JP7305378B2 - Machine learning method, heating furnace abnormality cause identification model, heating furnace abnormality cause identification method, and heating furnace abnormality cause identification device - Google Patents

Description

The present invention relates to a heating furnace abnormality cause identification method, a heating furnace abnormality cause identification device, a machine learning method, and a heating furnace abnormality cause identification model.

In general, when a heating furnace is operated, operation data such as gas and air usage amounts and temperatures are collected, and the conditions of the heating furnace are constantly monitored based on the collected operation data. When an abnormality occurs in the heating furnace, the collected operational data indicates an abnormal value. Against this background, Patent Literature 1 describes a method for determining the presence or absence of an abnormality in a heating furnace based on the deviation of the values of collected operation data from the values of normal operation data.

Japanese Unexamined Patent Application Publication No. 2011-12928

Among the abnormalities in the heating furnace, there are abnormalities such as oxygen concentration abnormalities that cannot be identified in detail from the value of the operation data even if the presence or absence of occurrence is known. Therefore, according to the method described in Patent Literature 1, even if the presence or absence of occurrence of an abnormality can be determined, it is necessary to investigate the cause of the abnormality on site for an abnormality whose cause cannot be specified. As a result, the work load increases due to on-site work, and if the investigation period is prolonged, the abnormal state will continue for a long time, and the heating furnace unit consumption will continue to deteriorate. In order to solve this problem, it is conceivable to use machine learning technology to learn the values of operation data during normal times and when abnormalities occur, thereby identifying the cause of the abnormality from the values of the operation data. . However, even if machine learning technology is used, it is not possible to identify the cause of anomalies that have not occurred in the past because there is no actual value of operation data, and therefore the causes of the anomalies cannot be learned. In addition, if it is not possible to diagnose equipment with only the operation data detection sensors that are currently used, it is necessary to add operation data detection sensors. It is difficult to assess the progress made.

The present invention has been made in view of the above problems, and its object is to be able to identify the cause of an abnormality, including the cause of an abnormality that has not occurred in the past, and to diagnose the situation of equipment when an operation data detection sensor is added. It is also an object of the present invention to provide a heating furnace abnormality cause identification method, a heating furnace abnormality cause identification device, a machine learning method, and a heating furnace abnormality cause identification model capable of evaluating equipment improvement conditions.

A heating furnace abnormality cause identification method according to the present invention is a heating furnace abnormality cause identification method for identifying the cause of an abnormality that has occurred in a heating furnace, comprising an acquisition step of acquiring operation data of the heating furnace; A model in which operation data is an input variable and the cause of an abnormality in a heating furnace is an output variable. For the cause, input the operation data acquired in the acquisition step to the abnormality cause identification model machine-learned using the simulated operation data created based on the actual operation data in normal times. includes two steps of an identification step of identifying the cause of an abnormality that occurred in the heating furnace, and the abnormality cause identification model uses operation data newly acquired by adopting an operation data detection sensor that has not been adopted. Machine learning is performed using data added to the simulated operation data.

A method for identifying the cause of abnormality in a heating furnace according to the present invention is characterized in that, in the above invention, the cause-of-abnormality identification model is machine-learned using deep learning.

An abnormality cause identification device for a heating furnace according to the present invention is an abnormality cause identification device for a heating furnace that identifies the cause of an abnormality that has occurred in a heating furnace. A model in which operation data is an input variable and the cause of an abnormality in a heating furnace is an output variable. Regarding the cause, by inputting the operation data acquired by the acquisition means to the abnormality cause identification model machine-learned using the simulated operation data created based on the actual operation data in normal times. , and identifying means for identifying the cause of an abnormality that occurred in the heating furnace, and the abnormality cause identification model uses the operation data newly acquired by adopting an operation data detection sensor that has not been adopted. It is characterized by being machine-learned using data added to simulated operation data.

The machine learning method according to the present invention includes a step of creating operation data of the heating furnace in the case where a cause of an abnormality that has not been recorded in the past occurs as simulated data based on the operation data of the heating furnace during normal operation; Using the operation data in normal time, the simulated data, and the operation data newly acquired by adopting the unadopted operation data detection sensor, the operation data of the heating furnace is used as an input variable, and the cause of the abnormality of the heating furnace. is characterized by including two steps of machine learning of an abnormality cause identification model having as an output variable.

In the machine learning method according to the present invention, in the step of creating the simulated data of the above invention, a model between operation data is created based on the material balance and / or heat balance in the heating furnace, and a model between operation data is created. and applying actual or simulated operational data to the model between operational data to create simulated data.

An abnormality cause identification model for a heating furnace according to the present invention is machine-learned by the machine-learning method according to the present invention.

According to the heating furnace abnormality cause identification method, the heating furnace abnormality cause identification device, the machine learning method, and the heating furnace abnormality cause identification model according to the present invention, the cause of abnormality including the cause of abnormality that has not occurred in the past can be identified. It is possible to identify and evaluate the facility diagnosis status and facility improvement status when the operation data detection sensor is added.

FIG. 1 is a block diagram showing the configuration of an abnormality cause identification system for a heating furnace, which is an embodiment of the present invention. FIG. 2 is a diagram showing an example of operational data and causes of abnormalities in the heating furnace. FIG. 3 is a schematic diagram showing the configuration of an operation data DB. FIG. 4 is a flow chart showing the flow of machine learning processing, which is an embodiment of the present invention. FIG. 5 is a diagram for explaining the relationship of operation data. FIG. 6 is a flow chart showing the flow of abnormality cause identification processing, which is an embodiment of the present invention.

Hereinafter, the configuration and operation of a system for identifying the cause of abnormality in a heating furnace, which is an embodiment of the present invention, will be described with reference to the drawings.

〔composition〕
First, with reference to FIGS. 1 to 3, the configuration of an abnormality cause identification system for a heating furnace, which is an embodiment of the present invention, will be described. FIG. 1 is a block diagram showing the configuration of an abnormality cause identification system for a heating furnace, which is an embodiment of the present invention. FIGS. 2(a) and 2(b) are diagrams showing an example of operation data and an abnormality cause of the heating furnace, respectively. FIG. 3 is a schematic diagram showing the configuration of an operation data DB.

As shown in FIG. 1, a heating furnace abnormality cause identification system 1 according to one embodiment of the present invention is a system for identifying the presence or absence of abnormality and the cause of the abnormality in a heating furnace 2 that heats steel materials. A data acquisition unit 3 , an abnormality cause identification device 4 , an output device 5 , an operation data database (operation data DB) 6 and a machine learning unit 7 are provided.

The operation data acquisition unit 3 acquires operation data (absolute value or relative value with respect to the absolute value at a predetermined time) of the heating furnace 2 from various sensors provided in the heating furnace 2 and an information processing device such as a process computer. Operation data is output to the abnormality cause identification device 4 . The operation data acquired by the operation data acquisition unit 3 include M gas (a mixed gas of by-product gases generated from a blast furnace, a coke oven, a converter, etc.) supplied to the heating furnace 2 and the flow rate of air, Oxygen concentration ( _O2 concentration) of the steel, time in the furnace, etc. can be exemplified.

The abnormality cause identification device 4 is configured by an information processing device. The cause-of-abnormality identification device 4 executes the cause-of-abnormality identification process described later using the cause-of-abnormality identification model 41 machine-learned by the machine-learning unit 7, based on the operation data output from the operation-data acquisition unit 3. The presence or absence of an abnormality in the heating furnace 2 and the cause of the abnormality are identified, and information on the identified cause of the abnormality is output to the output device 5 when an abnormality occurs.

In this embodiment, the abnormality cause identification model 41 includes an input layer 41a, an intermediate layer 41b, and an output layer 41c. The input layer 41a receives operation data acquired by the operation data acquiring unit 3, for example, as shown in FIG. 2(a). The parameters of the intermediate layer 41b are machine-learned using learning data including operation data, normality/abnormality determination results, and data indicating the cause of abnormality. The output layer 41c outputs the result of normal/abnormal determination for the operation data input to the input layer 41a, and data indicating the cause of the abnormality when the operation data is determined to be abnormal.

Here, when an O ₂ concentration abnormality occurs in the heating furnace 2, the cause of the abnormality is, for example, as shown in FIG. air flow meter abnormality, air flow meter abnormality in heating furnace 2, O ₂ concentration meter abnormality in heating furnace 2, exhaust gas switching valve leak, heating _furnace 2 Examples include an increase in the amount of air that enters the duct, damage to the burner tile, and the like.

The output device 5 is composed of a well-known output device such as a display device and a printing device. The output device 5 notifies the operator of information about an abnormality that has occurred in the heating furnace 2 and includes the cause of the abnormality identified by the cause-of-abnormality identification device 4 .

As shown in FIG. 3, the operation data DB 6 stores the operation data acquired by the operation data acquisition unit 3, the determination result of normality/abnormality of the operation data, and the data indicating the cause of the abnormality at the time of determination of abnormality in association with each other. are doing. Here, the operational data shown in FIG. 3 are data independent of each other, using instantaneous values of the operational data or average values within a certain period of time. A determination of normality or abnormality is given to each of these instantaneous values or average values within a certain period of time. It should be noted that determination of normality or abnormality of operation data acquired in advance is made based on a specific threshold value, disturbance of correlation, etc., based on past measurement results.

The machine learning unit 7 is configured by an information processing device, and performs a machine learning process, which will be described later, using a machine learning technique such as deep learning. The specific model 41 is machine-learned. In addition, in the present embodiment, the machine learning unit 7 uses operation data when an abnormality cause that has not occurred in the past and operation data related to an abnormality cause for which a predetermined number or more of operation data has not been obtained as simulated data. It has a simulated data generation unit 71 that generates from operating data of time.

[Machine learning processing]

Next, machine learning processing, which is an embodiment of the present invention, will be described with reference to FIGS. 4 and 5. FIG. FIG. 4 is a flow chart showing the flow of machine learning processing, which is an embodiment of the present invention. FIG. 5 is a diagram for explaining the relationship of operation data.

The flowchart shown in FIG. 4 starts at a predetermined timing such as when new operational data is added to the operational data DB 6, and the machine learning process proceeds to step S1.

In the processing of step S1, the machine learning unit 7 determines whether or not there is a cause of abnormality such as no operation data being obtained or no operation data equal to or exceeding a predetermined number being obtained. As a result of determination, if there is such an abnormality cause (step S1: Yes), the machine learning unit 7 advances the machine learning process to the process of step S2. On the other hand, if there is no such cause of abnormality (step S1: No), the machine learning unit 7 advances the machine learning process to step S3.

In the process of step S2, the simulated data generating unit 71 generates simulated operation data when an abnormality has occurred in which no operation data is obtained or a predetermined number or more of operation data are not obtained. Specifically, since the operational data affect each other, the simulated data generation unit 71 creates an inter-operation data model that shows the relationship between the operation data based on the material balance and/or heat balance in the heating furnace 2, and creates Simulated data is generated using the operational data inter-model. If it is difficult to calculate the material balance and/or the heat balance, simulation data may be created using a multivariable analysis technique such as multiple regression calculation.

More specifically, as shown in FIG. 5, there is a correlation between the flow rate of M gas and the flow rate of air in the heating furnace 2, and when the flow rates of M gas and air change, the flow rate in the heating furnace 2 _O2 concentration changes. Therefore, first, the simulated data generation unit 71 creates a model between operation data that indicates the relationship between such operation data. Next, when creating simulated data with an abnormality in the M gas flowmeter as the cause of the abnormality, the simulation data generation unit 71 changes the flow rate of the M gas by ±10% or more from the flow rate in the normal state. The flow rate of the gas is calculated, and the flow rate of the M gas at the time of abnormality is input to the operation data inter-model described above to calculate the abnormal values of the other operation data. By this processing, simulated data can be created in which the abnormality is caused by an abnormality in the M gas flow meter. Thereby, the processing of step S2 is completed, and the machine learning processing proceeds to the processing of step S3.

In the process of step S3, the machine learning unit 7 uses the operation data, the result of normal/abnormal determination of the operation data, and the data indicating the cause of the abnormality when determining the abnormality, which are stored in the abnormality cause identification device 4. The abnormality cause identification model 41 is machine-learned. Also, at this time, the machine learning unit 7 adds newly acquired operation data to the simulated data by employing an unadopted operation data detection sensor, and machine-learns the abnormality cause identification model 41 . Equipment diagnosis status and equipment improvement after adopting the unadopted operational data detection sensor by adding the newly acquired operational data to the existing simulated data by adopting the unadopted operational data detection sensor Able to assess the situation. Thereby, the processing of step S3 is completed, and a series of machine learning processing ends.

[Abnormal cause identification process]
Next, with reference to FIG. 6, an abnormality cause identification process, which is an embodiment of the present invention, will be described. FIG. 6 is a flow chart showing the flow of abnormality cause identification processing, which is an embodiment of the present invention.

The flowchart shown in FIG. 6 starts at the timing when the operation data is output from the operation data acquisition unit 3 to the abnormality cause identification device 4, and the abnormality cause identification process proceeds to the process of step S11.

In the process of step S<b>11 , the abnormality cause identification device 4 acquires the operational data acquired from the operational data acquisition section 3 . Thereby, the process of step S11 is completed, and the abnormality cause identification process proceeds to the process of step S12.

In the process of step S12, the abnormality cause identification device 4 inputs the operation data acquired in the process of step S11 to the abnormality cause identification model 41, thereby determining whether the operation data acquired in the process of step S11 is normal/abnormal. , data relating to the cause of the abnormality are acquired when the abnormality is determined. Thereby, the process of step S12 is completed, and the abnormality cause identification process proceeds to the process of step S13.

In the process of step S13, based on the information output from the cause-of-abnormality identification model 41 in the process of step S12, the abnormality cause identification device 4 determines whether the operation data obtained in the process of step S11 is normal or abnormal. Determine what it is. As a result of the determination, if the operation data is normal, the abnormality cause identification device 4 terminates the series of abnormality cause identification processing. On the other hand, if the operation data is abnormal operation data, the abnormality cause identification device 4 advances the abnormality cause identification process to the process of step S14.

In the process of step S<b>14 , the abnormality cause identification device 4 outputs to the output device 5 the abnormality determination result and the data regarding the abnormality cause. As a result, the processing of step S14 is completed, and the series of abnormality cause identification processing ends.

As is clear from the above description, in the abnormality cause identification system 1 for the heating furnace, which is one embodiment of the present invention, the abnormality cause identification device 4 uses the operation data of the heating furnace 2 as an input variable, In a model in which the cause of an abnormality is an output variable, there is no record of operation data corresponding to the cause of the abnormality, or no record of operation data exceeding a predetermined number has been obtained. By inputting the operation data acquired by the operation data acquisition unit 3 to the abnormality cause identification model 41 machine-learned using the simulated operation data created based on the operation data, generated in the heating furnace 2 The cause of the abnormality is identified, and the abnormality cause identification model 41 is machine-learned using data obtained by adding operation data newly acquired by adopting an unadopted operation data detection sensor to the simulated operation data. Therefore, it is possible to identify the cause of the abnormality including the cause of the abnormality that has not occurred in the past, and to evaluate the facility diagnosis status and the facility improvement status when the operation data detection sensor is added.

〔Example〕
In this example, the cause of the abnormal O ₂ concentration in the heating furnace was identified. Causes of _O2 concentration abnormalities in the heating furnace include M gas flow meter abnormalities, air flow meter abnormalities, _O2 concentration meter abnormalities, increased intruding air, switching valve abnormalities, and burner tile damage. , and the M gas flowmeter, no cause of abnormality has occurred in the past, and there is no actual operational data, so simulated data was generated. As for the M gas flow meter anomaly, there is actual operational data, but in order to confirm the determination accuracy of the simulated data, simulated data was generated and machine learning was performed. For machine learning, we used simulated data for each cause of anomalies and operational data during normal times. Judgment was made on this learned content using actual operation data when an abnormality occurred in the M gas flow meter. As a result, the determination accuracy was 100%, and it was confirmed that the cause of abnormality can be determined with high accuracy even using simulated data. In addition, by installing non-adopted thermometers and pressure gauges on the exhaust gas inlet and outlet sides of the regenerative heat storage body, new data on the temperature and pressure of the exhaust gas on the inlet and outlet sides can be obtained. , it is possible to evaluate the details of the switching valve abnormality of the exhaust gas, and as a facility improvement status, by adding the damage situation to the heat balance and mass balance of the regeneration heat storage body and calculating it, the regeneration heat storage body at each operation It is assumed that it will be possible to calculate the optimum temperature and manage it with high accuracy.

Although the embodiments to which the inventions made by the present inventors are applied have been described above, the present invention is not limited by the descriptions and drawings forming part of the disclosure of the present invention according to the embodiments. That is, other embodiments, examples, operation techniques, etc. made by those skilled in the art based on this embodiment are all included in the scope of the present invention.

1 heating furnace abnormality cause identification system 2 heating furnace 3 operation data acquisition unit 4 abnormality cause identification device 5 output device 6 operation data database (operation data DB)
7 machine learning unit 41 abnormality cause identification model 71 simulated data generation unit

Claims (5)

Operation data that simulates the operation data of the heating furnace when the cause of the abnormal oxygen concentration has occurred. a step that creates as
operation data of the heating furnace during normal operation, the simulated operation data, and operation data newly acquired after adoption by adopting an unadopted operation data detection sensor; Machine learning of an abnormality cause identification model using data indicating the cause of abnormality at the time of abnormality determination, with operation data of the heating furnace as an input variable and the cause of the oxygen concentration abnormality in the heating furnace as an output variable,
In the step of creating the simulated operational data, an inter-operational data model showing the relationship between the operational data is created based on the material balance and/or heat balance in the heating furnace, and the operation in which the cause of the abnormal oxygen concentration occurs. creating the simulated operational data by inputting data into the operational data inter-model;
Causes of oxygen concentration abnormality in the heating furnace include mixed gas flow meter abnormality, air flow meter abnormality, oxygen concentration meter abnormality, infiltration air increase, switching valve abnormality, and burner tile breakage.
A machine learning method characterized by: 2. A heating furnace abnormality cause identification model machine-learned by the machine-learning method according to claim 1. 3. The heating furnace abnormality cause identification model according to claim 2, which is machine-learned using deep learning. A heating furnace abnormality cause identification method for identifying the cause of an oxygen concentration abnormality occurring in a heating furnace,
an acquisition step of acquiring operating data of the heating furnace;
an identification step of identifying the cause of the oxygen concentration abnormality occurring in the heating furnace by inputting the operation data acquired in the acquisition step into the abnormality cause identification model according to claim 2 or 3;
A method for identifying the cause of an abnormality in a heating furnace, comprising: An abnormality cause identification device for a heating furnace that identifies the cause of an oxygen concentration abnormality that has occurred in a heating furnace,
Acquisition means for acquiring operation data of the heating furnace;
identifying means for identifying the cause of the oxygen concentration abnormality occurring in the heating furnace by inputting the operation data acquired by the acquiring means into the abnormality cause identifying model according to claim 2 or 3;
An abnormality cause identification device for a heating furnace, comprising:

Images