equation (9) described later are the model parameters.

[0107] The label data indicates whether or not solidified shell remelting has occurred for each explanatory variable record. That is, a label indicating the occurrence of solidified shell remelting is attached to the explanatory variable record of the tracking surface related to the occurrence of solidified shell remelting among the explanatory variable records created from the data acquired in a casting in which a breakout has occurred among the castings performed in the past.

[0108] In learning the machine learning model of the solidified shell remelting detection unit 142, optimal values ​​of the model parameters are calculated using the maximum likelihood method that maximizes the likelihood function for the training data.

[0109] The likelihood function L is the training data number t and zt If the labels of the training data are (abnormal = 1, normal = 0), then it is expressed by equation (9).

[0110]

number

[0111]

number

[0112]

number

[0113] (Data used in each detector) FIG. 12 is a diagram explaining data of items in an explanatory variable record used when each of the detection units, the short side collision flow just below the molten metal surface detection unit 141 and the solidified shell remelting detection unit 142, calculates a probability value.

[0114] In FIG. 12, the leftmost column shows the item names of the explanatory variable records, and the two columns on the right side show the items used in the short side collision flow just below the molten metal surface detection unit 141 and the items used in the solidified shell re-melting detection unit 142, respectively.

[0115] The items of the explanatory variable record include "temperature or temperature change rate", which is composed of "temperature or temperature change rate" at each position (depth from the molten metal surface). In the example of Fig. 12, nine positions (depth from the molten metal surface) are shown: "-35mm", "-15mm", ... "125mm".

[0116] For example, the temperature at a depth of -35 mm from the molten metal surface corresponds to "temperature at the position of thermometer No. 1" in Figure 11, and the temperature change rate at a depth of -35 mm from the molten metal surface corresponds to "temperature change rate at the position of thermometer No. 1" in Figure 11. Similarly, the temperature at a depth of -15 mm from the molten metal surface corresponds to "temperature at the position of thermometer No. 2" in Figure 11, and the temperature change rate at a depth of -15 mm from the molten metal surface corresponds to "temperature change rate at the position of thermometer No. 2" in Figure 11.

[0117] The items of the explanatory variable record include "magnetic field strength," "molten steel superheat," "tundish height," "mold width," and "casting speed." These are items that correspond to the casting control data.

[0118] 12, a circular symbol is attached at the position corresponding to the item in the column of items used by the short side impingement flow detection unit 141 just below the molten metal surface. In this example, it is shown that the temperature at a depth of -35 mm from the molten metal surface, the temperature at a depth of -15 mm from the molten metal surface, ..., the temperature at a depth of 45 mm from the molten metal surface are used by the short side impingement flow detection unit 141 just below the molten metal surface. It is also shown that the magnetic field strength, the degree of molten steel superheat, the height of the tundish, the width of the mold, and the casting speed are used by the short side impingement flow detection unit 141 just below the molten metal surface.

[0119] 12, a circular symbol is attached at the position corresponding to the relevant item in the column of items used by the solidified shell remelting detection unit 142. In this example, it is shown that the temperature at a depth of -35 mm from the molten metal surface, the temperature at a depth of -15 mm from the molten metal surface, ..., the temperature at a depth of 45 mm from the molten metal surface are used by the solidified shell remelting detection unit 142. It is also shown that the temperature change rate at a depth of 65 mm from the molten metal surface, the temperature change rate at a depth of 85 mm from the molten metal surface, the temperature change rate at a depth of 105 mm from the molten metal surface, and the temperature change rate at a depth of 125 mm from the molten metal surface are used by the solidified shell remelting detection unit 142.

[0120] As described above, the solidified shell remelt detection unit 142 detects features similar to the distribution of the temperature change rate on the breakout occurrence side in Fig. 7, mainly based on the temperature change rate calculated from the temperature measurement value of the thermometer. However, the distribution of the temperature change rate as shown in Fig. 7 may be observed regardless of the remelting of the solidified shell SH. For this reason, the detection by the solidified shell remelt detection unit 142 also uses the temperature at a depth of -35 mm from the molten metal surface, the temperature at a depth of -15 mm from the molten metal surface, ..., the temperature at a depth of 45 mm from the molten metal surface.

[0121] It is also shown that the solidified shell remelting detection unit 142 uses magnetic field strength, molten steel superheat, tundish height, mold width, and casting speed.

[0122] (Parameter setting section) The parameter setting unit 126 sets the value of an abnormality probability threshold parameter used by the abnormality determination unit 127 described later. The abnormality probability threshold parameter is set using verification data consisting of explanatory variable records acquired in previously performed castings in which an abnormality occurred and normal castings, and labels corresponding to each detection unit.

[0123] The parameter setting unit 126, for example, uses data of explanatory variable records acquired in past castings as verification data, and searches for an optimal value that minimizes the probability that the anomaly determination unit 127 will determine all explanatory variable records whose labels are abnormal as abnormal and will mistakenly determine explanatory variable records whose labels are normal as abnormal, and sets this value as the anomaly probability threshold parameter.

[0124] The parameter setting unit 126 sets an abnormality probability threshold parameter used for the abnormality probability calculated by the short side collision flow just below the molten metal surface detection unit 141, and an abnormality probability threshold parameter used for the abnormality probability calculated by the solidified shell remelting detection unit 142.

[0125] Depending on the validation data, it may not be possible to search for a value that can correctly determine all explanatory variable records whose labels are abnormal as abnormal within the range of values ​​taken by the anomaly probability threshold parameter. In such a case, an optimal value is searched for in order to minimize the probability that explanatory variable records whose labels are abnormal are erroneously determined to be normal and explanatory variable records whose labels are normal are erroneously determined to be abnormal, while a predetermined value is not exceeded.

[0126] The optimal value (threshold) searched for by the parameter setting unit 126 is the abnormality probability threshold parameter p θ The signal is supplied to the abnormality determination unit 127 as the abnormality detection signal.

[0127] In addition, the parameter setting unit 126 may also store model parameters of the machine learning models of the short side collision flow just below the molten metal surface detection unit 141 and the solidified shell remelting detection unit 142, and supply these model parameters to the probability value calculation unit 125.

[0128] (Abnormality judgment section) The abnormality determination unit 127 determines whether or not a breakout will occur, in which a part of the slab released from the mold breaks, based on the probability value calculated by the probability value calculation unit 125. The abnormality determination unit 127 is configured with a short side impingement flow determination unit 151 directly below the molten metal surface corresponding to the short side impingement flow detection unit 141 directly below the molten metal surface, and a solidified shell remelt determination unit 152 corresponding to the solidified shell remelt detection unit 142.

[0129] The short side impingement flow judgment unit 151 judges whether the abnormality is normal or abnormal by comparing the abnormality probability calculated by the short side impingement flow detection unit 141 directly below the molten metal surface with a threshold value. The solidified shell re-melting judgment unit 152 judges whether the abnormality is normal or abnormal by comparing the abnormality probability calculated by the solidified shell re-melting detection unit 142 with a threshold value.

[0130] The short side impingement flow judgment unit 151 just below the molten metal surface and the solidified shell remelting judgment unit 152 judge the abnormality probability threshold parameter p θ (0 <p θ<1) is used to judge whether each abnormality probability indicates an abnormality or normality. As described above, the abnormality probability threshold parameter used by the short side impingement flow just below the molten metal surface judgment unit 151 and the abnormality probability threshold parameter used by the solidified shell re-melting judgment unit 152 are set by the parameter setting unit 126.

[0131] In other words, if the abnormality probability calculated by the short side collision flow detection unit 141 just below the molten metal surface or the solidified shell remelting detection unit 142 is pT, then pT ≧ p θ If pT <p θ If so, the abnormality probability value is determined to represent normality.

[0132] This determination is made before the tracking surface corresponding to the explanatory variable record for which the abnormality probability is calculated moves outside the mold 2.

[0133] In this way, the abnormality determination unit 127 determines whether or not a breakout will occur before the tracking surface moves out of the mold 2.

[0134] As an example, when the abnormality probability calculated by the short side collision flow detection unit 141 just below the molten metal surface is equal to or greater than a threshold value and the abnormality probability calculated by the solidified shell remelting detection unit 142 is equal to or greater than a threshold value, the abnormality determination unit 127 determines that a breakout has occurred, in which part of the cast piece released from the mold has broken (i.e., an abnormality has occurred).

[0135] The unit for judging the narrow side impingement flow just below the molten metal surface 151 and the unit for judging the solidified shell remelting 152 may simultaneously output an abnormality risk that auxiliary indicates the degree of abnormality in the abnormality probability calculated by the unit for detecting the narrow side impingement flow just below the molten metal surface 141 and the unit for detecting the solidified shell remelting 142, respectively. The abnormality risk is, for example, θ When pT is equal to 1, the output is adjusted to 50%, when pT=1, the output is adjusted to 100%, and when pT=0, the output is adjusted to 0%.

[0136] The abnormality risk y for the abnormality probability pT may be calculated by the formula (12), where α is a constant that takes a positive value for adjusting the gradient of the abnormality risk at an abnormality risk of 50% with respect to the abnormality probability.

[0137]

number

[0138] (Alarm output section) The warning output unit 128 controls the output of a warning regarding the occurrence of a breakout. For example, the warning output unit 128 outputs a warning notifying that a breakout is highly likely to occur based on the determination result of the abnormality determination unit 127. The warning may be issued using, for example, a sound, an image, a vibration, or the like.

[0139] The alarm output unit 128 may be configured to output an alarm when, for example, the abnormality determination unit 127 determines that the abnormality probabilities output by the below-surface narrow side collision flow detection unit 141 and the solidified shell remelting detection unit 142 are both abnormal. Alternatively, the alarm may be output when a state in which the abnormality probabilities output by each of the units are both determined to be abnormal continues a predetermined number of times.

[0140] The alarm output unit 128 may also issue an alarm notifying the user of the high possibility of breakout occurrence by further using the abnormality risk calculated by each of the unit for determining narrow side impingement flow just below the molten metal surface 151 and the unit for determining solidified shell remelting 152. For example, an alarm may be issued when the abnormality risk of the abnormality probability output by each of the unit for detecting narrow side impingement flow just below the molten metal surface 141 and the unit for detecting solidified shell remelting 142 is 80% or higher for a predetermined number of times in succession.

[0141] (Abnormality detection processing) Next, a description will be given of the flow of an anomaly detection process performed by the anomaly detection device 100. Fig. 13 is a flowchart illustrating an example of the flow of an anomaly detection process.

[0142] In step S21, the data acquiring unit 121 acquires temperature measurements from the thermometers arranged on the copper plates constituting the short sides of the mold 2. The data acquiring unit 121 also acquires casting control information from the continuous casting machine. The casting control information is information that indicates, for example, the magnetic field strength of the electromagnetic stirrer 5, the degree of superheat of the molten steel, the height of the tundish, the width of the mold, the casting speed, etc. The casting speed corresponds to the speed at which the slab moves.

[0143] In step S22, the casting length calculation unit 123 calculates the casting length at the position of each thermometer based on the casting speed included in the control information acquired in step S21.

[0144] At this time, the casting length calculation unit 123 integrates the casting speed with respect to time from the casting start time to calculate the casting length at the molten metal surface inside the mold 2. Then, the casting length calculation unit 123 calculates the distance between the position of each thermometer and the target position of the molten metal surface level, and specifies the casting length corresponding to the position of each thermometer at each time by subtracting the calculated distance from the casting length at the target position of the molten metal surface level.

[0145] In step S23, the temperature change rate calculation unit 122 calculates the temperature change rate at the position of each thermometer based on the temperature measurement value from each thermometer acquired in step S21. The temperature change rate may be calculated by differential approximation of the time derivative of each temperature measurement value, or may be calculated by estimation using a Kalman filter for a state equation model that represents the relationship between temperature and change rate in each temperature measuring device.

[0146] In step S24, the tracking surface management unit 124 generates a tracking surface.

[0147] At this time, the tracking surface management unit 124 virtually generates a tracking surface, which is a surface perpendicular to the casting direction and the short side, i.e., a surface parallel to the molten metal surface, at a predetermined time interval in the mold 2. Then, the tracking surface management unit 124 moves the tracking surface in the casting direction at a casting speed that corresponds to the moving speed of the slab.

[0148] In step S25, the tracking surface management unit 124 records explanatory variable records corresponding to each tracking surface.

[0149] At this time, the tracking surface management unit 124 specifies the temperature and the temperature change rate within the copper plate of the mold 2 when the tracking surface passes the positions of the thermometers. At this time, as described above, using equations (3) to (7), the tracking surface management unit 124 specifies the temperature and the temperature change rate within the copper plate of the mold 2 when the tracking surface passes the positions of each thermometer.

[0150] That is, the tracking surface management unit 124 calculates the k-th tracking surface TR k The tracking surface management unit 124 calculates the temperature change rate and temperature at the time when the thermometer passes the position of each thermometer, and stores these values ​​as data in the explanatory variable record corresponding to the k-th tracking surface. k+1 The temperature change rate and temperature at the time when the object passes the position of each thermometer are calculated, and these values ​​are stored as data in the explanatory variable record corresponding to the k+1-th tracking surface.

[0151] As shown in FIG. 1, the mold 2 has short sides on the left and right. Therefore, the explanatory variable record corresponding to the left short side and the explanatory variable record corresponding to the right short side are aligned along the tracking plane TR. k , tracking surface TR k+1 , ... are recorded corresponding to each of them.

[0152] In step S26, a process for detecting a narrow side impingement flow just below the molten metal surface is executed by the narrow side impingement flow detection unit 141. As a result, a probability (abnormality probability) indicating a possibility that a narrow side impingement flow just below the molten metal surface is occurring is calculated for each tracking surface.

[0153] In step S27, a solidified shell remelting detection process is executed by the solidified shell remelting detection unit 142. As a result, a probability (abnormality probability) indicating the possibility that remelting of the solidified shell has occurred is calculated for each tracking surface.

[0154] In step S28, an abnormality determination process is executed by the abnormality determination unit 127. As a result, the abnormality probabilities calculated in the processes of steps S26 and S27 are compared with the abnormality probability threshold parameter, and it is determined whether each abnormality probability represents an abnormality or a normality.

[0155] In addition, the unit for judging the narrow side impingement flow just below the molten metal surface 151 and the unit for judging the solidified shell remelting 152 may simultaneously output an abnormality risk that auxiliary represents the degree of abnormality in the abnormality probability calculated by the unit for detecting the narrow side impingement flow just below the molten metal surface 141 and the unit for detecting the solidified shell remelting 142, respectively.

[0156] In step S29, the warning output unit 128 controls whether or not to output a warning notifying that a breakout is highly likely to occur, based on the result of the determination in step S28.

[0157] In step S29, for example, an alarm may be output when the abnormality probabilities output by the below-surface narrow side collision flow detection unit 141 and the solidified shell remelting detection unit 142 are both determined to be abnormal in the process of step S28. Alternatively, an alarm may be issued when a state in which both of the abnormality probabilities output by each of the units are determined to be abnormal continues a predetermined number of times.

[0158] Furthermore, the presence or absence of an alarm output may be controlled using the abnormality risk calculated by each of the unit for determining the narrow side impingement flow just below the molten metal surface 151 and the unit for determining the solidified shell re-melting 152. For example, an alarm may be issued when the abnormality risk of the abnormality probability output by each of the unit for detecting the narrow side impingement flow just below the molten metal surface 141 and the unit for detecting the solidified shell re-melting 142 is 80% or more for a predetermined number of times in succession.

[0159] In this manner, the abnormality detection process is carried out.

[0160] (Detection of collision flow on the short side directly below the molten metal surface) Next, a description will be given of processing for detecting a narrow side impingement flow just below the molten metal surface by the narrow side impingement flow detection unit 141. Fig. 14 is a flowchart for explaining an example of the flow of processing for detecting a narrow side impingement flow just below the molten metal surface.

[0161] In step S51, the short side collision flow detection unit 141 just below the molten metal surface extracts temperature data for the positions of the first N thermometers from the top out of the thermometers arranged on the short side of the mold 2 from the explanatory variable records recorded in step S25.

[0162] In step S52, the below-the-molten-surface narrow-side impingement flow detection unit 141 acquires the casting control data from the explanatory variable records recorded in step S25.

[0163] In step S53, the below-the-molten-surface narrow-side impingement flow detection unit 141 inputs the temperature data extracted in step S51 and the casting control data extracted in step S52 into the machine learning model.

[0164] In step S54, the unit 141 for detecting narrow side impingement flow just below the molten metal surface calculates the abnormality probability.

[0165] In this manner, the process of detecting the narrow side impingement flow just below the molten metal surface is executed for each of the explanatory variable records recorded in step S25.

[0166] Next, a description will be given of a solidified shell remelting detection process performed by the solidified shell remelting detection unit 142. Fig. 15 is a flowchart illustrating an example of the flow of the solidified shell remelting detection process.

[0167] In step S71, the solidified shell remelting detection unit 142 extracts temperature data for the positions of the thermometers up to the top Nth from the explanatory variable records recorded in step S25.

[0168] In step S72, the solidified shell remelting detection unit 142 extracts data on the temperature change rates at the positions of the M-th thermometers from the bottom, from the explanatory variable records recorded in step S25.

[0169] In step S73, the solidified shell remelting detection unit 142 acquires the casting control data from the explanatory variable records recorded in step S25.

[0170] In step S74, the solidified shell remelt detection unit 142 inputs the temperature data extracted in step S71, the temperature change rate data extracted in step S71, and the casting control data extracted in step S73 into a machine learning model.

[0171] In step S75, the solidified shell remelting detection unit 142 calculates the abnormality probability.

[0172] In this manner, the solidified shell re-melting detection process is executed for each of the explanatory variable records recorded in step S25.

[0173] (Abnormality determination process) Next, a description will be given of the abnormality determination process performed by the abnormality determination unit 127. Fig. 16 is a flowchart illustrating an example of the flow of the abnormality determination process.

[0174] In step S91, the abnormality determination unit 127 acquires an abnormality probability threshold parameter. As a result, the abnormality probability threshold parameter used by the short side impingement flow just below the molten metal surface determination unit 151 and the abnormality probability threshold parameter used by the solidified shell re-melting determination unit 152 are acquired from the parameter setting unit 126.

[0175] In step S92, the unit for judging the narrow side impingement flow just below the molten metal surface 151 calculates the abnormality probability calculated by the unit for detecting the narrow side impingement flow just below the molten metal surface, and sets the abnormality probability threshold parameter p θ Compare with 1.

[0176] In step S93, the judging unit 151 for judging the narrow side impingement flow just below the molten metal surface judges whether the flow is normal or abnormal based on the comparison result in step S92. At this time, if the abnormality probability calculated by the detecting unit 141 for the narrow side impingement flow just below the molten metal surface in step S54 is pT1, then pT1≧p θ If 1, the abnormality probability value is determined to represent an abnormality, and pT1 <p θ If it is 1, the abnormality probability value is determined to represent normal.

[0177] This judgment is made after the tracking surface corresponding to the explanatory variable record for which the abnormality probability to be judged has been calculated has passed the positions of the Nth thermometer from the top, but before it passes the bottom end of the short side of the mold 2.

[0178] In step S94, the unit for judging the narrow side impingement flow just below the molten metal surface 151 simultaneously outputs an abnormality risk level which auxiliary indicates the degree of abnormality of the abnormality probability calculated in step S54 by the unit for detecting the narrow side impingement flow just below the molten metal surface 141. Note that the process of step S94 may not be executed.

[0179] In step S95, the solidified shell remelting determination unit 152 converts the abnormality probability calculated by the solidified shell remelting detection unit 142 into an abnormality probability threshold parameter p θ Compare with 2.

[0180] In step S96, the solidified shell remelting determination unit 152 determines whether the solidified shell remelting is normal or abnormal based on the comparison result of step S95. At this time, if the abnormality probability calculated by the solidified shell remelting detection unit 142 in step S75 is pT2, pT2≧p θ In the case of 2, the abnormality probability value is determined to represent an abnormality, and pT2 <p θIn the case of 2, the abnormality probability value is determined to represent normal.

[0181] This determination is made after the tracking surface corresponding to the explanatory variable record for which the abnormality probability being determined has been calculated has passed the position of the lowest placed thermometer, but before it passes the bottom end of the short side of the mold 2.

[0182] In step S97, the solidified shell remelting determination unit 152 simultaneously outputs an abnormality risk that auxiliary represents the degree of abnormality in the abnormality probability calculated in step S75 by the solidified shell remelting detection unit 142. Note that the process of step S97 may not be executed.

[0183] In this manner, the abnormality determination process is executed. The abnormality determination process is executed for each of the abnormality probabilities calculated in the process of detecting the narrow side impingement flow just below the molten metal surface and the process of detecting remelting of the solidified shell. Therefore, for each of the explanatory variable records recorded in step S25, the presence or absence of an abnormality related to the occurrence of the narrow side impingement flow just below the molten metal surface and the presence or absence of an abnormality related to the occurrence of remelting of the solidified shell are determined.

[0184] (Effects of this embodiment) According to this embodiment, the cause of the occurrence of a breakout can be detected more accurately. That is, it is possible to detect both the delay in the growth of the solidified shell caused by the collision of the molten steel flow with the short side of the mold just below the molten steel surface, and the progress of remelting of the solidified shell. This makes it possible to more accurately predict the occurrence of a breakout.

[0185] Furthermore, the occurrence of a breakout requires the halting of casting, the withdrawal of the cast piece, the removal of the molten steel that has flowed out, and the replacement of the segments, and therefore it takes a long time to return to normal operation, which becomes a factor in reducing productivity.

[0186] According to the present embodiment, it is possible to output an alarm for an abnormality before a breakout occurs in the slab that has come out of the mold, thereby making it possible to improve the efficiency of operations related to casting.

[0187] Furthermore, in this embodiment, an alarm is output by referring to the abnormality probability value related to the occurrence of a narrow side impingement flow just below the molten metal surface and the abnormality probability value related to the occurrence of remelting of the solidified shell. This makes it possible to reliably output an alarm when an abnormality occurs and to minimize the output of an alarm due to erroneous detection.

[0188] (Verification results based on actual measurements) The machine learning model of the short side collision flow detection unit 141 just below the molten metal surface and the machine learning model of the solidified shell remelting detection unit 142 were verified using verification data consisting of 2,536 pieces of past casting data prepared separately from the training data.

[0189] In the verification of the machine learning model of the short side impingement flow detection unit 141, 127 pieces of data were labeled with an abnormality (occurrence of a short side impingement flow just below the molten metal surface) out of 2536 pieces of verification data. θ When the abnormality probability was calculated while changing p, the maximum value that can accurately determine that all abnormal data is abnormal was p θ This value was supplied to the short side impingement flow judgment unit 151 just below the molten metal surface as an abnormality probability threshold parameter to judge an abnormality.

[0190] In the verification of the machine learning model of the solidified shell remelting detection unit 142, 11 pieces of data were labeled with an abnormality (occurrence of solidified shell remelting) out of 2536 pieces of verification data. θ When the abnormality probability was calculated while changing p, the maximum value that can accurately determine that all abnormal data is abnormal was p θ =0.357. This value was supplied to the solidified shell remelting determination unit 152 as the abnormality probability threshold parameter to perform abnormality determination.

[0191] FIG. 17 shows the verification results of the machine learning model of the short side impingement flow detection unit 141 just below the molten metal surface and the verification results of the machine learning model of the solidified shell remelting detection unit 142.

[0192] 17, in the judgment by the short side impingement flow judgment unit 151, 5 out of 2409 data items labeled as normal (normal data items) were erroneously judged as abnormal. On the other hand, all of the 127 data items labeled as abnormal (abnormal data items) were judged as abnormal.

[0193] In addition, 37 out of 2525 data items labeled as normal (normal data items) were erroneously determined to be abnormal by the solidified shell remelting determination unit 152. On the other hand, all 11 data items labeled as abnormal (abnormal data items) were determined to be abnormal.

[0194] (Simulation results) Furthermore, a simulation was performed on data acquired during casting using the anomaly detection device 100. The target casts were 2 casts in which a breakout occurred and 101 cast in which casting was completed normally.

[0195] The simulation was performed for three cases: a case in which an alarm is output based only on the judgment result of the short side collision flow judgment unit 151 just below the molten metal surface (case a); a case in which an alarm is output based only on the judgment result of the solidified shell re-melting judgment unit 152 (case b); and a case in which an alarm is output based on the judgment results of both the short side collision flow judgment unit 151 just below the molten metal surface and the solidified shell re-melting judgment unit 152 (case c).

[0196] Here, a tracking surface is generated every 25 mm of casting, and when it reaches a position 125 mm from the molten metal surface, the abnormality determination unit 127 (the unit 151 for determining short side collision flow just below the molten metal surface or the unit 152 for determining solidified shell re-melting) is operated to determine an abnormality based on the data of the explanatory variable record. Also, the alarm output unit 128 is configured to output an alarm if the determination result of the abnormality determination unit 127 is abnormal 10 times in a row.

[0197] The simulation results are shown in Fig. 18. Fig. 18 shows the number of alarm output casts, the number of alarm output (casts with BO occurrence), the number of alarm output (normal casts), and the alarm output casting length ratio (casts with alarm occurrence) for each of case a, case b, and case c.

[0198] The number of casts for which an alert has been issued indicates the number of casts for which an alert has been issued out of all casts.

[0199] The number of times an alarm was output (cast where a breakout occurred) indicates the number of times an alarm was output during a cast where a breakout (BO) occurred.

[0200] The number of times an alarm was output (normal cast) indicates the number of times an alarm was output in a cast in which no breakout occurred (normal cast).

[0201] The alarm output casting length ratio (alarm generating cast) indicates the ratio of the casting length for which an alarm was output to the total casting length in the cast for which an alarm was output.

[0202] In addition, in all of the simulations of cases a, b, and c, an alarm was issued for all casts in which a breakout occurred.

[0203] In case a, the total number of times that the alarm was issued was 30 times in 7 casts, the number of times that the alarm was issued in the casts where BO occurred was 22, and the number of times that the alarm was issued in the normal casts was 8. However, the ratio of the casting length at which the alarm was issued was 4.68%, which means that false positives are not sufficiently suppressed.

[0204] In case b, the total number of times that an alarm was issued was 3641 times in 83 casts, the number of times that an alarm was issued in casts where BO occurred was 66, and the number of times that an alarm was issued in normal casts was 3575. In other words, in case b, an alarm was issued in more than 80% of casts, and it cannot be said that false positives were sufficiently suppressed.

[0205] In case c, the total number of alarms was 33 in 7 casts, 10 alarms in BO casts, and 23 alarms in normal casts. The alarm output casting length ratio was 0.32%. Both the alarm output number and the alarm output casting length ratio showed that false positives were sufficiently suppressed.

[0206] In this way, outputting an alarm based on the judgment result of only one of the judgment sections results in a large number of false positives. In contrast, it was confirmed that outputting an alarm based on the logical product of the judgment results of both judgment sections can significantly reduce false positives.

[0207] <Example of software implementation> The functions of the anomaly detection device 100 can be realized by a program for causing a computer to function as the device, and by a program for causing a computer to function as each block of the device.

[0208] Fig. 19 is a block diagram illustrating a physical configuration of a computer 500 used as the anomaly detection device 100. As shown in Fig. 19, the anomaly detection device 100 can be configured by a computer 500 including a bus 510, a processor 501, a main memory 502, an auxiliary memory 503, a communication interface 504, and an input / output interface 505. The processor 501, the main memory 502, the auxiliary memory 503, the communication interface 504, and the input / output interface 505 are connected to each other via the bus 510. An input device 506 and an output device 507 are connected to the input / output interface 505.

[0209] The processor 501 may be, for example, a CPU (Central Processing Unit), a microprocessor, a digital signal processor, a microcontroller, or a combination of these.

[0210] The main memory 502 may be, for example, a semiconductor RAM (random access memory).

[0211] For example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination of these may be used as the auxiliary memory 503. A program for causing the processor 501 to execute the above-described operation of the anomaly detection device 100 is stored in the auxiliary memory 503. The processor 501 loads the program stored in the auxiliary memory 503 onto the main memory 502, and executes each instruction included in the loaded program.

[0212] The communication interface 504 is an interface for connecting to a network. As the input / output interface 505, for example, a USB interface, a short-range communication interface such as infrared or Bluetooth (registered trademark), or a combination of these is used.

[0213] The input device 506 may be, for example, a keyboard, a mouse, a touch pad, a microphone, or a combination of these. The output device 507 may be, for example, a display, a printer, a speaker, or a combination of these.

[0214] When the functions of the anomaly detection device 100 are realized by a program for causing a computer to function as the device, the functions described in each of the above embodiments are realized by executing the above program by a processor 501 and a main memory 502.

[0215] The program may be non-transitory and may be recorded in one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be provided to the device via any wired or wireless transmission medium.

[0216] In addition, some or all of the functions of the above-mentioned blocks may be realized by logic circuits. For example, an integrated circuit in which a logic circuit that functions as each of the above-mentioned control blocks is formed is also included in the scope of the present invention.

[0217] Furthermore, some or all of the functions of each of the above blocks may operate on the above device, or may operate on another device (for example, an edge computer or a cloud server, etc.).

[0218] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

[0219] 〔summary〕 An anomaly detection device according to a first aspect of the present invention is an anomaly detection device that detects anomalies in casting by a continuous casting machine that casts a rectangular slab, the continuous casting machine having a rectangular mold having long sides and short sides, an immersion nozzle that supplies molten steel into the mold from an outlet opening toward the short side of the mold, and thermometers that are provided at a plurality of positions on each of the two short sides of the mold including positions higher and lower than a target level of the molten steel surface of the molten steel and measure the temperature inside the short sides of the mold, the anomaly detection device having a temperature change rate calculation unit that calculates a temperature change rate at each thermometer position from the temperature measured by the thermometer, a casting length calculation unit that calculates a casting length from the start of casting based on the casting speed of the continuous casting machine, and a tracking surface that is a surface parallel to the molten steel surface in the mold at predetermined time intervals. the tracking surface management unit imaginarily generates a tracking surface and moves the tracking surface in the casting direction at the casting speed, thereby acquiring data on the temperature and temperature change rate inside the short sides when the tracking surface passes the position of the thermometer, and stores the acquired temperature and temperature change rate data in explanatory variable records corresponding to each short side of the tracking surface, together with casting control data related to control of the continuous casting machine; a probability value calculation unit calculates a probability value representing the possibility that an abnormality has occurred for each short side of the tracking surface based on the data stored in the explanatory variable records; and an abnormality determination unit determines whether a breakout has occurred, in which a part of the cast piece released from the mold has broken, based on the probability value calculated by the probability value calculation unit.

[0220] The anomaly detection device according to aspect 2 of the present invention is in the above-mentioned aspect 1, wherein the probability value calculation unit comprises a short side collision flow detection unit just below the molten metal surface that calculates, for each short side of the tracking surface that has passed a thermometer that is positioned higher than a predetermined first position among the thermometers, a probability value representing a possibility that a short side collision flow just below the molten metal surface has occurred, in which a molten steel flow from the molten steel supplied from the submerged nozzle collides with the short side just below the molten metal surface, and a solidified shell remelt detection unit that calculates, for each short side of the tracking surface that has passed a thermometer that is positioned lowest among the thermometers, a probability value representing a possibility that re-melting of the solidified shell has occurred, and the anomaly determination unit determines whether or not the breakout will occur before the tracking surface moves outside the mold.

[0221] In the anomaly detection device according to aspect 3 of the present invention, in the above aspect 2, the narrow side collision flow detection unit directly below the molten metal surface calculates a probability value representing the possibility that the narrow side collision flow directly below the molten metal surface is occurring based on the temperature data measured by each of the thermometers at a position higher than the first position and the casting control data, and the solidified shell re-melting detection unit calculates a probability value representing the possibility that re-melting of the solidified shell is occurring based on the temperature data measured by each of the thermometers at a position higher than the first position and the temperature change rate data calculated based on the temperature values ​​measured by each of the thermometers at a position lower than a predetermined second position, and the casting control data.

[0222] The anomaly detection device according to a fourth aspect of the present invention, in accordance with the above-mentioned second or third aspect, further comprises an alarm output unit which controls output of an alarm related to the occurrence of a breakout, and when the abnormality probability calculated by the narrow side collision flow detection unit just below the molten metal surface is equal to or greater than a threshold value and the abnormality probability calculated by the solidified shell remelting detection unit is equal to or greater than a threshold value, the abnormality determination unit determines that a breakout will occur, in which a hole will be broken in a part of the slab released from the mold, and the alarm output unit outputs the alarm based on the determination result by the abnormality determination unit.

[0223] The anomaly detection device according to aspect 5 of the present invention, in accordance with aspect 4 above, uses data of explanatory variable records acquired in past castings as verification data, and the anomaly determination unit determines all states in which anomalies occur in the verification data as abnormal, and further includes a threshold setting unit that sets an optimal threshold value in the sense of minimizing the probability of erroneously determining a normal state as abnormal.

[0224] The anomaly detection device according to aspect 6 of the present invention, in accordance with aspect 2 or 3 above, further comprises an alarm output unit which controls output of an alarm related to the occurrence of a breakout, wherein the anomaly determination unit further calculates an abnormality risk which auxiliary represents the degree of abnormality on the tracking surface using the abnormality probability calculated by the short side collision flow detection unit just below the molten metal surface and the solidified shell remelting detection unit, respectively, and the alarm output unit outputs the alarm based on the determination result by the anomaly determination unit and / or the abnormality risk.

[0225] In the anomaly detection device according to aspect 7 of the present invention, in the above aspects 2 to 6, the probability value calculation unit uses a machine learning model that learns model parameters using data of explanatory variable records obtained in past castings as training data.

[0226] An anomaly detection method according to an eighth aspect of the present invention is an anomaly detection method for an anomaly detection device that detects an anomaly in casting by a continuous casting machine that casts a rectangular slab, the continuous casting machine having a rectangular mold having long sides and short sides, an immersion nozzle that supplies molten steel into the mold from an outlet opening toward the short side of the mold, and thermometers that are provided at a plurality of positions on each of the two short sides of the mold including positions higher and lower than a target level of the molten steel surface of the molten steel and measure the temperature inside the short sides of the mold, the anomaly detection method including a temperature change rate calculation step of calculating a temperature change rate at each thermometer position from the temperature values ​​measured by the thermometers, a casting length calculation step of calculating a casting length from the start of casting based on the casting speed of the continuous casting machine, and a tracking surface that is a surface parallel to the molten steel surface in the mold at predetermined time intervals. The method includes a tracking surface management step of virtually generating a tracking surface and moving the tracking surface in the casting direction at the casting speed, thereby acquiring data on the temperature and temperature change rate inside the short sides when the tracking surface passes the position of the thermometer, and storing the acquired temperature and temperature change rate data in explanatory variable records corresponding to each short side of the tracking surface, together with casting control data related to control of the continuous casting machine; a probability value calculation step of calculating a probability value indicating the possibility that an abnormality has occurred for each short side of the tracking surface based on the data stored in the explanatory variable records; and an abnormality determination step of determining whether a breakout has occurred, in which a hole has been broken in a part of the cast piece released from the mold, based on the probability value calculated in the probability value calculation step.

[0227] A program according to an eighth aspect of the present invention causes a computer to execute each step of the anomaly detection method according to the eighth aspect. [Explanation of symbols]

[0228] 1. Continuous casting machine 2. Mold 3 Submerged Nozzle 4 Discharge hole 100 Anomaly detection device 121 Data Acquisition Section 122 Temperature change rate calculation section 123 Casting length calculation section 124 Tracking surface management unit 125 Probability Value Calculation Unit 126 Parameter setting section 127 Abnormality determination section 128 Alarm output section 141 Detector for collision flow on narrow side directly below molten metal surface 142 Solidified shell remelting detector 151 Short side collision flow judgment section just below the molten metal surface 152 Solidified shell remelting judgment part

Claims

1. An abnormality detection device for detecting abnormalities in casting by a continuous casting machine that casts a rectangular cast piece, The continuous casting machine comprises: A rectangular mold having a long side and a short side; an immersion nozzle for supplying molten steel into the mold from a discharge port that opens toward a short side of the mold; a thermometer provided at a plurality of positions including a position higher than a target level of the molten steel surface on each of the two short sides of the mold and a position lower than the target level of the molten steel surface on each of the two short sides of the mold for measuring a temperature inside the short sides of the mold; a temperature change rate calculation unit that calculates a temperature change rate at the position of each thermometer from the temperature measurement value of the thermometer; a casting length calculation unit that calculates a casting length from the start of casting based on the casting speed of the continuous casting machine; a tracking surface management unit which virtually generates a tracking surface, which is a surface parallel to the molten metal surface, at a predetermined time interval within the mold, moves the tracking surface in the casting direction at the casting speed, thereby acquiring data on the temperature and the temperature change rate inside the short side when the tracking surface passes the position of the thermometer, and stores the acquired temperature and temperature change rate data in explanatory variable records corresponding to each short side of the tracking surface together with casting control data related to the control of the continuous casting machine; a probability value calculation unit that calculates a probability value representing a possibility that an abnormality has occurred for each short side of the tracking surface based on the data stored in the explanatory variable record; an abnormality determination unit that determines whether or not a breakout occurs, in which a part of a cast piece ejected from the mold breaks, based on the probability value calculated by the probability value calculation unit; An anomaly detection device comprising:

2. The probability value calculation unit a short side collision flow detection unit directly below the molten metal surface that calculates a probability value representing a possibility that a short side collision flow directly below the molten metal surface occurs, in which a molten steel flow caused by molten steel supplied from the immersion nozzle collides with the short side directly below the molten metal surface, for each short side of the tracking surface that has passed through a thermometer located higher than a first position among the thermometers; and a solidified shell remelting detection unit that calculates a probability value representing a possibility that remelting of the solidified shell has occurred for each short side of the tracking surface that has passed through the lowest thermometer among the thermometers, The abnormality detection device according to claim 1 , wherein the abnormality determination unit determines whether or not the breakout occurs before the tracking surface moves out of the mold.

3. The short side collision flow detection unit directly below the molten metal surface is calculating a probability value representing a possibility that the narrow side impingement flow just below the molten metal surface is occurring based on data of temperatures measured by each of the thermometers at a position higher than the first position and the casting control data; The solidified shell remelting detection unit includes:

3. The anomaly detection device according to claim 2, further comprising: a temperature detection unit that detects whether or not the solidified shell has remelted; a temperature change rate data calculated based on the temperature data measured by each of the thermometers at a position higher than the first position, and the temperature change rate data calculated based on the temperature values ​​measured by each of the thermometers at a position lower than a predetermined second position; and a probability value indicating a possibility that the solidified shell has remelted, based on the casting control data.

4. An alarm output unit that controls output of an alarm related to the occurrence of a breakout, The abnormality determination unit is when the abnormality probability calculated by the narrow side collision flow detection unit just below the molten metal surface is equal to or greater than a threshold value and the abnormality probability calculated by the solidified shell remelting detection unit is equal to or greater than a threshold value, it is determined that a breakout will occur, in which a hole is broken in a part of the slab that has come out of the mold; The abnormality detection device according to claim 2 , wherein the alarm output unit outputs the alarm based on a result of the determination made by the abnormality determination unit.

5. 5. The anomaly detection device according to claim 4, further comprising a threshold setting unit that uses data of explanatory variable records acquired in past castings as verification data, and that determines, as abnormal, all states in which an anomaly occurs in the verification data, and sets an optimal threshold value in the sense of minimizing a probability of erroneously determining a normal state as abnormal.

6. An alarm output unit that controls output of an alarm related to the occurrence of a breakout, The abnormality determination unit is further calculating an abnormality risk level which auxiliary represents the degree of abnormality on the tracking surface by using the abnormality probability calculated by the short side collision flow detection unit just below the molten metal surface and the solidified shell remelting detection unit, The anomaly detection device according to claim 2 , wherein the alarm output unit outputs the alarm based on a result of determination by the anomaly determination unit and / or the degree of abnormality risk.

7. The anomaly detection device according to claim 2 , wherein the probability value calculation unit uses a machine learning model that learns model parameters by using data of explanatory variable records acquired in past castings as training data.

8. An abnormality detection method for an abnormality detection device that detects abnormalities in casting by a continuous casting machine that casts a rectangular cast piece, The continuous casting machine comprises: A rectangular mold having a long side and a short side; an immersion nozzle for supplying molten steel into the mold from a discharge port that opens toward a short side of the mold; a thermometer provided at a plurality of positions including a position higher than a target level of the molten steel surface on each of the two short sides of the mold and a position lower than the target level of the molten steel surface on each of the two short sides of the mold for measuring a temperature inside the short sides of the mold; a temperature change rate calculation step of calculating a temperature change rate at each thermometer position from the temperature measurement value of the thermometer; A casting length calculation step of calculating a casting length from the start of casting based on the casting speed of the continuous casting machine; a tracking surface management step of virtually generating a tracking surface, which is a surface parallel to the molten metal surface, in the mold at a predetermined time interval, and moving the tracking surface in the casting direction at the casting speed to obtain data on the temperature and the temperature change rate inside the short side when the tracking surface passes the position of the thermometer, and storing the obtained temperature and temperature change rate data in explanatory variable records corresponding to each short side of the tracking surface together with casting control data related to the control of the continuous casting machine; a probability value calculation step of calculating a probability value representing a possibility that an abnormality has occurred for each short side of the tracking surface based on the data stored in the explanatory variable record; an abnormality determination step of determining whether or not a breakout occurs, in which a part of a cast piece discharged from the mold breaks, based on the probability value calculated in the probability value calculation step; The anomaly detection method includes:

9. A program causing a computer to execute each step of the anomaly detection method according to claim 8.

Images