JP2010172962A - Method of predicting characteristics of rolled product - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method of predicting the characteristics of a rolled product, by which the prediction is carried out with high accuracy without especially bringing about the increase of calculation load. <P>SOLUTION: The method of predicting the characteristics of the rolled product includes: a step for obtaining data for learning; a step for calculating a learning term; and a predicting step. The step for obtaining the data for learning has a step for collecting rolling data, a step for calculating metallic structure, a step for calculating mechanical property and a step for measuring the mechanical property. The learning term calculating step is for obtaining the learning term by which the relationship between a calculated value for learning and a measured value for learning is shown. The predicting step has the step for collecting the rolling data, the step for calculating the metallic structure, the step for calculating the mechanical property and a step for correcting the calculated value for prediction. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a method for predicting properties of a rolled product, and more particularly to a method for predicting properties of a rolled product, which can be predicted with high accuracy without causing an increase in calculation load.

  In recent years, the specifications required by customers for rolled products have become stricter. In particular, it is important to keep properties such as strength and ductility of the rolled product within an allowable range. For this reason, conventionally, characteristic control has been attempted by predicting characteristics of a rolled product using a characteristic prediction model using manufacturing conditions such as heating, processing, and cooling in a rolling process as inputs.

  For example, Patent Document 1 discloses a method for controlling manufacturing conditions in order to predict characteristics for each of heating, rough rolling, finish rolling, and cooling processes in a rolling process, and to obtain target rolled product characteristics. ing. In this method, in order to improve the accuracy of prediction, first, the manufacturing conditions are obtained from the component values and the required characteristics. Then, after the heating process, the actual manufacturing conditions such as the heating time are taken in and recalculated, and the process after the heating is reviewed. After the rough rolling step, the manufacturing conditions of the rough rolling result are taken in and recalculated, and the subsequent steps are similarly reviewed. After the finish rolling process, the manufacturing conditions of finish rolling results are taken in and recalculated, and the subsequent processes are reviewed in the same manner.

  Patent Document 2 discloses a method for obtaining manufacturing conditions that satisfy the required specifications of a rolled product based on a database that stores the range of manufacturing conditions and the actual values of the characteristics of the rolled product in the range of the manufacturing conditions. Has been.

  Furthermore, Patent Document 3 discloses a method for predicting characteristics of a rolled product using a characteristic prediction model and obtaining manufacturing conditions that satisfy the required specifications when there is no rolled product having characteristics that satisfy the required specifications. It is shown.

  However, in the prediction by the characteristic prediction model, there is a limit to the prediction accuracy, and it is not always possible to obtain a highly accurate prediction result for any steel type and production condition. For this reason, a mechanism for improving the accuracy of the characteristic prediction model is required.

  As a method for realizing this mechanism, a method has been known in which parameter values of various manufacturing conditions are input and an output is obtained by a neural network regardless of a physical phenomenon that determines characteristics. However, this method has a problem that the process until the output is obtained is unclear, and when the parameter range is extrapolated, the accuracy is significantly deteriorated. And in order to solve this problem, the method of using the local neighborhood regression model currently disclosed by patent document 4 was known.

Japanese Patent No. 2509481 Japanese Patent No. 3053251 Japanese Patent No. 3053252 JP 2006-309709 A

  However, this method using a local neighborhood regression model is considered to increase the calculation load because a local neighborhood regression model is constructed using a large amount of data when predicting the characteristics of a rolled product.

  The present invention has been made to solve the above-described problems, and an object thereof is to provide a rolled product property prediction method capable of predicting the properties of a rolled product with high accuracy without causing an increase in calculation load. To do.

  A rolled product property prediction method according to a first aspect of the present invention includes a learning data acquisition step, a learning term calculation step, and a prediction step, wherein the learning data acquisition step selects a rolled product to be learned in a rolling line. Applying a step of collecting rolling data for manufacturing as rolling data for learning and a metallurgical phenomenon model obtained by formulating the metallurgical phenomenon, and learning the metal structure of the rolled product to be learned based on the rolling data for learning Applying the calculated mechanical property model to calculate the mechanical properties, and calculating the mechanical properties, and based on the learning rolling data and the learned metallic structure calculated values, A step of calculating the mechanical properties of the rolled product as a calculated value for learning, and after manufacturing the rolled product to be learned, the machine of the rolled product to be learned Measuring a property as an actual measurement value for learning, and the learning term calculation step acquires a learning term representing a relationship between the calculated calculation value for learning and the actual measurement value for learning, and the prediction A step of collecting rolling data when producing a rolling product to be predicted in the rolling line as rolling data for prediction; and applying the metallurgical phenomenon model, and applying the prediction rolling data based on the rolling data for prediction A step of calculating a metal structure of a rolled product as a predicted metal structure calculated value; and applying the mechanical property model, based on the predicted rolling data and the predicted metal structure calculated value, Calculating a physical property as a calculated value for prediction, and correcting the calculated value for prediction using the learning term. It is.

  The rolled product property prediction method according to the second aspect of the present invention includes a learning data acquisition step, a learning term calculation step, and a prediction step, wherein the learning data acquisition step selects a rolled product to be learned in a rolling line. The step of collecting rolling data at the time of manufacturing as rolling data for learning, the step of measuring the metal structure of the learning target rolled product as a learning metal structure measured value after manufacturing the rolled product of the learning target, Applying a mechanical property model that has been formulated in order to calculate mechanical properties, and calculating for learning the mechanical properties of the rolled product to be learned based on the learning rolling data and the actual measurement value of the metal structure for learning And a step of measuring the mechanical properties of the learning target rolled product as a learning actual value after manufacturing the learning target rolling product. The learning term calculation step acquires a learning term representing a relationship between the learning calculation value and the learning actual measurement value, and the prediction step predicts in the rolling line. The step of collecting rolling data when manufacturing the target rolled product as rolling data for prediction, and after manufacturing the rolling product of the learning target, the metal structure of the rolling product of the prediction target is used as the actual measurement value of the prediction metal structure Applying the mechanical property model, calculating the mechanical properties of the rolling product to be predicted as a prediction calculation value based on the prediction rolling data and the prediction metal structure actual measurement value; Correcting the predicted calculation value using the learning term.

  A rolled product property prediction method according to a third aspect of the present invention includes a learning data acquisition step, a learning term calculation step, and a prediction step, wherein the learning data acquisition step selects a rolled product to be learned in a rolling line. Applying a step of collecting rolling data for manufacturing as rolling data for learning and a metallurgical phenomenon model obtained by formulating the metallurgical phenomenon, and learning the metal structure of the rolled product to be learned based on the rolling data for learning A step of calculating as a calculated value, and a step of actually measuring a metal structure of the rolled product as the learning target as an actual value for learning after manufacturing the rolled product as the learning target, and the learning term calculating step includes: A learning term representing a relationship between the learning calculation value and the learning actual measurement value is acquired, and the prediction step is a prediction target in the rolling line. A step of collecting rolling data when producing a rolled product as rolling data for prediction; applying the metallurgical phenomenon model; and calculating a prediction microstructure of the metal structure of the rolling product to be predicted based on the rolling data for prediction And a step of correcting the calculated value for prediction using the learning term.

  According to the present invention, it is possible to predict the characteristics of a rolled product with high accuracy without causing an increase in calculation load.

Embodiment 1 FIG.
Embodiment 1 of the present invention relates to a method for predicting mechanical properties of a rolled product. FIG. 1 is a schematic diagram of a rolling line and a mechanical property prediction system according to the first embodiment. The mechanical property prediction method for a rolled product according to Embodiment 1 is realized by the mechanical property prediction system shown in FIG.

[Rolling line and mechanical property prediction system of Embodiment 1]
Hereinafter, the rolling line and mechanical property prediction system according to Embodiment 1 will be described. In FIG. 1, a broken line arrow indicates the direction in which the rolled product flows, and a solid line arrow indicates the direction in which the process proceeds in the mechanical property prediction system.

  First, the rolling line will be described. As shown in FIG. 1, the rolling line 10 is a line for producing a rolled product (including a state in the middle of being completed as a product from the rolled material, the same applies hereinafter) from the rolled material (slab). The rolling line 10 includes a heating device 12, a processing device (rolling mill) 14, a cooling device 16, a winding device 18, and a conveyance table 20 that connects these devices. Each device of the rolling line 10 is driven by, for example, an electric motor or a hydraulic device.

  The heating device 12 is provided with a heating furnace for heating the rolling material. The processing device 14 includes a single or a plurality of stands. The processing apparatus 14 includes, for example, a 1-stand reversible rough rolling mill and a 7-stand tandem finish rolling mill. The cooling device 16 has a function of controlling the temperature of the rolled product. The cooling device 16 is a device that cools the rolled product with cooling water. The cooling device 16 includes, for example, a runout table, a cooling table, a forced cooling device, and the like.

  The winding device 18 is a device for winding the rolled product into a coil shape in order to convey the rolled product manufactured in the rolling line 10 to a subsequent process. The conveyance table 20 is an apparatus used for conveying the rolled product to the next process in the rolling line 10 and an apparatus used for conveying the rolled product in each process in the rolling line 10. The conveying device 22 is a device for conveying the rolled product manufactured by the rolling line 10 (taken by the winding device 18) to a subsequent process of the rolling line 10. As this conveying apparatus 22, a walking beam, a conveyor, a crane, a motor vehicle, etc. are used, for example. Of the rolled products produced on the rolling line 10, several to several tens of rolled products are sent to the sampling area 24 shown in FIG. Rolled products other than those sent to the sampling area 24 are sent to the lower process 26. The lower process 26 is a generic name of processes downstream from the rolling process. The lower process 26 includes, for example, a dividing line and a skin pass line. The dividing line is a line including a step of cutting a rolled product manufactured by the rolling line 10 into an appropriate length. The rolling product is cut by, for example, a request from a customer or convenience of carrying out such as weight restriction. The skin pass line is a line including a step of rolling under light pressure from the viewpoint of aesthetics of the rolled product and the improvement of properties such as prevention of stretcher strain.

  The sampling area 24 is an area including an apparatus for unwinding and cutting out a part of the rolled product in order to cut out a test piece to be subjected to a mechanical property test from the rolled product. Note that the rolled product sent to the sampling area 24 is also sent to the lower process after the test strip has been cut off, in the same manner as other rolled products.

  Next, the mechanical property prediction system 30 will be described. The mechanical property prediction system 30 includes rolling data collection means 32, metal structure calculation means 34, mechanical property results collection means 36, and mechanical property prediction means 38.

  The rolling data collection means 32 is not only the actual values such as the temperature and load when the rolling material is rolled by the processing device 14 of the rolling line 10, but also the temperature history of each stand of the processing device 14, the temperature history of the cooling device 16, etc. Collect rolling data including model calculation values using actual values. Furthermore, the rolling data collection means 32 collects the positional information in the longitudinal direction of the rolled product and the chemical components of the rolled material as rolling data.

  The metal structure calculation means 34 calculates the metal structure by performing calculation using a model obtained by formulating the metallurgical phenomenon based on the actual values such as the chemical composition, temperature, load, etc. included in the rolling data described above. To do. The characteristics of the metal structure to be calculated include volume ratios of ferrite, pearlite, bainite, and martensite, and particle sizes of ferrite and austenite. Various models have been proposed for formulating metallurgical phenomena, and a wide range of mathematical formulas representing static recovery, static recrystallization, dynamic recovery, dynamic recrystallization, grain growth, etc. is widely known. It has been. An example of this model is published in Plastic Working Technology Series 7 Plate Rolling (Corona) P198-229. By using this model, the volume fraction of ferrite, pearlite, bainite, martensite, etc., austenite grain size, and ferrite grain size can be calculated.

  The mechanical property result collecting means 36 measures the mechanical property of the rolled product from the test piece cut out from the rolled product in the sampling area 24, and collects the measured value of the mechanical property.

  The mechanical property predicting unit 38 predicts the mechanical property based on the chemical composition and the like included in the rolling data described above and the calculated metal structure value obtained from the metal structure calculating unit 34.

  Hereinafter, the mechanical property achievement collecting unit 36 and the mechanical property predicting unit 38 will be described in detail. FIG. 2 is a block diagram showing a mechanical property achievement collecting unit and a mechanical property predicting unit according to the first embodiment.

  As shown in FIG. 2, the mechanical property result collecting unit 36 includes a mechanical property result measuring unit 40 and a mechanical property result value transmitting side storage unit 42. The mechanical property predicting unit 38 includes a mechanical property calculating unit 44, a mechanical property result calculation value storage unit 46, a mechanical property result value receiving side storage unit 48, a mechanical property new data confirmation unit 50, , Mechanical property learning term calculation means 52 and mechanical property learning term storage means 54 are provided.

  The mechanical property result measuring means 40 cuts out a test piece from the rolled product, performs tests such as a tensile test, a compression test, and a hardness test on the test piece, and sets the mechanical property of the rolled product as a measured mechanical property value. measure. The mechanical properties to be measured include tensile strength and yield strength. Then, the mechanical property result value transmission side storage means 42 stores the measured mechanical property value.

  The mechanical property calculation means 44 calculates the mechanical property of the rolled product as the mechanical property calculation value based on the chemical composition contained in the rolling data described above and the metal structure calculation value obtained from the metal structure calculation means 34. To do. The mechanical properties to be calculated are the same as those measured by the mechanical property performance measuring means 40. As an example of the mechanical property calculation means 44, there is an example published in the 173rd and 174th Nishiyama Memorial Technology Course "Structural Change and Material Prediction of Hot Rolled Steel" (Japan Steel Institute) P124-127. is there. The mechanical property result calculation value storage means 46 stores the mechanical property calculation value.

  The mechanical property actual value receiving side storage unit 48 receives and stores the actual measured mechanical property value stored in the mechanical property actual value transmission side storage unit 42.

  The mechanical property new data confirmation unit 50 sets a new data reception flag when the mechanical property actual value reception side storage unit 48 receives the mechanical property actual measurement value. Thereby, the timing when the mechanical property actual measurement value is acquired is detected.

  The mechanical property learning term calculation means 52 calculates a learning term from the calculated mechanical property value and the actual mechanical property value. The mechanical property learning term storage means 54 stores learning terms.

[Mechanical property prediction method of Embodiment 1]
Hereinafter, a method for predicting the mechanical properties of the rolled product using the above-described mechanical property prediction system 30 will be described. This method includes a learning data acquisition step, a learning term calculation step, and a prediction step. In the following, a method for predicting the mechanical properties of a rolled product for each section value of the thickness of the rolled product will be described as an example.

  As the learning data acquisition step, first, the rolling data collection means 32 collects learning rolling data for manufacturing the learning target rolled product (step S101).

  Next, the metal structure calculation means 34 applies a metallurgical phenomenon model obtained by formulating the metallurgical phenomenon, and calculates the metal structure of the rolled product to be learned as a learning metal structure calculation value based on the learning rolling data ( Step S102).

  Next, the mechanical property calculation means 44 applies a mechanical property model that is formulated to calculate the mechanical property, and based on the learning rolling data and the learning metal structure calculated value, the learning target rolling Calculate the mechanical properties of the product as calculated mechanical properties for learning. And the mechanical property calculation value for learning is memorize | stored by the mechanical property performance calculation value memory | storage means 46 (step S103). At this time, a calculated mechanical property value for learning is calculated and stored for each section value of the thickness of the rolled product.

  Next, the mechanical property result measuring means 40 manufactures a rolled product to be sampled as a learning target, and then measures the mechanical property of the rolled product from the test piece as an actual measured mechanical property value for learning. And the mechanical property actual value for learning is memorize | stored by the mechanical property performance value transmission side memory | storage means 42 (step S104). In this case, the actual measured mechanical property value for learning is measured and stored for each section value of the thickness of the rolled product.

  Further, the stored mechanical property actual value for learning is transmitted to the mechanical property actual value receiving side storage means 48. Then, the actual mechanical property value for learning is received and stored by the mechanical property result value receiving side storage means 48 (step S105). At this time, the actual measured mechanical property value is received and stored for each section value of the thickness of the rolled product.

  And step S101-step S105 are performed in order about several rolled products of learning object. As a result, the calculated mechanical property value for learning and the actual measured mechanical property value for learning for a plurality of learning target rolled products are acquired and stored in each storage means.

  As described above, after storing the learning mechanical property calculation values and the learning mechanical property actual measurement values for the plurality of learning target rolled products, the learning term calculation step is executed.

  The learning term calculation step is executed at a timing when both the latest learning mechanical property calculation value and the learning mechanical property actual measurement value are stored. However, since the measurement by the mechanical property result measuring means 40 is a destructive inspection, it takes several hours to several days to perform a test piece cutting operation and a measurement test. On the other hand, the calculation by the mechanical property calculation means 44 is completed promptly after collecting the rolling data for learning. For this reason, the timing at which the actual measured mechanical property value for learning is acquired is different from the timing at which the calculated mechanical property value for learning is acquired.

  Therefore, in order to detect the timing at which both the latest calculated mechanical property value for learning and the actual measured mechanical property value for learning are stored, the actual measured mechanical property value value is received by the storage means 48 for the actual measured mechanical property value. In accordance with the timing at which the value is received, a new data reception flag is set by the mechanical property new data confirmation means 50. At this time, a new data reception flag is set for each section value of the thickness of the rolled product. Then, in accordance with the timing when the new data reception flag is set, the learning term calculation step is executed by the mechanical property learning term calculation means 52 (step S106). This learning term calculation step may be executed at the timing of the first rolling roll replacement after the new data reception flag is set. Processing by the learning term calculation step is as follows.

  First, among the stored learning mechanical property calculation values, a specified number of learning mechanical property calculation values for the rolled product obtained by actually measuring the actual learning mechanical property value are acquired in order from the newest one. . Similarly, a predetermined number of new ones are acquired in order from the memorized mechanical property actual values for learning. And the ratio of the mechanical property actual value for learning with respect to the mechanical property calculated value for learning is calculated for every rolling product. And the average value of each ratio is calculated and acquired as a learning term. The learning term is stored by the mechanical property learning term storage means 54. FIG. 3 is a diagram illustrating a configuration of a learning term storage table in which learning terms are stored according to the first embodiment. As shown in FIG. 3, the learning term is stored for each section value and mechanical property of the thickness of the rolled product.

The prediction step is executed after the above learning term calculation step is completed.
As the prediction step, first, the rolling data collecting means 32 collects rolling data for producing a rolling product to be predicted on the rolling line as rolling data for prediction (step S107).

  Next, the metal structure calculation means 34 applies a metallurgical phenomenon model obtained by formulating the metallurgical phenomenon, and calculates the metal structure of the rolled product to be predicted as a prediction metal structure calculation value based on the prediction rolling data (step) S108).

  Next, the mechanical property calculation means 44 applies a mechanical property model formulated to calculate the mechanical property, and based on the prediction rolling data and the predicted microstructure calculation value, the rolling of the prediction target The mechanical property of the product is calculated as a predicted mechanical property calculation value (step S109). At this time, the mechanical property calculation value for prediction is calculated for each section value of the thickness of the rolled product to be predicted.

  Furthermore, a learning term corresponding to the section value of the thickness of the rolled product to be predicted is acquired from the learning terms stored in the learning term calculation step. Then, the prediction mechanical property calculation value is corrected by multiplying the prediction mechanical property calculation value by the learning term (step S110).

[Effect of Embodiment 1]
By the above method, the mechanical properties of the rolled product can be predicted with high accuracy without increasing the load on the computer.

[Modification of Embodiment 1]
In this Embodiment 1, it is not limited to the method of estimating a mechanical property for every division value of the plate | board thickness of a rolled product. In the first embodiment, a method of predicting mechanical properties for each section value other than the plate thickness can be applied. This modification can also be applied to the following other embodiments.

  In the first embodiment, the average value of the ratio of the measured mechanical property value for learning to the calculated mechanical property value for learning is acquired as a learning term. In the first embodiment, the average value is acquired as a learning term. The thing is not limited to the average value of the ratio. A result of regression analysis between the ratio and a parameter having a large influence on the ratio may be acquired as a learning term. This modification can also be applied to the following other embodiments.

  In the first embodiment, in the learning term calculation step, the rolling product to be learned and the section value of the sheet thickness match from the stored learning mechanical property calculation value and the stored learning mechanical property actual measurement value. We have a specified number of things to do. The first embodiment is not limited to this method. If the specified number of items that match the learned rolling product and the thickness value of the section value are not stored, they may be acquired from adjacent section values. This modification can also be applied to the following other embodiments.

  In the first embodiment, the mechanical property result calculation value storage means 46, the mechanical property result value reception side storage means 48, the mechanical property result value transmission side storage means 42, and the mechanical property learning term storage means 54 are: It may be composed of one storage medium or may be composed of a plurality of storage media. The mechanical property prediction system 30 according to the first embodiment may be configured from a single computer or may be configured from a plurality of computers connected to each other via a network. Here, the computer is not limited to a personal computer, but generically refers to devices that can implement the mechanical property prediction system 30 of the first embodiment by a program, including arithmetic means, microcomputers, and the like included in information processing devices. .

Embodiment 2. FIG.
The second embodiment of the present invention relates to a method for predicting the metal structure of a rolled product. Below, the description which overlaps with Embodiment 1 is abbreviate | omitted, and demonstrates the metal structure prediction method of the rolled product which concerns on Embodiment 2. FIG.

[Metal structure prediction system of Embodiment 2]
FIG. 4 is a block diagram illustrating a characteristic part of the metallographic structure prediction system according to the second embodiment. The rolled structure metal structure prediction method according to Embodiment 2 is realized by the metal structure prediction system shown in FIG. As shown in FIG. 4, the metal structure prediction system according to the second embodiment includes rolling data collection means 32, metal structure prediction means 56, and metal structure performance collection means 58. Further, the metal structure prediction means 56 includes a metal structure calculation means 34, a metal structure result calculation value storage means 60, a metal structure result value receiving side storage means 62, a metal structure new data confirmation means 64, and a metal structure learning term. Calculation means 66 and metal structure learning term storage means 68 are provided. The metal structure record collecting unit 58 includes a metal structure record measuring unit 70 and a metal structure record value transmitting side storage unit 72. In FIG. 4, a broken line arrow indicates a direction in which processing proceeds in the metal structure prediction system.

  The metal structure calculation means 34 performs calculation using a model obtained by formulating the metallurgical phenomenon based on the actual values such as the chemical composition, temperature, and load included in the rolling data described above, thereby converting the metal structure into a metal. Calculate as an organization calculation value. The characteristics of the metal structure to be calculated include volume ratios of ferrite, pearlite, bainite, and martensite, and particle sizes of ferrite and austenite. The metal structure actual calculation value storage means 60 stores the metal structure calculation value.

  The metal structure performance measuring means 70 cuts out a test piece from the rolled product, and measures the characteristics of the metal structure from the test piece as a measured value of the metal structure. The characteristics to be measured are the same as those calculated by the metal structure calculation means 34. As a measuring method, there are a method of measuring from a micrograph of a test piece and a method of measuring by a sensor using ultrasonic waves. The metal structure actual value transmission side storage means 72 stores the metal structure actual measurement value.

  The metal structure actual value receiving side storage means 62 receives and stores the metal structure actual value stored in the metal structure actual value transmission side storage means 72.

  The metal structure new data confirmation unit 64 sets a new data reception flag when the metal structure actual value is received by the metal structure actual value receiving side storage unit 62. Thereby, the timing at which the measured value of the metal structure is acquired is detected.

  The metal structure learning term calculation means 66 calculates a learning term from the metal structure calculation value and the metal structure actual measurement value. The metal structure learning term storage means 68 stores learning terms.

[Metallic structure prediction method of Embodiment 2]
Hereinafter, a method for predicting the metal structure of a rolled product using the above-described metal structure prediction system will be described. This method includes a learning data acquisition step, a learning term calculation step, and a prediction step. In the following, a method for predicting the metal structure of a rolled product for each section value of the thickness of the rolled product will be described as an example.

  As the learning data acquisition step, first, the rolling data collection means 32 collects learning rolling data for manufacturing a learning target rolled product (step S201).

  Next, the metal structure calculation means 34 applies a metallurgical phenomenon model obtained by formulating the metallurgical phenomenon, and calculates the metal structure of the rolled product to be learned as a learning metal structure calculation value based on the learning rolling data. And the metal structure calculation value for learning is memorize | stored by the metal structure performance calculation value storage means 60 (step S202). At this time, the calculated metal structure calculated value is stored for each section value of the thickness of the rolled product.

  Next, after manufacturing the learning target rolled product to be sampled by the metal structure performance measuring means 70, the metal structure of the rolled product is measured as a learning metal structure actual measurement value from the test piece. And the metal structure actual value for learning is memorize | stored by the metal structure performance value transmission side memory | storage means 72 (step S203). At this time, the actual measurement value of the learning metal structure is measured and stored for each section value of the thickness of the rolled product.

  Next, the stored mechanical property actual value for learning is transmitted to the metal structure actual value receiving side storage means 62. The metal structure actual measurement value receiving side storage means 62 receives and stores the learning metal structure actual measurement value (step S204). At this time, the actual measurement value of the learning metal structure is received and stored for each section value of the thickness of the rolled product.

  And step S201-step S204 are performed in order about several rolled products of learning object. In this way, the learning metal structure calculated values and the learning metal structure actual measurement values for the plurality of learning target rolled products are acquired and stored in the respective storage means.

  As described above, the learning term calculation step is executed after storing the learning metal structure calculation values and the learning metal structure actual measurement values for a plurality of learning target rolled products.

  As in the first embodiment, in order to detect the timing at which both the latest learning metal structure calculated value and the learning metal structure actual measurement value are stored, the metal learning result value receiving side storage means 62 uses the learning metal. A new data reception flag is set by the metal structure new data confirmation means 64 in accordance with the timing of receiving the measured structure value. Then, in accordance with the timing when the new data reception flag is set, the learning term calculation step is executed by the metal structure learning term calculation means 66 (step S205). This learning term calculation step may be executed at the timing of the first rolling roll replacement after the new data reception flag is set. Processing by the learning term calculation step is as follows.

  First, from the stored learning metal structure calculated values, the learning metal structure calculated values for the rolled product obtained by actually measuring the learning metal structure actual measurement values are acquired in order from the new one. Similarly, a predetermined number of new ones are acquired in order from the stored actual measured metal structure values. Then, for each rolled product, the ratio of the learning metal structure actual measurement value to the learning metal structure calculation value is calculated. And the average value of each ratio is calculated and acquired as a learning term. The metal structure learning term storage means 68 stores the learning term. The learning term is stored for each section value of the thickness of the rolled product and each characteristic of the metal structure.

The prediction step is executed after the above learning term calculation step is completed.
As a prediction step, first, the rolling data collecting means 32 collects rolling data when manufacturing a rolling product to be predicted in the rolling line as rolling data for prediction (step S206).

  Next, the metal structure calculation means 34 applies a metallurgical phenomenon model obtained by formulating the metallurgical phenomenon, and calculates the metal structure of the rolled product to be predicted as a prediction metal structure calculation value based on the prediction rolling data (step) S207).

  Furthermore, a learning term corresponding to the section value of the thickness of the rolled product to be predicted is acquired from the learning terms stored in the learning term calculation step. Then, the prediction metal structure calculation value is corrected by multiplying the prediction metal structure calculation value by the learning term (step S208).

[Effect of Embodiment 2]
By the above method, the metal structure of the rolled product can be predicted with high accuracy without increasing the load on the computer.

Embodiment 3 FIG.
Embodiment 3 of the present invention relates to a method for predicting mechanical properties of a rolled product. Below, the description which overlaps with Embodiment 1 is abbreviate | omitted, and demonstrates the mechanical property prediction method of the rolled product which concerns on Embodiment 3. FIG.

[Features of Embodiment 3]
FIG. 5 is a block diagram illustrating a characteristic part of the mechanical property prediction system according to the third embodiment. The rolled product mechanical property prediction method according to the third embodiment is realized by the mechanical property prediction system shown in FIG. As shown in FIG. 5, the mechanical property prediction system according to the third embodiment includes a metal structure actual value input means 74 unlike the first embodiment. In FIG. 5, a broken line arrow indicates a direction in which processing proceeds in the mechanical property prediction system.

  The metal structure actual value input means 74 is a means for inputting a metal structure actual measurement value such as a ferrite particle diameter of a metal structure of a rolled product and a volume ratio of each phase, which is obtained by measurement with a sensor using ultrasonic waves or the like. . Unlike the first embodiment, the mechanical property calculation unit 44 according to the third embodiment calculates the mechanical property based on the measured value of the metal structure.

  Therefore, in the third embodiment, the mechanical property calculation means 44 is not affected by an error caused by a model obtained by formulating the metallurgical phenomenon used in the metal structure calculation means 34 described in the first embodiment. For this reason, in Embodiment 3, even when predicting the mechanical properties of the rolled product, it is not affected by this error.

[Effect of Embodiment 3]
By the above method, the mechanical properties of the rolled product can be predicted with high accuracy without increasing the load on the computer.

1 is a schematic diagram of a rolling line and a mechanical property prediction system according to Embodiment 1. FIG. It is a block diagram which shows the mechanical property results collection means and mechanical property prediction means which concern on Embodiment 1. FIG. It is a figure which shows the structure of the learning term storage table which memorize | stores the learning term based on Embodiment 1. FIG. It is a block diagram which shows the characteristic part of the metallographic structure prediction system which concerns on Embodiment 2. FIG. It is a block diagram which shows the characteristic part of the mechanical property prediction system which concerns on Embodiment 3. FIG.

DESCRIPTION OF SYMBOLS 10 Rolling line 24 Sampling area 30 Mechanical property prediction system 32 Rolling data collection means 34 Metal structure calculation means 36 Mechanical property results collection means 38 Mechanical property prediction means 40 Mechanical property results measurement means 44 Mechanical property calculation means 50 Machine New property confirmation data 52 mechanical property learning term calculation means 54 mechanical property learning term storage means 56 metal structure prediction means 58 metal structure results collection means 64 metal structure new data confirmation means 66 metal structure learning term calculation means 68 metal structure Learning term storage means 70 Metal structure performance measurement means 70
74 Metal structure actual value input means

Claims (4)

A learning data acquisition step, a learning term calculation step, and a prediction step,
The learning data acquisition step includes
Collecting rolling data when manufacturing a rolled product to be learned in the rolling line as rolling data for learning;
Applying a metallurgical phenomenon model that formulates the metallurgical phenomenon, calculating a metal structure of the rolled product to be learned as a learning metal structure calculated value based on the learning rolling data;
Applying a mechanical property model that has been formulated in order to calculate mechanical properties, and calculating for learning mechanical properties of the rolled product to be learned based on the rolling data for learning and the calculated value of the microstructure for learning Calculating as a value;
Measuring the mechanical properties of the learning target rolled product as a learning actual value after manufacturing the learning target rolled product, and
The learning term calculation step is to acquire a learning term representing a relationship between the learning calculation value and the learning actual measurement value,
The prediction step includes
Collecting rolling data when producing a rolling product to be predicted in the rolling line as rolling data for prediction;
Applying the metallurgical phenomenon model, calculating a metal structure of the rolled product to be predicted as a prediction metal structure calculation value based on the prediction rolling data;
Applying the mechanical property model, calculating mechanical properties of the prediction rolling product based on the prediction rolling data and the prediction microstructure calculation value as a prediction calculation value;
And a step of correcting the calculated value for prediction using the learning term. A learning data acquisition step, a learning term calculation step, and a prediction step,
The learning data acquisition step includes
Collecting rolling data when manufacturing a rolled product to be learned in the rolling line as rolling data for learning;
After manufacturing the rolled product to be learned, measuring the metal structure of the rolled product to be learned as a measured metal structure measured value;
Applying a mechanical property model that has been formulated in order to calculate mechanical properties, and calculating for learning the mechanical properties of the rolled product to be learned based on the learning rolling data and the actual measurement value of the metal structure for learning Calculating as a value;
Measuring the mechanical properties of the learning target rolled product as a learning actual value after manufacturing the learning target rolled product, and
The learning term calculation step is to acquire a learning term representing a relationship between the learning calculation value and the learning actual measurement value,
The prediction step includes
Collecting rolling data when producing a rolling product to be predicted in the rolling line as rolling data for prediction;
After manufacturing the rolled product of the learning target, measuring the metal structure of the rolled product of the prediction target as a predicted metal structure measured value,
Applying the mechanical property model, calculating mechanical properties of the rolling product to be predicted as a predicted calculated value based on the rolling data for prediction and the actual measurement value of the prediction metal structure; and
And a step of correcting the calculated value for prediction using the learning term. A learning data acquisition step, a learning term calculation step, and a prediction step,
The learning data acquisition step includes
Collecting rolling data when manufacturing a rolled product to be learned in the rolling line as rolling data for learning;
Applying a metallurgical phenomenon model that formulates the metallurgical phenomenon, and calculating a learning structure as a learning calculated value based on the learning rolling data,
Measuring the metal structure of the learning target rolled product as a learning actual value after manufacturing the learning target rolled product, and
The learning term calculation step is to acquire a learning term representing a relationship between the learning calculation value and the learning actual measurement value,
The prediction step includes
Collecting rolling data when producing a rolling product to be predicted in the rolling line as rolling data for prediction;
Applying the metallurgical phenomenon model, calculating a metallographic structure of the rolling product to be predicted as a predicted calculation value based on the rolling data for prediction;
And a step of correcting the calculated value for prediction using the learning term. Performing the learning data acquisition step for at least one learning target rolled product to obtain the learning calculation value and the learning actual measurement value;
In the learning term calculation step, the ratio of the actual measurement value for learning to the calculated value for learning is calculated for each learning target rolled product, and an average value of the ratio is acquired as the learning term,
The method for predicting a property of a rolled product according to any one of claims 1 to 3, wherein, in the step of correcting the calculation value for prediction, the calculation value for prediction is multiplied by the learning term.

Images