JPH04172160A - Method for predicting restrained breakout in continuous casting - Google Patents

Abstract

PURPOSE:To improve prediction accuracy by forming the positional distribution pattern of information indicating the presence of abnormal temp. rising from temp. information at the different positions on the mold wall and recognizing similarity with the prepared reference pattern. CONSTITUTION:Plural temp. detecting means are buried in the mutually different positions on the mold wall in continuous casting equipment. From the temp. information obtd. by each temp. detecting means, the presence of the abnormality temp. rising at each position is identified and the positional distribution pattern of information indicating the presence of the abnormal temp. rising is formed. The presence of the similarity between this pattern and the prepared reference pattern is decided and in the case the similarity is present thereon, the presence of breakout is decided. The positional distribution patterns at the time of developing the breakout and at the time of developing no breakout, are inputted in neural net, respectively, and learning effect is held in the reference pattern to improve the hitting ratio of the prediction.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to the prediction of restrictive breakouts that may occur in continuous casting.

[Prior Art] There are various types of breakouts that occur during continuous casting, but regarding restrictive breakouts, there are
It is difficult to predict and prevent it because it occurs suddenly even when there is no apparent change in operating conditions. In other words, restraint breakout is thought to occur when a part of the shell formed by solidifying molten steel sticks to the mold and is torn off. There are no changes above.

A method for predicting restrictive breakout is disclosed, for example, in Japanese Patent Publication No. 47545/1983. In this method, the temperature of each part of the mold is measured using multiple thermocouples embedded in the mold, and if the temperature detected by one thermocouple rises by one point above the average detected temperature and then falls,
When a similar temperature change pattern is detected by other adjacent thermocouples, it is considered that there is a possibility of a breakout.

[Problem to be solved by the invention] However, in the conventional restrictive breakout prediction method, the prediction accuracy is relatively low, and even if a breakout does not actually occur, there is a high possibility that it will be mistakenly recognized as a breakout. .

If a restraining breakout is predicted to occur in continuous casting, drawing of the cast steel must be stopped immediately to prevent it. However, - Temporary stopping of drawing of cast steel is accompanied by a partial change in the material of the cast steel, so the parts formed by stopping drawing cannot be made into products, resulting in a decrease in yield. . Therefore,
If there is a high possibility of erroneously recognizing a breakout, the yield will be reduced more than necessary.

In reality, the breakout accuracy rate in conventional prediction methods is about 2 to 4%.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to improve the prediction accuracy of a method for predicting restraint breakout in continuous casting, prevent breakout from occurring, and at the same time prevent a decrease in yield.

[I! Means for Solving the Problems] In order to solve the above problems, the present invention monitors the temperature at each position detected by a plurality of temperature detection means embedded at different positions in the mold wall of continuous casting equipment. In a method for predicting a restrictive breakout, the presence or absence of an abnormal temperature rise at each position is identified from the temperature information obtained by each temperature detection means, and a position distribution pattern of information indicating the presence or absence of an abnormal temperature rise is formed. By recognizing the similarity between this position distribution pattern and a reference pattern prepared in advance, the presence or absence of a breakout is predicted.

[Function] In the present invention, attention is paid to the fact that the positional distribution pattern of the occurrence points of the abnormal change in the cast steel temperature (temporal temperature rise) that appears prior to the restraint breakout has a 7-shape or a U-shape. are doing. That is, by pattern recognition of this position distribution pattern, it is possible to identify whether the pattern is a breakout or not, and to predict a breakout with a high accuracy rate.

In a preferred embodiment of the present invention, which will be described later, the following method is further used to improve breakout prediction accuracy.

(a) The position distribution pattern to be processed is constructed by combining instantaneous detection patterns obtained at different times. In the instantaneous detection pattern, there is only a single 7-shape, but by superimposing one time series data, a more complex pattern can be recognized, improving the accuracy of prediction.

(b) The position of each pattern is made to coincide with a predetermined reference point, which is the position of the element where an abnormal temperature rise was first detected. In pattern recognition, a positional shift between a reference pattern and a recognition target pattern causes a large error in the recognition result.

In continuous casting, the position where restrictive breakout occurs is indefinite, so if the position of each element of the reference pattern is fixedly associated with the element position of each detector on a one-to-one basis,
Although positional shifts occur in response to actual pattern positional changes and errors in pattern recognition are likely to occur, the accuracy of breakout prediction can be improved by aligning the reference pattern and the detection target pattern.

(c) After inputting the position distribution pattern when a breakout occurred and the position distribution pattern when a breakout did not occur, which were actually measured in the past, to the neural network and learning them as training data. Input the position distribution pattern of the detection target and predict the presence or absence of a breakout.

By learning both the position distribution pattern when a breakout occurs and the position distribution pattern when a breakout does not occur, the incidence of misrecognition is reduced and the accuracy of prediction is improved.

Other objects and features of the present invention will become clear from the following description of embodiments with reference to the drawings.

[Example] Figure 5a shows the appearance of a mold used in continuous casting.

A cooling water passage is formed inside the wall of this mold.
The cooling water cools the molten steel in the mold. The molten steel injected into the mold gradually solidifies within the mold as cooling progresses, and is slowly drawn out from below the mold at a constant speed. The casting width varies depending on the specifications of the finished product, but in this example, the maximum casting width is 222cm.
It is.

Next, the progress of the process when a restrictive breakout occurs will be described with reference to FIG. In Figure 6, the process is (1)-(2)-(3)
Proceed in the order of -(4) and -(5).

The molten steel in the mold E solidifies sequentially from the portion along the inner wall of the mold that is cooled to form a shell B. Since the shell B is pulled out from below the mold, it progresses downward, and as it moves downward, it grows and becomes thicker. Since there is an inclusion called powder between shell B and mold E, shell B is separated from the wall surface of mold E. However, in the vicinity of meniscus F, part A of the shell sometimes falls into mold E. It may stick. The shell fixed to mold E will be restrained and will not move.
When a force is applied to pull out shell B from below, the shell breaks in the middle, and part of shell A remains stuck to the mold wall.
The other part of the shell B advances downward (1).

Opening formed by shell rupture (between A and B)
In this case, the molten steel directly contacts the die wall and is rapidly cooled, so a new shell C is formed in that area (2).

The new shell C sticks to both the lower shell B and the upper restrained shell A, but as the lower shell B moves downward, the new shell C, which is thinner, is torn off by the force. Part C fixed to lower shell B
It is divided into a and a part cb fixed to the upper shell A (
3).

The molten steel again enters between the separated shells Ca and Cb, contacts the mold wall, is cooled, and forms a new shell, which is then torn off again.

These processes are repeated (4).

The location where the new shell is formed gradually moves down the mold. Breakout occurs when the position reaches the bottom of mold E (5).

FIG. 7 shows the appearance of a slab that suffered a restrictive breakout during casting. Referring to FIG. 7, it can be seen that the portion where the restrictive breakout has occurred forms a substantially V-shape. Because the shell thickness is very thin where breakout occurs, the temperature at the mold wall is higher than normal. The part where such an abnormal temperature rise occurs (the thin part of the shell) gradually moves downward.

Therefore, in this invention, the temperature is measured at each position of the mold, attention is paid to the part where an abnormal temperature rise has occurred, and the restraint breakout is predicted by pattern recognition of the position distribution pattern. .

In this embodiment, a large number of thermocouples TC are embedded in each of the four sides of the mold, as shown in FIGS. 5b and 5c. Specifically, a rod made of constantan is coated with an insulating material except for the tip, and the rod is inserted and fixed into holes formed at various positions on a copper plate constituting a mold. By processing the electrical signals obtained from each thermocouple, the temperature of each part can be measured.

In this example, among the large number of thermocouples shown in Figures 5b and 5C, only those in the first to fourth stages are used, and the remaining thermocouples are not used. do not have.

FIG. 8 shows an example of a change in temperature detected by the thermocouple at each position when a restrictive breakout occurs. Furthermore, the 8th
In the figure, the reference point is one measurement point, the upwardly related point is one measurement point located above the reference point, and the horizontally related point is one measurement point located laterally with respect to the reference point. be. When a restraining breakout occurs, when the shell rupture occurs, the temperature at that location increases more abnormally than before and then decreases. In Figure 8,
Abnormal temperature increases are cursed at all of the reference point, upward related point, and lateral related point. However, the timing at which the temperature rises is different for each point.

In this embodiment, processing as shown in FIG. 1 is performed to predict a restrictive breakout. Although Figure 1 is in the form of a flowchart, Figure 1 shows the flow of information, and the order in which each process is executed is not accurate; for example, multiple processes may be executed at the same time. Please note that there may be cases where

Each process will be explained with reference to FIG. In process P1, M
The voltage output from each thermocouple attached to the mold is sampled at predetermined intervals, the obtained voltage is converted into digital data, and the data is further processed to obtain temperature information. Further, when the time-series temperature information obtained from each thermocouple is used as learning data, it is temporarily stored and saved in an external storage device.

In process 2, the time-series information on a large number of temperatures obtained in process P1 is processed to detect an abnormal temperature rise at each point. That is,
The time-series change in temperature data at each location is monitored, and it is determined whether the change satisfies the condition of the following equation (1).

T b (t) - T b a (t') ≧ Ts
・-(1) T b (t): Current detected temperature (instantaneous value) Tba (t') Moving average temperature before the knee period (2 seconds) Ts: Temperature deviation threshold The moving average temperature is is determined as the average value of all the instantaneous temperature values detected at n points in the past (in this example, n=
16:32 seconds average).

If the condition of formula (1) is satisfied, then the following formula (
Using equations 2) and 3, it is determined whether a predetermined temperature drop occurs within a predetermined time.

(Tb(t)-Tb(t'))/2 < b...
(2) tbl −tbo ≦ c ・
-・(3) T b (t): Current detected temperature (instantaneous value) T b (t'): Previous detected temperature (instantaneous value) Tba (t') Moving average temperature before the knee period (2 seconds) tbl: Time tb□ when the condition of equation (2) is satisfied
: The time at which the condition of the formula (1) is satisfied, c: The constant formula (2) is the time when the condition of the formula (1) is satisfied continuously for a predetermined period of time Process P
In step 4, for each measurement point that satisfies all the conditions of equations (1), (2), and (3) above, identify the time at which the time-series change in temperature peaks, and calculate the time at which the temperature changes over time reach a peak. The curve of the equimixture line passing through that measurement point is determined by calculating the temperature detected at each measurement point.

That is, 2. For example, as shown in FIG. 2a, on the plane coordinates in which the mold is expanded in the horizontal direction, an equimixture line (C1) passing through the measurement point (Xl) indicated by a black circle is determined. Figure 2a shows five equal crosstalk lines 0.
1 to C5 are drawn, and each of these equimixture lines is generated based on the temperature distribution at different points in time. That is, since the process P4 in FIG. 1 is repeatedly executed based on temperature distribution data at different points in time, a plurality of equal crosstalks are obtained. These equimixture lines are superimposed on one processing plane by processing P5 in FIG. This results in a group of isotherms as shown in FIG. 2a, for example.

In process P6 in FIG. 1, an input data area (in a two-dimensional array) having cells that correspond one-to-one with each cell of an input layer of a neural network, which will be described later, is generated. That is, a memory area for data arrangement is generated that is associated with each position of the cell so that each position coordinate of the two-dimensional pattern to be recognized matches each input cell of the neural network. In this embodiment, the input data area is composed of 25 (horizontal) x lO (vertical) cells, as shown in FIG. 3a.

Further, in process P6, the position is normalized. That is, as is clear from the difference between, for example, FIG. 2a and FIG. 2b, the location where the abnormal temperature change appears is not constant, and this distribution pattern occurs in any part of the developed mold plane. Generally, in pattern recognition, if the positions of a reference pattern and a recognition target pattern are misaligned, the correlation between the two becomes small, making recognition difficult. Therefore, in this example, as shown in Figure 9, for example, the horizontal position of the pattern is aligned to a predetermined reference position (in Figure 9, the center of the pattern is the center of the data area) to facilitate recognition. ing. Also, since this alignment is performed, there is no need to input the entire data of the mold, and parts other than those where abnormal temperature change patterns appear are ignored, and data is partially extracted and processed. Therefore, the amount of data is small, and the number of neurons in the input layer of the neural network is small.

In process P7 of FIG. 1, data is written into each cell of the input data area generated in process P6. That is, the equimixing line obtained in process P4 and each cell position are calculated, and all the cells at the positions where the equimixing line passes are set to the firing state (the black part in FIG. 3a).

Figure 3a shows the cell arrangement of the input data area as a result of processing only the isothermal llAC3 shown in Figure 2a, and Figure 3b shows the result of processing for all the isothermal lines in Figure 2a. .

The data obtained through process P7 is input to the neural network in this embodiment. Neural net is
In this example, as shown in FIG. 1, it is composed of three layers: an input layer, an intermediate layer, and an output layer. The result of processing P7 is applied to the input layer of the neural network. The output layer is provided with two cells: one indicating whether the output is normal or not, and the other indicating the presence or absence of a breakout.

To use a neural network, it must be trained in advance. In this example, saved past data whose results (whether or not a breakout has occurred) are already known is input from an external storage device, data as shown in Figure 3b is generated from that data, and this is used as learning data. is applied to the input layer of the neural network, and a learning process is performed, that is, the weighting of each cell in each layer and the strength of connections are adjusted so that the correct result appears in the output layer. Known pack propagation is used as an algorithm for this learning process. Further, as the neural network device, a personal neurocomputer NEURO-07 manufactured by NEC Corporation was used in this embodiment.

In actual learning, the input data is one-occurrence data (in which a breakout actually occurred) as shown in Figures 2a and 2b, and false alarm data (as shown in Figures 2C and 2d). We used a plurality of samples of each sample (in which no breakout occurred but whose temperature distribution pattern was similar to that of the occurrence data, so a false alarm was likely to occur), and performed learning as many times as necessary to obtain the correct result.

After the above learning process is completed, input the temperature data during the actual continuous casting process and perform the first
By executing the process of PL-P7 shown in the figure and applying the obtained data as prediction target data to each corresponding cell in the input layer of the neural network, the signal appearing in the output layer can be obtained as the prediction result. can.

In the above embodiments, a neural network is used as a pattern recognition means, but the present invention can also be implemented using a general pattern recognition method such as pattern matching. In addition, in the example, as a pattern to be recognized, a plurality of equidistant lines obtained at different times were superimposed, but for example, as shown in FIGS. 3a and 4b, equidistant lines at a certain time The input cover pattern may be generated only by

[Effect] As a result of predicting a restrictive breakout using the method described above, the prediction accuracy was significantly improved compared to the conventional method, and the occurrence of false alarms was reduced.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the contents of processing in an embodiment of the present invention. FIGS. 2a, 2b, 2c, 2d, and 4a are schematic diagrams showing isotherm patterns generated from different input temperature data. FIGS. 3a and 3b are schematic diagrams of data arrays showing the results of converting the data in FIG. 2a into data for each cell in the input data area. FIGS. 4b and 4c are schematic diagrams of data arrays showing the results of converting the data in FIG. 4a into data for each cell in the input data area. FIG. 5a, FIG. 5b, and FIG. 5c are a perspective view, a front view, and a side view, respectively, showing the appearance of the mold of the example. FIG. 6 is a longitudinal sectional view showing the process of occurrence of restrictive breakout. FIG. 7 is a perspective view showing the appearance of a slab in which a restraining breakout has occurred. FIG. 8 is a timing chart showing temperature changes at each measurement point when a restrictive breakout occurs. FIG. 9 is a schematic diagram showing changes in pattern position due to position normalization. TC: Thermocouple Xl: Measurement points 01 to C5: Equal crosstalk applicant Nippon Steel Corporation (1 other person) - Figure L$ BO Door 6 Diagram Voice 7

Claims (4)

[Claims] (1) In a method of monitoring the temperature at each position detected by a plurality of temperature detection means embedded at different positions in a mold wall of continuous casting equipment and predicting a restrictive breakout: by each of the temperature detection means From the temperature information obtained, the presence or absence of an abnormal temperature rise at each location is identified, a position distribution pattern of information indicating the presence or absence of an abnormal temperature rise is formed, and the similarity between this position distribution pattern and a reference pattern prepared in advance is determined. A method for predicting the presence or absence of a breakout in continuous casting by recognizing the following. (2) The method for predicting a restraint breakout in continuous casting according to claim 1, wherein the position distribution pattern of information indicating the presence or absence of abnormal temperature rise is constructed by combining instantaneous detection patterns obtained at different times. (3) By aligning the position of each element in the position distribution pattern and the reference pattern with a predetermined reference point, the position of the element in which abnormal temperature rise was detected first, the position distribution pattern and the reference pattern can be adjusted. 2. The continuous casting restraint breakout prediction method according to claim 1, wherein the position of the continuous casting is adjusted. (4) After inputting the position distribution pattern when a breakout occurred and the position distribution pattern when a breakout did not occur, which were actually measured in the past, to the neural network and learning them as training data. 2. The continuous casting restraint breakout prediction method according to claim 1, wherein the presence or absence of a breakout is predicted by inputting a position distribution pattern of a detection target.