JP2718800B2 - Breakout prediction processing method for continuous casting - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a continuous casting breakout prediction processing method for predicting the occurrence of breakout of a slab in a casting mold in continuous casting. The present invention relates to a method for predicting breakout of continuous casting, which makes it possible to predict with high accuracy and speed, and to make the adjustment process for construction extremely easy.

In the continuous casting process, during the process of molten steel flowing through the mold (mold) and gradually cooling from the outside and solidifying, the thickness of the solidified portion of the molten steel is reduced due to abnormal cooling such as the molten steel burning on the inner wall of the mold. When a thin portion is formed and the portion breaks out of the mold, a breakout occurs in which molten steel flows out. When this breakout occurs, not only the casting process is interrupted but also a serious accident such as damage to the equipment is caused due to the steam explosion of the molten steel or the cooling water that has flowed out. This breakout can be avoided by reducing the speed of drawing the molten steel and lengthening the time required for cooling, but it is not possible to reduce the drawing speed of the molten steel so much that the productivity of the subsequent process does not decrease. That is the current situation.
Therefore, there is a strong demand for providing a high-precision breakout prediction processing method that makes it possible to predict the occurrence of breakout in a continuous casting process with high precision.

[Conventional technology]

Before going into the description of the prior art, referring to FIG.
The mechanism of the occurrence of a breakout will be briefly described using a restrictive breakout as an example.

If normal solidification of molten steel is performed, the 15th
As shown in the figure, the shell thickness increases toward the lower part of the mold, and stable solidification proceeds. During this process, if a restraining force is generated in the vicinity of the meniscus for some reason and the restraining force becomes larger than the shell strength, the shell breaks as shown in FIG. continue,
As shown in (1), molten steel flows out of the gap between the broken shells, and the shell breaks continuously during the positive strip while forming a new thin shell.

In the negative strip period, the shell thickness increases because no breaking force acts on the shell. However, in the positive strip period, a breaking force acts again on the shell, and the shell is broken. As described above, the shell rupture occurs for each mold oscillation, and the shell rupture portion sequentially propagates below the mold and in the width direction according to the casting speed as shown in FIG. Then, as shown in the figure, a breakout (BO in the figure) occurs when the shell break portion is below the lower end position of the mold.

In this way, when the restrictive breakout occurs, as shown in FIG. 16, hot wrinkles generated by solidification of the molten steel flowing out to the slab surface are recognized.

In order to predict this breakout, conventionally, as disclosed in JP-A-58-148064, a plurality of thermocouples are embedded in a mold wall of a continuous casting facility,
An operation is performed in which the detected temperature of one of the thermocouples once rises from the average detected temperature and then falls, and the detected temperature of at least one other thermocouple adjacent to the thermocouple is changed as described above. A method has been adopted in which the occurrence of a breakout is predicted by detecting that an operation with a large temperature change is performed.

That is, when a shell break occurs at the position of the thermocouple buried in the mold, the temperature detected by the thermocouple rapidly rises due to the molten steel flowing out of the mold surface, and then the molten steel flowing out is cooled by the mold to form a new shell. Utilizing the characteristic that the detected temperature decreases by forming a temperature, the detected temperature of one of the thermocouples is set to the detected temperature of the adjacent thermocouple based on the average detected temperature of the embedded thermocouple. In addition, if the operation of increasing and decreasing the average detected temperature is performed, it is determined that a breakout has occurred, and the process of predicting the occurrence of the breakout has been performed.

Further, as another prior art, as disclosed in Japanese Patent Application Laid-Open No. 55-84259, the temperature is measured at one or more places on two or more surfaces among four surfaces of a mold copper plate for continuous casting. And comparing these temperatures with each other, and predicting the occurrence of a breakout while using the temperature difference as an index.

[Problems to be solved by the invention]

However, in such a conventional breakout prediction processing method according to the related art, the difference in the temperature pattern between the time when the breakout symptom occurs and the time when the breakout is normal is determined by the temperature deviation, the temperature rise rate, the temperature fluctuation time difference between adjacent thermocouples, and the like. However, there are the following problems in order to identify the information by a combination of simple level judgments.

That is, the mold surface temperature does not always show a stable constant value even in a steady state, and greatly fluctuates due to changes in operating conditions such as the type of steel, casting speed, metal surface temperature, and inflow of powder. Further, the temperature rise at the time of breakout occurrence is not constant but has a large variation such as 10 ° C. to 100 ° C. Therefore, it is difficult to detect a sign of breakout with high accuracy by using a simple algorithm and determination conditions as in the past, so that false detection and non-detection increase, which is not practical. was there.

To solve this, it is conceivable to adopt a method of building more complicated algorithms and judgment conditions,
Not only is it extremely difficult to construct, but if too complicated algorithms and decision conditions are used, there will be a new problem that signal processing will take too long and breakouts cannot be predicted. Become.

Also, in order to absorb such a change in operating conditions, it is necessary to change the algorithm and determination conditions built under specific operating conditions or adjust the set values in each determination condition. If the method of the prior art is adopted, it is necessary to examine the obtained performance data for each event and perform a change / adjustment process, so that it becomes a trial-and-error process. There was a problem that it would require a lot of labor and time.

The present invention has been made in view of such circumstances,
It is an object of the present invention to provide a new continuous casting breakout prediction method that can quickly and accurately predict occurrence of a breakout in a continuous casting process and that can easily perform an adjustment process for construction.

[Means for solving the problem]

The present invention has been studied recently, while the present inventor has found that although there are variations such as temperature level differences and change timing differences, the breakout phenomenon can be grasped as a unique pattern phenomenon common to all breakouts. Focusing on the fact that a neurocomputer of a certain parallel distributed processing is suitable for identifying this pattern phenomenon,
It is configured to perform a process of predicting the occurrence of a breakout according to a neurocomputer.

FIG. 1 illustrates the principle configuration of the present invention. In the figure, 10 is a mold, 20 is a temperature detecting end disposed on the side wall of the mold, 30 is a network structure part provided for the temperature detecting end 20 and having a temperature detection value detected by the temperature detecting end 20 as an input signal. So,
The one which calculates and outputs an output signal corresponding to the input signal according to the data conversion function of the network structure to which the internal state value is allocated, 31 is an internal state value management unit provided in the network structure unit 30, and executes data conversion processing 32 is a differential operation unit which manages the internal state value required at the time, and calculates differential information such as primary differential, secondary differential and deviation information of the temperature detection value detected by the temperature detection end 20. , 33
Is a time-series data storage unit that holds time-series data of the temperature detection value detected by the temperature detection end 20 and time-series data of the differential information of the temperature detection value output by the differential operation unit 32;
Reference numeral 34 denotes a first operation unit which receives an output signal of the network structure unit 30 associated with at least two or more adjacent temperature detecting terminals 20 as inputs, and performs mutual operation of input output signal values of the network structure unit 30. From the relationship, it is detected whether or not the sign of the breakout has occurred substantially simultaneously at the adjacent temperature detection end 20. Reference numeral 35 denotes a breakout prediction unit, which is connected to the network structure unit 30 and the first calculation unit 34. Equivalent, 36 is a second arithmetic unit, which is a break corresponding to the temperature detecting end 20 located at the same part corresponding to the row of the temperature detecting ends 20 arranged in a two-row configuration or a part near the same. The output signal of the output predicting unit 35 is used as an input, and it is detected from the correlation of the output signal values whether or not the sign of the breakout occurs at the temperature detecting end 20 at the substantially same position in the upper and lower portions at the same time, 40 Is a learning process A network structure unit such that an output signal group from the network structure unit 30 output in response to presentation of a learning input signal group according to a back propagation method becomes a desired teacher signal group, for example. A learning signal storage device 41 for learning the internal state value of the learning processing device 40 is provided for managing a learning signal used by the learning processing device 40.

The first arithmetic unit 34 holds the time-series data of the output signal value of the network structure unit 30 and a buffer unit 341 that outputs the held data indicating the most likely sign of breakout among the held data. , This buffer part 3
The hierarchical network structure unit 342 (which receives the output signal of 41 as an input and outputs a judgment value of the simultaneous occurrence of the sign of breakout in accordance with the data conversion function of the hierarchical network structure (manages the internal state values required when executing the data conversion process) (Including an internal state value management unit 343).

Before describing the operation of the present invention configured as described above, the network structure unit 30 and the hierarchical network structure unit 34 will be described.
A hierarchical network configuration data processing device that executes adaptive data processing and a back propagation method that the learning processing device 40 executes as one example will be described.

In the hierarchical network configuration data processing device, a hierarchical network is configured from a kind of node called a basic unit and an internal connection having a weight value corresponding to an internal state value. FIG. 17 shows the basic configuration of the basic unit 1. The basic unit 1 is a multi-input, one-output system, and a multiplication unit 2 for multiplying a plurality of inputs by respective weights of internal connections, and an accumulation unit 3 for adding all the multiplication results thereof.
And a threshold processing unit 4 that performs a non-linear threshold process on the accumulated value and outputs one final output.

Assuming that the h-th layer is the front tomographic layer and the i-th layer is the rear tomographic layer, the accumulation processing unit 3 executes the operation of the following expression (1), and the threshold processing unit 4 executes the operation of the following expression (2).

Here, h: unit number of h layer i: unit number of i layer p: pattern number of input signal θ _i : threshold value of i unit of i layer W _ih : weight value of internal coupling between _{hi and} i layers x _pi : y _ph : The output from the h-th unit of the h-th layer with respect to the input signal of the p-th pattern y _pi : The output of the i-th layer with respect to the input signal of the p-th pattern Output from the i-th unit Then, in the hierarchical network configuration data processing device,
A large number of basic units 1 having such a configuration are hierarchically connected as shown in FIG. 18 by using an input unit 1 'for directly distributing and outputting an input signal value as an input layer to form a hierarchical network. Thus, a parallel data processing function of converting an input pattern into a corresponding output pattern is exhibited.

The learning processing device 40 performs a process of obtaining a weight value of the hierarchical network, which defines the data conversion function of the hierarchical network configuration data processing device, by a learning process according to the back propagation method.

In the back propagation method, learning is performed by adaptively and automatically adjusting the weight value W _ih and the threshold value θ _i of the hierarchical network by error feedback. (1)
As is apparent from the equation (2), the adjustment of the weight value W _ih and the threshold value θ _i needs to be performed simultaneously, but this operation is a difficult operation that interferes with each other. Accordingly, one of the applicants of the present application has filed a patent application entitled “Japanese Patent Application No. 62-333484 (December 28, 1987)
As disclosed in “Japanese Patent Application,“ Network Configuration Data Processing Device ”), by providing a unit that always outputs“ 1 ”to the h layer on the input side and assigns a threshold θ _i to the output as a weight value. It was suggested to treat the threshold theta _i as the weighting value incorporates threshold theta _i in the weight value W _ih. By doing so, the above (1)
Equation (2) is Will be represented by

Next, a description will be given of a weight value learning processing method by the back propagation method according to the description format of the expressions (3) and (4). Here, this description is made on the assumption that the hierarchical network unit has a hierarchical network having a three-layer structure of the h layer-i layer-j layer shown in FIG.

By analogy with equations (3) and (4), the following equations (5) and (6)
An expression is obtained. That is, Where j: the unit number of the j-th layer W _ji : the weight value of the internal connection between the ij-layers x _pj : the product sum of the inputs from each unit of the i-th layer to the j-th unit of the j-th layer y _pj : of the p-th pattern The output from the j-th unit of the j-th layer with respect to the input signal The learning processing device 40 outputs the output pattern y _pj from the output layer which is output when the input pattern for learning is presented,
When a teacher pattern d _pj (a teacher signal to the j-th unit and the j-th unit with respect to the input signal of the p-th pattern) which is a signal to be taken by the output pattern y _pj is received, first,
The difference between the output pattern y _pj and the teacher pattern d _pj [d _pj −y
_pj ], and then , And then Here, ε: learning constant ζ: momentum t: number of times of learning According to, the update amount ΔW _ji (t) of the weight value between the i-th layer and the j-th layer
Is calculated. Here, for adding those pertaining to the update amount of the weight values determined in the previous update cycle of "ζΔW _ji (t-1)" is order to speed up the learning.

Subsequently, the learning processing device 40 uses the calculated α _pj to
First of all, , Then , The update amount ΔW _ih (t) of the weight value between the h layer and the i layer
Is calculated.

Subsequently, the learning processing unit 40, the weighting value W _ji for according to the update amount calculated as described above in the next update cycle _{(t) = W ji (t} -1) + ΔW ji (t) W ih (t) = W ih By repeating the method of determining (t-1) + ΔW _ih (t), an output pattern y _pj from an output amount output when an input pattern for learning is presented, and the output pattern y weight value W _ji where the teacher pattern d _pj is the signal to be taken by _pj is to be consistent, the process to learn the W _ih.

Then, even in a case where the learning processing device 40 is further configured with multiple layers, according to the same processing, the update amount ΔW of the weight value between the preceding layers is used while using the value obtained from the subsequent stage of the output side and the network output data. It will be processed so as to be determined.

By allocating the weight values learned in this way to the internal connection of the hierarchical network, the hierarchical network configuration data processing device can output a desired teacher from the output layer when an input pattern for learning is presented to the input layer. In addition to processing to output a pattern, even if an unexpected input pattern is input, adaptive data processing of outputting a suitable output pattern from this hierarchical network is executed.

[Action]

When a sign of a breakout occurs, a characteristic temperature distribution pattern based on hot wrinkles occurs on the side wall of the mold 10 as described in FIG. On the other hand, when no sign of breakout occurs, the temperature distribution pattern on the side wall of the mold 10 shows a flat characteristic with relatively little change, although there is a difference depending on the operating conditions.

According to the present invention, an operator who plans a prediction process of a breakout will first arrange a plurality of temperature detecting ends 20, for example, in the lateral direction of the mold 10 in order to detect a temperature distribution pattern on the side wall of the mold 10. . Then, one network structure unit 30 that uses the temperature detection values of these temperature detection terminals 20 as input signals is prepared. The prepared network structure unit 30 performs adaptive data processing according to a data conversion function of a network structure to which internal state values (weight values, threshold values, and the like) are assigned, as in the above-described hierarchical network configuration data processing device. Things.

Next, the operator, from the operation data up to now, the data of the temperature distribution pattern detected by these temperature detection end 20 when a sign of breakout occurs, and the data when the sign of breakout does not occur A process of collecting the data of the temperature distribution pattern and storing it in the separately prepared learning signal storage device 41 in association with the presence / absence information of the occurrence of the sign of breakout is performed. Preferably, this collection data is collected as much as possible in accordance with breakouts that occur in various ways under various operating conditions.

Subsequently, the operator specifies that the temperature distribution pattern information stored in the learning signal storage device 41 is used as an input signal for learning, and that information on whether or not a sign of breakout has occurred at that time is used as a teacher signal for learning. Learning processing device
Start 40. Learning processing device started in this way
40 presents the temperature distribution pattern information stored in the learning signal storage device 41 to the network structure unit 30 as an input signal, and responds to this presentation to the network structure unit 30.
The learning process of the internal state value of the network structure unit 30 is performed so that the output signal from the network structure unit 30 becomes the presence / absence information of the occurrence of the corresponding breakout sign stored in the learning signal storage unit 41. This learning process is executed by, for example, the above-described back propagation method. Then, the operator performs a process of storing the internal state value obtained by the learning process of the learning processing device 40 in the internal state value management unit 31.

In this way, the network structure unit 30 to which the internal state value learned by the learning processing device 40 is assigned, receives the temperature distribution pattern information detected by the temperature detection end 20 during the actual operation, and Since the processing for adaptively outputting the presence / absence information of the occurrence of the breakout sign corresponding to the pattern information is performed, the occurrence of the breakout can be predicted according to the output signal of the network structure unit 30.

Further, when a sign of a breakout occurs, as described in the section of [Prior Art], an operation is performed in which the surface temperature of the mold at the point where the breakout occurs sharply rises and then the surface temperature falls. That is, a time-series pattern of characteristic temperatures is generated on the side wall of the mold 10 as shown in FIG. Then, corresponding to this, the first derivative of the surface temperature and 2
Differential information such as next derivative and deviation information (deviation from DC component) also rises sharply in response to the occurrence of a breakout sign,
Thereafter, a descending operation is performed. On the contrary,
When no signs of breakout occur, the time-series pattern of the temperature of the side wall of the mold 10 and its differential information shows a flat characteristic with relatively little variation, although there is a difference depending on the operating conditions. . Conventionally, the time series pattern of the temperature and its differential information is not captured as a pattern, but is captured only by a level difference. On the other hand, the present invention is configured to detect the occurrence of the sign of breakout by utilizing the feature as the pattern.

According to the present invention, the operator planning the breakout prediction process firstly arranges one or more temperature sensing ends 20 for measuring the temperature of the side wall of the mold 10, for example, in the lateral direction of the side wall of the mold 10. Set up. Then, when trying to detect the occurrence of a sign of a breakout with a time-series pattern of temperature, a time-series data storage unit 33 that holds time-series data of the detected temperature value of the disposed temperature detection end 20 is prepared. Also, when trying to detect the occurrence of a sign of breakout with a time-series pattern of temperature differential information, a differential operation unit 32 that calculates differential information of the detected temperature value of the disposed temperature detection end 20; A time-series data storage unit 33 for holding time-series data of differential information obtained by the differential operation unit 32 is prepared. In both cases, in addition to this, the network structure unit 30 using the temperature detection value stored in the time-series data storage unit 33 as an input signal is prepared in association with the time-series data storage unit 33.

Next, when trying to detect the occurrence of the sign of breakout from the operation data so far in a time-series pattern of the temperature, the detected temperature value of the temperature detecting end 20 at the time of the sign of the breakout occurs. The time-series pattern data and the time-series pattern data of the detected temperature value when no sign of breakout is generated are collected, and the learning signal storage device 41 provided is provided with the data of the sign of breakout. A process of storing the information in association with the presence / absence information is performed. Also, when trying to detect the occurrence of a sign of breakout using the time series pattern of the temperature differential information, the time series pattern of the differential information of the detected temperature value of the temperature detection end 20 when the sign of the breakout occurs is detected. Data and data of a time-series pattern of differential information of the detected temperature value when no sign of breakout occurs are collected and associated with the presence / absence information of the sign of breakout in the learning signal storage device 41. Perform the process of storing. This sampling data is also preferably collected as much as possible under various operating conditions to accommodate breakouts that occur in various ways.

Subsequently, when the operator intends to detect the occurrence of the sign of breakout with the temperature time-series pattern, the temperature time-series pattern information stored in the learning signal storage device 41 is used as an input signal for learning, and The learning processing device 40 is started by designating the use / non-occurrence information of the occurrence of the breakout symptom as a teacher signal for learning, and the occurrence of the breakout symptom is generated by a time-series pattern of the differential information of the temperature. When trying to detect, the time-series pattern information of the temperature differential information stored in the learning signal storage device 41 is used as an input signal for learning, and information on the presence / absence of a breakout sign at that time is used as a learning teacher signal. The learning processing device 40 is started by designating the use as a signal.
The learning processing device 40 started in this way performs the learning process of the internal state value of the network structure unit 30 to which the learning teacher signal is to be output according to the same process as the above-described process. Then, the operator performs a process of storing the internal state value obtained by the learning process of the learning processing device 40 in the internal state value management unit 31.

In this manner, the network structure unit 30 to which the internal state value learned by the learning processing device 40 is assigned is the time-series pattern information of the temperature detected by the temperature detecting end 20 during the actual operation (or the differential information of the temperature). When the time-series pattern information is input, the time-series pattern information of the temperature (or the time-series pattern information of the differential information of the temperature)
Is processed so as to adaptively output the presence / absence information of the occurrence of the breakout symptom corresponding to the above, so that the occurrence of the breakout can be predicted according to the output signal of the network structure unit 30. In this configuration, when a plurality of temperature detecting terminals 20 are provided, any one of the network structure units 30 outputs a breakout occurrence prediction signal by outputting breakout symptom occurrence information. It is possible to add various logical operation processes, such as adopting a configuration or adopting a configuration of outputting a prediction signal of the occurrence of a breakout by majority logic.

When it is necessary to further enhance the reliability of the breakout prediction processing based on the time-series pattern information of the temperature and the time-series pattern information of the differential information of the temperature,
A plurality of temperature detecting ends 20 are provided in the lateral direction of the side wall of the mold 10 (sufficient in the lateral direction is sufficient).
Is prepared. Then, it is detected whether or not the sign of the breakout occurs substantially simultaneously at the adjacent temperature detecting end 20 according to the prepared first arithmetic unit 34,
It is preferable to adopt a configuration in which the prediction display of the breakout occurrence is executed on condition that the simultaneous occurrence is detected. With such a configuration, even if a temperature condition similar to that at the time of the sign of breakout may occur due to a change in the molten metal level, a variation in powder inflow, an air gap between the solidified shell and the mold surface, and the like. Thus, it is possible to accurately predict the occurrence of a breakout symptom state without erroneous detection.

Further, when it is necessary to enhance the reliability of the breakout prediction process based on the temperature distribution pattern of the side wall of the mold 10, the time-series pattern information of the temperature, and the time-series pattern information of the differential information of the temperature, the operator operates the side wall of the mold 10 The temperature detection ends 20 are arranged in two rows in the flow direction of the molten steel (the flow direction is sufficient).
A network structure unit 30 and a first calculation unit 34 are prepared as a breakout prediction unit 35 in association with the temperature detection ends 20 located at the same portion or a portion in the vicinity of the column, and the breakout prediction unit 35 A second operation unit 36 having an output as an input is prepared. Then, it is detected whether or not the signs of the breakout occur substantially simultaneously at the temperature detection ends 20 of the same part in the upper and lower two rows in accordance with the prepared second calculation unit 36 to detect the simultaneous occurrence. It is preferable to adopt a configuration in which the prediction display of the occurrence of the breakout is executed under the condition of With this configuration, even when a temperature detection state similar to that at the time of the sign of a breakout may occur, the occurrence of the sign-out state of a breakout can be accurately predicted without erroneous detection.

The buffer unit 341 constituting the first arithmetic unit 34 converts the output signal of the network structure unit 30 in order to absorb a time lag of the timing of occurrence of a breakout sign appearing at the adjacent temperature detecting end 20. Processing is performed so as to hold for a certain period of time, and to output the sign information of the occurrence of a breakout within the period according to the stored data. Then, the hierarchical network structure unit 342 constituting the first arithmetic unit 34 stores the data of the hierarchical network structure provided by itself when all the buffer units 341 that provide the input signals output the symptom information of the occurrence of the breakout. By operating to output the simultaneous occurrence of the sign of breakout according to the processing function, the operation of realizing the function of the first arithmetic unit 34 is performed.

As described above, the present invention focuses on the point that the characteristics of the spatial and temporal temperature patterns of the side wall of the mold 10 are greatly different between when the sign of breakout occurs and when it does not occur, and the adaptive data Network structure part that executes processing
By storing the temperature patterns of the occurrence and non-occurrence of the breakout sign in the 30 data processing functions, it is possible to adaptively detect the occurrence of the breakout sign. It is assumed that. As a result, the ability to detect the occurrence of a breakout symptom state can be dramatically improved as compared with the conventional level judgment prediction processing.

Since the network structure unit 30 executes parallel distributed processing due to its structure, it becomes possible to detect the occurrence of a breakout symptom state faster than in the conventional sequential prediction algorithm. Moreover, the internal state value that defines the detection function of the network structure unit 30 is obtained mechanically by the learning processing device 40 by using data available as operation data as a learning signal without processing it, and However, even if another new learning signal is obtained, a higher-precision internal state value is obtained by using the previously obtained learning signal and the new learning signal as a new learning signal. Since it is required mechanically, the adjustment / change processing of the network structure unit 30 can be executed without using a trial and error method as in the related art, and thus the load on the operator can be significantly reduced.

〔Example〕

 Hereinafter, the present invention will be described in detail with reference to examples.

FIG. 3 illustrates a system configuration of a continuous casting system to which the present invention is applied. In the figure, the same components as those described in FIG. 1 are indicated by the same symbols. Reference numeral 11 denotes a roll, which pulls out the solidified cast iron on the surface from the mold 10, and 50 denotes a breakout prediction device that implements the present invention, which performs processing to predict the breakout of the slab in the mold 10. And 60, a roll control device for controlling the pulling speed of the roll 11, and a breakout prediction device 50.
Control to reduce the sign of the breakout caused by reducing the pulling speed of the roll 11 in response to the prediction signal from the control unit, and to determine the original speed of the roll 11 in response to the cancellation signal from the breakout prediction device 50. Is controlled to return to the high-speed one. As shown in this figure, in the system configuration of this embodiment, the mold
The temperature detecting ends 20 disposed on the side walls of the mold 10 are arranged in two rows in the lateral direction of the side wall of the mold 10.

FIG. 4 shows an embodiment of the breakout prediction device 50. In the figure, the same components as those described in FIG. 1 are indicated by the same symbols. Reference numeral 30a denotes a network structure provided in association with the temperature detecting end 20 in the upper row of the two-row configuration, 30b denotes a network structure provided in association with the temperature detecting end 20 in the lower row of the two-row configuration, and 37 denotes a hierarchical network. A prediction processing unit that receives the output value of the structure unit 342 as an input, specifies the maximum value of the output value of the input hierarchical network structure unit 342, and when the maximum value is larger than a predetermined determination value. A device for outputting a breakout prediction signal on the assumption that a sign of breakout has occurred, 38 is a prediction information display, and a prediction processing unit
Displaying foresight signals output by 37 to the operator, 39
Is a breakout occurrence confirming unit corresponding to the second calculating unit 36 in FIG. 1, which receives the output value of the network structure unit 30b and the output value of the prediction processing unit 37 as input and detects the temperature of the same upper and lower parts. It is determined whether or not the hierarchical network structure unit 342 and the network structure unit 30b provided in association with the end 20 detect the occurrence of a breakout sign simultaneously (simultaneously considering a certain delay time). At the same time, when the simultaneous occurrence of the signs of the breakout is not detected, the cancel signal is output, and when the simultaneous occurrence is detected, the cancel signal is not output.

The time-series data storage unit 33 provided before the network structure units 30a and 30b stores time-series data of the differential information of the temperature detection value output from the differential operation unit 32, as described in FIG. It is. In order to execute the holding process, the time-series data storage unit 33 includes, for example, ten buffer units, and holds the time-series data of the differential information of the temperature detection value obtained every three seconds, for example, while updating it. Process. The network structure units 30a and 30b are configured to use the time-series data of the differential information stored in the time-series data storage unit 33 as an input signal.

FIG. 5 shows an embodiment of the network structure units 30a and 30b. In this embodiment, a configuration in which the network structure units 30a and 30b have a hierarchical network configuration as described with reference to FIG. 18 is disclosed. That is, distribute the input signal
An input layer composed of ten input units 1 ', an intermediate layer composed of eight basic units 1 having a one-stage configuration,
An output layer composed of one basic unit 1 and an internal connection in which a weight value is set between the input layer and the intermediate layer, and between the intermediate layer and the output layer.
The network structure units 30a and 30b are configured. If the configuration of the breakout prediction device 50 is allowed to be complicated, it is also possible to adopt different hierarchical network configurations for the network structure unit 30a and the network structure unit 30b. It is also possible to adopt different hierarchical network configurations between the network structure units 30a and between the network structure units 30b.

The hierarchical network structure of the embodiment shown in FIG.
342, as shown in the figure, the network structure unit 30 provided in association with the two adjacent temperature detection ends 20 in the upper row.
A configuration is adopted in which the output value of a is input (input via the buffer unit 341). FIG. 6 shows an embodiment of the hierarchical network structure unit 342. In this embodiment, an input layer composed of two input units 1 'for distributing an input signal, an intermediate layer composed of four basic units 1 having a one-stage configuration, and one basic unit 1 And a hierarchical network structure unit 342 in which an internal connection having a weight value set between the input layer and the intermediate layer and between the intermediate layer and the output layer is configured. .

Network structure 30a, b and hierarchical network structure 3
As a configuration method using 42 hardware parts, one of the present applicants has filed “Japanese Patent Application No. 63-216865 (August 1988)
Filed on March 31, "Network configuration data processing device")
Can be used.

That is, the basic unit 1 is, as shown in FIG.
Multiplication type for multiplying the output from the preceding layer input through the input switch unit 7 and the weight value held by the weight value holding unit 8
A D / A converter 2a, an analog adder 3a that adds the output value of the multiplying D / A converter 2a and the previous accumulated value to calculate a new accumulated value, and an addition result of the analog adder 3a A sample-and-hold circuit 3b for holding, a non-linear function generation circuit 4a for non-linearly converting data held in the sample-and-hold circuit 3b when the accumulation processing is completed, and a non-linear function generation circuit 4a for outputting to a subsequent layer. This is realized by including an output holding unit 5 that holds an analog signal value, an output switch unit 6 that outputs data held by the output holding unit 5, and a control circuit 9 that controls these processing units.

The hierarchical network configuration is realized by a configuration in which the basic units 1 having this configuration are electrically connected by one common analog bus 70 as shown in FIG. In the figure, reference numeral 71 denotes a weight output circuit for giving a weight value to the weight holding unit 8 of the basic unit 1, 72 denotes an initial signal output circuit corresponding to the input unit 1 ', and 73 denotes a control signal for data transfer. A synchronous control signal line 74 for transmitting a signal to the weight output circuit 71, the initial signal output circuit 72, and the control circuit 9;
Is a main control circuit for transmitting a synchronization control signal.

In the hierarchical network configuration having this configuration, the main control circuit 74 sequentially selects the basic units 1 in the preceding layer in a time-series manner, and synchronizes the selection processing with the output holding unit 5 of the selected basic unit 1. The held final output is processed so as to be output to the multiplying D / A converter 2a of the basic unit 1 in the subsequent stage via the analog bus 70 in accordance with a time-division transmission format. Upon receiving this input, the multiplying D / A converter 2a of the subsequent-stage basic unit 1 sequentially selects corresponding weight values, performs multiplication processing of the input values and the weight values, and performs analog-adder 3a and sample-and-hold operations. The accumulation processing unit 3 constituted by the circuit 3b sequentially accumulates the multiplied values. Subsequently, when all the accumulating processes for the basic unit 1 of the preceding layer are completed, the main control circuit 74 activates the nonlinear function generating circuit 4a of the basic unit 1 of the subsequent layer to calculate the final output, and The output holding unit 5 performs processing to hold the final output of this conversion processing result. And the main control circuit
Reference numeral 74 designates a process in which an output pattern corresponding to the input pattern is output by repeating the same processing for the next subsequent layer using the subsequent layer as a new preceding layer.

The weight value of the internal connection between the network structure units 30a and 30b and the hierarchical network structure unit 342 is obtained by, for example, the learning processing device 40 of FIG. 1 that implements the back propagation method. FIG. 9 shows an embodiment of a system configuration for realizing this learning process. In the figure, the same components as those described in FIG. 1 are indicated by the same symbols. Reference numeral 300 denotes a network simulator, which is built on a computer and simulates the data conversion function of the network structure units 30a and 30b and the hierarchical network structure unit 342.
Reference numeral 2 denotes a learning signal presenting unit, which reads out a learning signal for learning from the learning signal storage device 41, presents the learning presenting signal to the network simulator 300, and generates another learning teacher signal forming a pair. Are presented to a learning execution unit 44 to be described later and a learning convergence determination unit 43 described below. Reference numeral 43 denotes a learning convergence determination unit, which is an output signal output from the network simulator 300 and a learning signal presentation unit 4.
In response to the learning teacher signal from step 2, it is determined whether or not an error in the data processing function of the network simulator 300 is within an allowable range, and the determination result is notified to the learning signal presenting unit 42. An execution unit, a learning teacher signal from the learning signal presentation unit 42 and a network simulator
Upon receiving the 300 network output data, the amount of weight value update is calculated according to the back propagation method, and the weight value is updated so that the weight value is learned to converge to a convergence value. Note that it is of course possible to prepare and use the network structure units 30a and 30b and the hierarchical network structure unit 342 themselves without using the network simulator 300.

Next, the processing of the present invention thus configured will be described in detail with reference to the processing procedure shown in FIG.

Operators planning breakout predictions
First, as shown in step 1 of the processing procedure in FIG. 10, the time series data of the temperature detection value of the temperature detection end 20 when the sign of the breakout occurs, and the breakout A process of collecting time-series data of the temperature detection value when no sign of the occurrence has occurred is executed. FIG. 11 shows an example of the time series data of the collected temperature detection values. Here, FIG. 11 (a)
However, when a sign of a breakout occurs, two adjacent temperature sensing terminals 20 arranged at a substantially middle position of the upper row.
FIG. 11B shows an example of time-series data of temperature detection values detected by the two temperature detection terminals 20 when no sign of breakout occurs. 4 shows time-series data of detected temperature values.
Such collected data is stored in the upper and lower rows of temperature detection terminals.
It is preferable to collect in accordance with each of the 20
It is preferable to collect as much as possible in response to breakouts that occur in various ways under various operating conditions.

Next, as shown in step 2, the operator calculates differential information of the detected temperature value collected in step 1 and uses the time-series data of the calculated differential information of the detected temperature value to generate a sign of breakout. Is stored in the learning signal storage device 41 in association with the presence / absence information. No.
FIG. 12 shows an example of the time series data of the differential information of the calculated temperature detection value. Here, FIG.
FIG. 11 (a) shows differential information of the time series data (this figure is primary differential information), and FIG. 12 (b) shows differential information of the time series data of FIG. 11 (b). (This diagram is first-order differential information). The time series data of the differential information of the detected temperature value may be directly collected from the differential operation unit 32.

Subsequently, as shown in step 3, the operator sends the learning processing device 40 the time-series data of the differential information of the temperature detection value stored in the learning signal storage device 41 as the learning presentation signal, and sets the breakpoint at that time. By specifying that the information indicating the presence / absence of an out sign is used as a learning teacher signal, an instruction is issued to start learning the weight value of the internal connection of the network structure units 30a and 30b. When started in this way, the learning signal presenting unit 42 reads out the time series data of the differential information of the temperature detection value from the learning signal storage device 41 and presents the time series data to the network simulator 300. (In this embodiment, "1" is present when there is information, and "0" is present when there is no information), and the learning execution unit 44 and the learning convergence unit 43 are executed.
To present. Then, the learning execution unit 44 outputs the output signal from the network simulator 300 output in response to the presentation of the time-series data of the differential information of the temperature detection value to the presence / absence information of the presented breakout sign. The weight value is updated as much as possible. This updating process is continued until the learning convergence determining unit 43 detects that the output signal from the network simulator 300 substantially matches the information indicating the presence / absence of the corresponding breakout sign.
When the learning signals are collected in correspondence with the temperature detection terminals 20 in the upper row and the lower row, the learning process is executed separately.

When the weight value of the internal connection between the network structures 30a and 30b is obtained by the learning processing device 40, the operator subsequently sends the learning processing device 40 Instructs to start learning the connection weight value. The hierarchical network structure section 342 executes a function of determining whether or not the network structure sections 30a provided corresponding to the two adjacent temperature detection ends 20 in the upper row have simultaneously detected the occurrence of a breakout sign. As shown in FIG. 13, in this learning, data of an area where the two input signal values both take values close to “1” (the BO area in the figure) is defined as a learning presentation signal. In this case, “1” indicating that a sign of breakout has occurred is used as a learning teacher signal, and one of the two input signal values takes a value close to “0” (see FIG. In the case where the data of the middle non-BO area) is used as the learning presentation signal, it is executed by using “0” indicating that no sign of breakout occurs.

When the weight value of the internal connection between the network structure units 30a and 30b and the hierarchical network structure unit 342 is determined by the learning processing device 40, the operator then converts the determined weight value into the network structure as shown in step 5. Part 30a,
The processing to be stored in the internal state value management unit 31 of b and the internal state value management unit 343 of the hierarchical network structure unit 342 is executed. Then, as shown in the next step 6, by activating the breakout prediction device 50, it is instructed to start execution of breakout prediction processing in the continuous casting system.

Next, returning to FIG. 4, the operation processing of the breakout prediction device 50 in which the learned weight value is set will be described in detail.

A temperature detection value detected according to a predetermined sampling period by the temperature detection end 20 disposed on the mold 10 is subjected to a differentiation process (differential processing for obtaining the same differential information as the differential information used as the learning presentation signal) by the differential operation unit 32. The processing is performed so as to be sequentially stored in the time-series data storage unit 33. The time-series data storage unit 33, as described above,
For example, it is provided with ten buffer means, and operates to hold time-series data of differential information of temperature detection values by holding differential information of temperature detection values of past ten points obtained every three seconds, for example. .

The network structure units 30a and 30b that receive the time-series data stored in the time-series data storage unit 33 perform a data conversion function using the learned weight values, thereby achieving a breakout of the time-series data. Is processed to output a display value of the occurrence state of the symptom. As described above, learning is performed such that the weight value of the network structure units 30a and 30b outputs “1” when a sign of breakout occurs, and outputs “0” when no sign of breakout occurs. Network structure part 30
“a” and “b” output a value close to “1” when the time-series data stored in the time-series data storage unit 33 indicates the occurrence state of the sign of breakout, and conversely, the non-occurrence state of the sign of breakout Is displayed so that a value close to "0" is output.

The output of the network structure unit 30a provided in association with the temperature detection end 20 in the upper row is sent to the buffer unit 341 provided for adjusting the deviation of the breakout detection timing with the temperature detection end 20 adjacent in the horizontal direction. Is entered. The buffer unit 341 has, for example, five buffer means to hold the output values of the past five points of the network structure unit 30a and to store the held data indicating the value closest to “1” among the held data. Processing to output. That is, as shown in FIG. 16, since the hot water wrinkles indicating the sign state of the occurrence of the breakout operate so as to propagate below the mold 10 with a substantially inverted triangular shape, Even at the adjacent temperature detecting end 20, the sign state of the occurrence of the breakout cannot be detected at the same time, so that the buffer unit 341 for absorbing the deviation of the detection timing is provided.

The hierarchical network structure unit 342 that receives the output value of the buffer unit 341 performs a data conversion function using the learned weight value, so that the two network structure units 30a that provide input signals simultaneously break out. Processing is performed to output a display value indicating whether or not the state of occurrence of the sign is being output. As described above, the weight value of the hierarchical network structure unit 342 outputs “1” when the symptom of the breakout occurs at the same time, and outputs “0” when the symptom of the breakout does not occur at the same time. Because it has been learned, the hierarchical network structure
342 indicates “1” when the two buffer units 341 that provide input signals together indicate the state of occurrence of the sign of breakout.
Is output, and conversely, when one of them indicates the non-occurrence state of the sign of breakout, processing is performed so as to output a value close to "0".

In response to the output of the hierarchical network structure unit 342, the prediction processing unit 37
Is specified, and if the maximum value is, for example, "0.5" or more, a breakout prediction signal is output assuming that a sign of breakout has occurred. In accordance with this prediction signal, the prediction information display 38 displays breakout prediction information to the operator of the continuous casting system, and instructs the roll control device 60 to reduce the pulling speed of the roll 11.

On the other hand, when the prediction processing unit 37 outputs a breakout prediction signal, the breakout occurrence confirmation unit 39 starts operating, and the hierarchical network structure unit 342 that has provided the prediction signal.
And a network structure section provided in association with the temperature detecting end 20 in the lower row at a position below the specified temperature detecting end 20.
The output value of the network structure unit 30b is monitored a predetermined number of times, and it is checked whether any output value of the network structure unit 30b shows a value larger than the predetermined value during the monitoring period. According to this check processing, if there is no value indicating a value larger than the predetermined value, the breakout symptom information detected in the upper row is determined to be erroneous detection, and is output to the roll control device 60 as a cancel signal. Conversely, when there is a value indicating a value larger than the predetermined value, it is determined that the breakout symptom information detected in the upper row is correct, and processing is performed so as not to output a cancel signal.

Thus, in the present invention, the network structure 30
In accordance with the adaptive data processing function of the device, the occurrence of a breakout symptom state is accurately predicted in real time.

FIG. 14 illustrates experimental data obtained by a simulation performed to verify the effectiveness of the present invention.
In this simulation, nine detected temperature values where a breakout occurred and no breakout occurred with respect to the breakout prediction device 50 built on the computer.
This is to simulate what kind of detection the breakout prediction device 50 according to the present invention has performed by inputting 25 temperature detection values.

This simulation is provided with a hierarchical network structure unit 342, assuming that there is a primary differentiation, a secondary differentiation, and a deviation (a deviation from the average value of the past eight points) as the differentiation processing executed by the differential operation unit 32. The test was performed assuming a case (experimental data in the figure) and a case without the provision (experimental data in the figure). Note that the conventional breakout prediction device has erroneously detected a breakout for all 25 detected temperature values for which no breakout has occurred.

As can be seen from the experimental data, it was confirmed that the use of the present invention can significantly reduce the number of times of erroneous detection as a breakout. This effect can be greatly improved by providing the hierarchical network structure section 342 (a method in which the detection value of the adjacent temperature detection end 20 is considered), and the differential information of the temperature detection value input to the network structure section 30 is obtained. It has been confirmed that the use of the first-order differential information greatly improves the above.

The present invention is characterized in that the occurrence of a breakout sign condition is detected by monitoring the state of the temperature pattern on the side wall of the mold 10 according to a data processing function employing a network configuration. As described above, since the temperature pattern state of the side wall of the mold 10 is characterized by a pattern, the embodiment only discloses the pattern grasp based on the time-series data of the differential information of the temperature detection value. However, it is also possible to configure the present invention by grasping a pattern based on time-series data of the temperature detection value itself without the differential operation unit 32, and as shown in FIG. Using the characteristics of the spatial temperature pattern at the time of occurrence, the temperature detection end 20 that detects this spatial temperature pattern is input to the network structure parts 30a and 30b by the temperature detection value. It is also possible to configure the present invention.

In this embodiment, in order to enhance the detection accuracy of the breakout, first, the occurrence of the sign of the breakout is detected with the temperature detection end 20 disposed in the lateral direction, and then, in the flow direction. Although the disclosure of checking the occurrence of a sign of breakout with the disposed temperature detecting end 20 has been disclosed, on the contrary, first, the temperature detecting end 20 disposed in the flow direction is checked. It is also possible to adopt a configuration in which the occurrence of a sign condition of a breakout is detected, and then the occurrence of the sign condition of a breakout is confirmed by the temperature detection end 20 disposed in the lateral direction. In other words, the present invention is not limited by the method of arranging the temperature detection ends 20 or the order of detection.

Although the illustrated embodiment has been described, the present invention is not limited to this. For example, the network structure 30a,
Although a hierarchical network configuration is disclosed as b, the present invention is not limited to this, and may adopt any other network configuration. In addition, as a learning algorithm executed by the learning processing device 40,
Although the method based on the propagation method is disclosed, the present invention is not limited to this, and may be based on any other learning algorithm.

〔The invention's effect〕

As described above, according to the present invention, according to the breakout prediction device configured by the data processing function of the network configuration that performs the adaptive data processing, the detection of the occurrence of the sign of the breakout is captured in a pattern. Since it is configured to execute adaptively, its detection ability can be dramatically improved compared to the conventional level judgment prediction processing. With the parallel distributed processing configuration, it is possible to detect the occurrence of a breakout symptom state faster than before. Moreover, since the operator can mechanically construct the breakout prediction device according to the present invention simply by obtaining the temperature pattern information obtained from the operation data, the operator is not required to perform the conventional trial-and-error adjustment load. No more compelling. And
If new temperature pattern information can be obtained, the detection capability of the breakout prediction device once constructed can be easily improved.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating the principle of the present invention, FIG. 2 is a diagram illustrating the surface temperature of a mold when a sign of breakout occurs, FIG. 3 is a diagram illustrating a continuous casting system to which the present invention is applied, FIG. FIG. 5 shows one embodiment of a breakout prediction device, FIG. 5 shows one embodiment of a network structure part, FIG. 6 shows one embodiment of a hierarchical network structure part, FIG. 7 shows one embodiment of a basic unit, FIG. Is an embodiment of a hierarchical network, FIG. 9 is a system configuration diagram for realizing a weight value learning process, FIG. 10 is an explanatory diagram of a processing procedure of the present invention, and FIG. 11 is a diagram of a sampled temperature detection value. An example of time series data, FIG. 12 is an example of time series data of differential information of a sampled temperature detection value, FIG. 13 is an explanatory diagram of a learning signal used for learning a weight value of an internal connection of a hierarchical network structure, Fig. 14 is an explanatory diagram of the experimental data, Figs. 15 and 16 Is an explanatory diagram of a restrictive breakout, FIG. 17 is a basic configuration diagram of a basic unit, and FIG. 18 is a basic configuration diagram of a hierarchical network. In the figure, 1 'is an input unit, 1 is a basic unit, 10 is a mold, 20 is a temperature detecting end, 30 is a network structure unit, 32 is a differential operation unit, 33 is a time series data storage unit, and 34 is a first operation unit. Section, 35 is a breakout prediction section, 36 is a second calculation section, 40
Is a learning processing device, and 41 is a learning signal storage device.

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Hiroyuki Tsuzuki 1015 Uedanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture Inside Fujitsu Limited (72) Inventor Takashi Kawasaki 1015 Uedanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture Fujitsu Limited ( 72) Inventor Shuji Naito 1-1-1, Edamitsu, Yawatahigashi-ku, Kitakyushu-city, Fukuoka Prefecture Inside Nippon Steel Corporation Yawata Works (72) Inventor Shinichi Fukunaga 1-1-1, Edamitsu, Yawata-higashi-ku, Kitakyushu-shi, Fukuoka Prefecture No. 1 Inside Nippon Steel Corporation Yawata Works (72) Inventor Hidetaka Konan 1-1 1-1 Edamitsu, Yawata Higashi-ku, Kitakyushu-shi, Fukuoka Prefecture Inside 1-1 Nippon Steel Corporation Yawata Works (72) Inventor Toru Takagi Nippon Steel Corporation Yawata Works, 1-1 1-1 Emitsu, Yahata-Higashi-ku, Kitakyushu-shi

Claims (6)

(57) [Claims] In a continuous casting breakout prediction processing method for predicting the occurrence of breakout of a slab in a mold in continuous casting, an input signal corresponding to an input signal is handled according to a data conversion function of a network structure to which internal state values are allocated. A network structure unit (30) for calculating and outputting an output signal is prepared, and the temperature detection values of a plurality of temperature detection ends (20) for detecting spatial temperature distribution information on the side wall of the mold are stored in the network structure unit. (30), and temperature distribution information detected by the temperature detection end (20) is used as the internal state value in the network structure unit (30).
When input to the network structure, the information on the occurrence of the sign of breakout indicated by the temperature distribution information is transmitted to the network structure section (3).
A breakout prediction processing method for continuous casting, characterized in that a value to be output from 0) is configured to be set. 2. A method for predicting the occurrence of breakout of a slab in a casting mold in continuous casting, wherein one or a plurality of temperature detection terminals (20) for detecting temperature information of a side wall of the casting mold are provided. A) a network structure section (30) for calculating and outputting an output signal corresponding to an input signal in accordance with a data conversion function of a network structure to which an internal state value is allocated, and A time-series data of a temperature detection value detected by the detection end (20) is input to the network structure section (30) prepared in association with the temperature detection end (20); When the time series data of the temperature detection value detected by the temperature detection end (20) is input to the network structure section (30) as a value, the breaker indicated by the time series data That generating information bets symptoms are configured such that a value to be output from the network structure (30) is set, breakout prediction processing method of continuous casting, characterized. 3. A continuous casting breakout predicting method for predicting the occurrence of breakout of a slab in a mold in continuous casting, wherein one or more temperature detecting ends (20) for detecting temperature information of a side wall of the mold are provided. A) a network structure section (30) for calculating and outputting an output signal corresponding to an input signal in accordance with a data conversion function of a network structure to which an internal state value is allocated, and A time-series data of differential information of a temperature detection value detected by the detection end (20) is input to the network structure section (30) prepared in association with the temperature detection end (20); When the time series data of the differential information of the temperature detection value detected by the temperature detection end (20) is input to the network structure unit (30) as the internal state value, the time series data That generating information signs of breakout pointed to data is configured such that a value to be output from the network structure (30) is set, breakout prediction processing method of continuous casting, characterized. 4. A method for predicting breakout of continuous casting according to claim 2 or 3, wherein the output signal value output from the network structure section (30) is
A first calculation unit (34) for determining whether a plurality of network structure units (30) provided in association with the adjacent temperature detection terminals (20) have detected the occurrence of the same breakout sign. A method for predicting the breakout of continuous casting. 5. A method for predicting a breakout in continuous casting according to claim 1, wherein the temperature detecting end is arranged in a side-by-side two-row configuration on a side wall of the mold and corresponds to a temperature detecting end for detecting temperature information of the side wall. In addition, a configuration having a network structure unit (30) is adopted, and the temperature detection terminals (20) of the corresponding same part in the two rows or a part in the vicinity thereof are obtained from output signal values output from the network structure part (30). ), A second computing unit (36) for determining whether or not the plurality of network structure units (30) detected the occurrence of the same breakout sign has been detected. Breakout prediction method for continuous casting. 6. The breakout prediction processing method for continuous casting according to claim 4, wherein the first arithmetic unit (34) is provided in association with the network structure unit (30). A) a buffer unit (341) for holding the time-series data of the output signal value of (3) and outputting the held data indicating the most broken-out state of the held data; and receiving the output signal of the buffer unit (341). And a hierarchical network structure section (342) for outputting a judgment value of the occurrence of a breakout symptom according to the same according to the data conversion function of the hierarchical network structure. Processing method.