JP7853566B2 - Prediction device, learning device, prediction program, and learning program - Google Patents

Description

This invention relates to a prediction device, a learning device, a prediction program, and a learning program for predicting the properties of steel materials.

Conventionally, a method has been known for using hierarchical neural networks to construct a nonlinear relationship between the material properties of high-strength cold-rolled steel sheets and multiple material-influencing factors, and then using this for material property prediction in continuous annealing processes (for example, Patent Document 1).

Japanese Patent Publication No. 2010-106314

However, the method described in Patent Document 1 above does not take into consideration the prediction of the toughness of the steel material.

This invention has been made in view of the above circumstances, and provides a prediction device, a learning device, a prediction program, and a learning program that can accurately predict the toughness of steel materials.

To achieve the above objective, a prediction device according to a first aspect of the present invention is a prediction device for predicting the toughness of steel material manufactured on a manufacturing line, comprising: an acquisition unit that acquires a plurality of measurement data measured by a plurality of sensors on the manufacturing line for the steel material; and a prediction unit that uses a pre-trained neural network to predict the toughness of the steel material from the plurality of measurement data acquired by the acquisition unit, wherein the neural network is a neural network for predicting parameters relating to a graph representing the toughness of the steel material based on the plurality of measurement data for the steel material manufactured on the manufacturing line.

According to the prediction device of the first aspect of the present invention, by acquiring multiple measurement data points measured by multiple sensors in the manufacturing line for steel materials, and predicting parameters related to a graph representing the toughness of the steel materials using a neural network, the toughness of the steel materials can be predicted with high accuracy.

Here, measurement data refers to data measured by sensors on steel materials being manufactured on the production line.

The prediction device according to the second aspect of the present invention is the same as the prediction device according to the first aspect, wherein the manufacturing line is a hot rolling line and the steel material is a hot-rolled steel sheet.

A prediction device according to a third aspect of the present invention, in a prediction device according to the first or second aspect, wherein the plurality of measurement data includes at least one of the temperature and size of the steel material, measured by the plurality of sensors in the manufacturing line.

The fourth aspect of the present invention is a prediction device according to any one of the first to third aspects, wherein the parameters relating to the graph representing the toughness of the steel material include the temperature at which the ductile fracture surface ratio in the Drop Weight Tear Test (DWTT) reaches a predetermined value, and the shape parameters of the regression curve of the fracture transition temperature in the DWTT.

A fifth aspect of the present invention is a learning device for training a neural network to predict the toughness of steel materials manufactured on a manufacturing line, comprising: an acquisition unit that acquires a plurality of measurement data measured by a plurality of sensors on the manufacturing line for each of a plurality of steel materials; and a learning unit that trains a neural network based on the plurality of measurement data acquired by the acquisition unit and the measured toughness of the steel materials, wherein the neural network is a neural network for predicting parameters relating to a graph representing the toughness of the steel materials based on the plurality of measurement data for the steel materials manufactured on the manufacturing line.

According to the fifth aspect of the present invention, a learning device can accurately predict the toughness of a steel material by acquiring multiple measurement data points from multiple sensors in the manufacturing line for each of the multiple steel materials, and by training a neural network for predicting parameters related to a graph representing the toughness of the steel material based on the multiple measurement data points and the measured toughness of the steel material.

The learning device according to the sixth aspect of the present invention is a learning device according to the fifth aspect, in which the toughness of the steel material is measured from a sample taken from at least one of the leading and trailing ends of the steel material.

A prediction program according to a seventh aspect of the present invention is a prediction program for predicting the toughness of steel manufactured on a manufacturing line. This program causes a computer to function as a prediction unit that uses an acquisition unit to acquire multiple measurement data points measured by multiple sensors on the manufacturing line for the steel, and a pre-trained neural network to predict the toughness of the steel from the multiple measurement data points acquired by the acquisition unit. The neural network is a neural network for predicting parameters related to a graph representing the toughness of the steel based on the multiple measurement data points for the steel manufactured on the manufacturing line.

The eighth aspect of the present invention is a learning program for training a neural network for predicting the toughness of steel materials manufactured on a manufacturing line. This program causes a computer to function as an acquisition unit that acquires multiple measurement data points measured by multiple sensors on the manufacturing line for each of the multiple steel materials, and as a learning unit that trains a neural network based on the multiple measurement data points acquired by the acquisition unit and the measured toughness of the steel materials. The neural network is a neural network for predicting parameters related to a graph representing the toughness of the steel materials based on the multiple measurement data points for the steel materials manufactured on the manufacturing line.

According to this invention, the toughness of steel materials can be predicted with high accuracy.

This figure shows an example of a general configuration of a hot rolling line. This is a diagram showing the configuration of the material prediction device. This diagram explains the parameters related to the graph representing the toughness of steel materials. This diagram explains the parameters related to the graph representing the toughness of steel materials. This is a flowchart of the learning process. This is a flowchart of the material prediction process. This figure shows the correlation between measured values and predicted values from a prediction model for the 50% transition temperature α, shape parameter β, yield stress (YP), and tensile strength (TS).

The embodiments of the present invention will be described below with reference to the drawings. In this embodiment, the example described will be one in which the prediction device and learning device are applied to a material prediction device that predicts the material properties of steel.

<Summary of Embodiments of the Invention>
Until now, the prediction and control of the mechanical properties of hot-rolled steel sheets for electric resistance welded pipes in manufacturing sites have mainly relied on observing the behavior of process data and replacing manufacturing site experience and results with simple regression equations, although physical metallurgical models based on fundamental principles have been referenced. However, this method still lacks sufficient prediction accuracy, and outliers can occur in actual manufacturing, especially for toughness required for thick-walled materials (evaluated by Charpy impact tests or DWTT (Drop Weight Tear Test)). Therefore, there is a need for a more accurate prediction technology for the mechanical properties of hot-rolled steel sheets based on existing manufacturing results.

Improving the accuracy of steel material prediction presents the following challenges: (1) to (5).

(1) It is necessary to improve the accuracy of physical models based on fundamental principles.

(2) It is necessary to improve the accuracy of process models that predict the cooling behavior of steel sheets in factories.

(3) It is necessary to identify measurement errors in process conditions such as process temperature.

(4) It is necessary to clarify the hidden variables in each factory.

(5) It is necessary to identify the probability factors caused by the disturbance.

Until now, material prediction has primarily relied on physical models based on fundamental principles. However, this alone has limitations in predicting the material properties of actual products. This is because factors other than the aforementioned physical models significantly influence prediction accuracy.

In recent years, neural network (NN) models have gained attention as a replacement for physical models for material prediction. In actual manufacturing settings, while some variability exists, certain characteristics are obtained for each manufacturing condition. Therefore, a correlation must exist between manufacturing conditions and material properties. If this correlation can be modeled using NNs with big data on manufacturing conditions linked to factory materials as training data, it may be possible to construct a material prediction model with higher accuracy than physical models. Several prior documents have already explored this approach. For example, Patent Document 1 describes a manufacturing method that uses NNs to derive a nonlinear relationship between the material properties of cold-rolled steel sheets and multiple material-influencing factors. By substituting the target material properties of the cold-rolled steel sheet and the remaining material-influencing factors (excluding one intentional control factor) into this nonlinear relationship, a target value for the intentional control factor that yields the target material properties is determined. This target value is then used to control the intentional control factor, resulting in extremely small material variability between cold-rolled steel sheets.

However, conventional technologies primarily focus on the properties and predictions of thin sheets, and there are currently no technologies for predicting the performance of products with thicker walls and requiring superior toughness.

Furthermore, NN models have a problem. In manufacturing environments, various disturbances exist, and the relationship between measurement data related to process conditions and material properties is not deterministic. Therefore, conventional NNs may suffer from overfitting, potentially failing to minimize generalization errors. As a countermeasure, a Bayesian Neural Network (BNN), which combines NN with Bayesian estimation, is effective, and in this embodiment, a BNN is used as the predictive model.

Furthermore, the toughness of the steel material is predicted using this BNN-based prediction model. Here, toughness (ductile fracture surface ratio) changes rapidly near its transition temperature, and in temperature ranges far from the transition temperature, it only takes on very high or very low values, making it unsuitable for model prediction. Therefore, the transition curve of the toughness value is determined, the transition of that curve is predicted by the model, and the value at a given temperature is determined from the resulting curve.

<Outline of the hot rolling line configuration>
Figure 1 shows an example of a schematic configuration of a hot rolling line, which is an example of an application site for the material prediction device 10.

In Figure 1, the hot rolling line comprises a heating furnace 11, a widthwise rolling mill 13, a roughing mill 14, a finish rolling mill 15, a cooling device (runout table) 16, and a coiling device (coiler) 17.

The heating furnace 11 heats the slab (cast slab) S.

The widthwise rolling mill 13 rolls the slab S, heated in the heating furnace 11, in the widthwise direction.

The roughing mill 14 rolls the slab S, which has been rolled in the width direction by the width-direction rolling mill 13, from above and below to form a rough bar. In the example shown in Figure 1, the roughing mill 14 has a rolling stand 14a consisting only of work rolls, and rolling stands 14b to 14e having work rolls and backup rolls.

The finishing rolling mill 15 further performs hot finishing rolling on the rough bar produced by the rough rolling mill 14 to a predetermined thickness. In the example shown in Figure 1, the finishing rolling mill 15 has seven rolling stands 15a to 15g.

The cooling device 16 cools the hot-rolled steel sheet H (hereinafter simply referred to as steel sheet H) that has undergone hot finishing rolling by the finishing rolling mill 15 with cooling water.

The winding device 17 winds the steel plate H, cooled by the cooling device 16, into a coil shape to manufacture a hot-rolled coil.

Furthermore, the hot rolling line can be realized using known technologies and is not limited to the configuration shown in Figure 1.

Furthermore, a sensor 20 is provided downstream of the heating furnace 11 to measure the size of the slab S extracted from the heating furnace 11. The sensor 20 measures and outputs the size of the slab S. Here, the size of the slab S refers to its thickness.

Furthermore, a sensor 22 is provided downstream of the roughing mill 14 to measure the temperature and size of the rough bar. The sensor 22 measures and outputs the temperature and size of the rough bar. Here, the size of the rough bar refers to its thickness.

Furthermore, a sensor 23 is provided in the upstream section of the finishing rolling mill 15 to measure the temperature of the rough bar. The sensor 23 measures and outputs the temperature of the rough bar.

Furthermore, a sensor 24 is provided downstream of the finishing rolling mill 15 to measure the temperature and size of the steel plate H. The sensor 24 measures and outputs the temperature and size of the steel plate H. Here, the size of the steel plate H refers to its thickness.

Furthermore, a sensor 25 is provided downstream of the cooling device 16 to measure the temperature of the cooled steel plate H. The sensor 25 measures and outputs the temperature of the steel plate H.

Furthermore, a sensor 26 is provided at the winding point of the winding device 17 to measure the temperature of the coiled steel plate H. The sensor 26 measures and outputs the temperature of the coiled steel plate H.

Sensors 20, 22, 23, 24, 25, and 26 are configured using, for example, infrared sensors for measuring temperature, and sensors that capture images, analyze the captured images, and measure size.

<Configuration of the material prediction device>
Next, the configuration of the material prediction device will be described. Figure 2 shows the functional configuration of the material prediction device 10.

As shown in Figure 2, the material prediction device 10 comprises a measurement data acquisition unit 32, a material data acquisition unit 34, a learning unit 36, a prediction model storage unit 38, a prediction unit 40, and a display unit 42.

During the learning phase, the measurement data acquisition unit 32 acquires at least one of the temperature and size measurements for multiple steel materials, measured by multiple sensors 20, 22, 23, 24, 25, and 26 in the hot rolling line.

Specifically, the measurement data acquisition unit 32 acquires the chemical composition input by the input unit (not shown) during learning. The measurement data acquisition unit 32 also acquires the output (size) of sensor 20 for the slab S. Furthermore, the measurement data acquisition unit 32 acquires the output (size and temperature) of sensor 22 and the output (temperature) of sensor 23 for the rough bar. Finally, the measurement data acquisition unit 32 acquires the output (size and/or temperature) of sensors 24, 25, and 26 for the steel plate H.

The measurement data acquisition unit 32 acquires at least one of the temperature and size of the steel material, measured by multiple sensors 20, 22, 23, 24, 25, and 26 in the hot rolling line, during prediction.

Specifically, the measurement data acquisition unit 32 acquires the chemical composition input by the input unit (not shown) during prediction. The measurement data acquisition unit 32 also acquires the output (size) of sensor 20 for the slab S. Furthermore, the measurement data acquisition unit 32 acquires the output (size and temperature) of sensor 22 and the output (temperature) of sensor 23 for the rough bar. Finally, the measurement data acquisition unit 32 acquires the output (size and/or temperature) of sensors 24, 25, and 26 for the steel plate H.

The material data acquisition unit 34 acquires material data measured from samples taken from at least one of the leading and trailing ends of each of the multiple hot-rolled steel sheets, which are input by the input unit (not shown) during learning. The material data relates to toughness, and specifically, it is a parameter related to the graph representing the toughness of the steel material. More specifically, the graph representing the toughness of the steel material is a graph that shows the relationship between the test temperature and the ductile fracture surface ratio in the DWTT (Drop Weight Tear Test), and the material data includes the 50% transition temperature α, which is the temperature at which the ductile fracture surface ratio in the DWTT reaches a predetermined value (50%), and the shape parameter β of the regression curve of the fracture transition temperature in the DWTT (see Figure 3A). Figure 3A shows an example of the 50% transition temperature α and the shape parameter β in the graph that shows the relationship between the test temperature and the ductile fracture surface ratio in the DWTT. The relationship between the test temperature, ductile fracture surface ratio, 50% transition temperature α, and shape parameter β is expressed by the following equation.

Furthermore, the effect of the shape parameter β on the graph curve is shown in Figure 3B. Figure 3B shows an example where the 50% transition temperature α = -40°C and the shape parameter β = 1, 2, 5, and 10.

The sample is collected offline, while unwinding the hot-rolled coil wound by the winding device 17. The material data may further include yield stress and tensile strength.

The learning unit 36 creates training data for each of the multiple hot-rolled steel sheets, consisting of various measurement data acquired by the measurement data acquisition unit 32 and material data acquired by the material data acquisition unit 34.

The learning unit 36 learns a prediction model, which is a neural network for predicting the toughness of steel materials based on various measurement data of steel materials manufactured in the hot rolling line, using the created training data, and stores it in the prediction model storage unit 38. In this embodiment, a Bayesian neural network is used as the neural network. The chemical composition, the size of the slab S at the downstream location of the heating furnace 11, the temperature and size of the rough bar at the downstream location of the rough rolling mill 14, the temperature of the rough bar at the upstream location of the finish rolling mill 15, the temperature and size of the steel plate H at the downstream location of the finish rolling mill 15, the temperature of the cooled steel plate H at the downstream location of the cooling device 16, and the temperature of the steel plate H at the winding location by the winding device 17 are used as inputs to the neural network. Parameters related to the graph representing the toughness of the steel material are used as outputs to the neural network.

Furthermore, the learning unit 36 trains the neural network for each training data set so that the loss function, which represents the difference between the output when various measurement data of the training data is input to the neural network and the material data of the training data (parameters related to the graph representing the toughness of the steel), is minimized.

The prediction unit 40 uses the prediction model stored in the prediction model storage unit 38 to predict parameters related to the graph representing the toughness of the steel material as material data for the steel material, based on various measurement data acquired by the measurement data acquisition unit 32. If the material data further includes yield stress and tensile strength, then yield stress and tensile strength are also predicted.
The prediction unit 40 then uses parameters related to the predicted toughness graph of the steel material to determine the curve of the graph representing the toughness of the steel material and predicts the ductile fracture surface ratio at a predetermined temperature.

The display unit 42 displays the prediction results from the prediction unit 40.

Next, the learning process performed by the material prediction device 10 will be explained with reference to the flowchart shown in Figure 4. The learning process shown in Figure 4 is executed when at least one of the temperature and size measured by multiple sensors 20, 22, 23, 24, 25, and 26 in the hot rolling line for multiple steel materials is input, along with parameters related to a graph representing the toughness of the steel materials, as material data measured from samples taken from the multiple steel materials.

In step S100, the measurement data acquisition unit 32 acquires at least one of the temperature and size measured by multiple sensors 20, 22, 23, 24, 25, and 26 in the hot rolling line for each of the multiple steel materials. The measurement data acquisition unit 32 also acquires the chemical composition input by the input unit (not shown) for each of the multiple steel materials.

In step S102, the material data acquisition unit 34 acquires parameters related to the graph representing the toughness of the steel, which are measured as material data for samples taken from multiple steel materials that were input.

In step S104, the learning unit 36 creates training data for each of the multiple steel materials, consisting of various measurement data acquired by the measurement data acquisition unit 32 and material data acquired by the material data acquisition unit 34.

In step S106, the learning unit 36 trains a prediction model, which is a neural network for predicting the material properties of steel produced on a hot rolling line, based on the created training data, and stores it in the prediction model storage unit 38.

Next, the material prediction process performed by the material prediction device 10 will be explained with reference to the flowchart shown in Figure 5. Note that the material prediction process shown in Figure 5 is performed when at least one of the temperature and size measured by multiple sensors 20, 22, 23, 24, 25, and 26 in the hot rolling line is input for steel materials.

In step S110, the measurement data acquisition unit 32 acquires at least one of the temperature and size of the steel material, which are measured by multiple sensors 20, 22, 23, 24, 25, and 26 in the hot rolling line. The measurement data acquisition unit 32 also acquires the chemical composition input by the input unit (not shown).

In step S112, the prediction unit 40 uses the prediction model stored in the prediction model storage unit 38 to predict parameters related to the toughness graph of the steel material as material data from various measurement data acquired by the measurement data acquisition unit 32. Then, the prediction unit 40 obtains a curve representing the toughness graph of the steel material from the predicted parameters and predicts the ductile fracture surface ratio at a predetermined temperature.

In step S114, the display unit 42 displays the prediction result from the prediction unit 40, and the material prediction process ends.

<Examples>
First, in the training process for building the predictive model, we use measurement data from multiple steel materials and material data measured from samples taken from the tip or tail end as training data.

By inputting measurement data of specific steel materials into a predictive model built through learning, the material properties of the steel are predicted.

Examples of training data used for learning include the following measurement and material data.

The measurement data includes shape data (size (thickness) of the slab S at the downstream section of the heating furnace 11, size (thickness) of the rough bar at the downstream section of the roughing mill 14, and size (thickness) of the steel plate H at the downstream section of the finishing mill 15) and process data (slab heating temperature and heating time of the heating furnace 11, temperature of the rough bar at the downstream section of the roughing mill 14, temperature of the rough bar at the upstream section of the finishing mill 15, temperature of the steel plate H at the downstream section of the finishing mill 15, temperature of the steel plate H at the downstream section of the cooling device 16, temperature of the steel plate H on the cooling table of the cooling device 16, and temperature of the steel plate H at the winding section by the winding device 17).

The input includes measurement data, chemical composition (C, Si, Mn, P, S, Ni, Cr, Mo, Nb, V, Ti, Al, Ca, N), and compositional formula (hardenability (carbon equivalent: e.g., Ceq, see Reference 1)).
[Reference 1]: Suzuki, Iron and Steel 70 (1984) pp. 2179-2187

The material data includes yield stress, tensile strength, 50% transition temperature (α), and shape parameter (β).

The Bayesian neural network used as the prediction model has three hidden layers between the input and output layers. The number of neurons in each hidden layer is 128, 64, and 32, respectively. Furthermore, the activation function for all layers is the hyperbolic tangent function.

Figure 6 shows an example of predicting material data using a trained neural network. Figure 6 shows the correlation between the measured value of the 50% transition temperature α and the predicted value from the prediction model, the measured value of the shape parameter β and the predicted value from the prediction model, the measured value of the yield stress (YP) and the predicted value from the prediction model, and the measured value of the tensile strength (TS) and the predicted value from the prediction model. As shown in Figure 6, good correlations are obtained, indicating that the predictions are accurate.

Thus, in this embodiment, by acquiring multiple measurement data points from multiple sensors in a hot rolling line for the steel material, and using a neural network to predict parameters related to the graph representing the toughness of the steel material, the toughness of the steel material can be predicted with high accuracy.

Furthermore, by acquiring multiple measurement data points from multiple sensors in the hot rolling line for multiple steel materials, and training a neural network to predict parameters related to a graph representing the toughness of the steel based on these multiple measurement data points and the measured material properties of the steel, it is possible to accurately predict the toughness of the steel.

Furthermore, because the toughness of the steel can be predicted with high accuracy, the optimal manufacturing conditions can be determined.

Furthermore, since preferred manufacturing conditions can be determined from chemical components measured in advance, product yield can be improved.

Therefore, in this embodiment, the toughness of the steel material is predicted using a Bayesian neural network.

Furthermore, the neural network training uses training data consisting of measurement data and material data of the steel. Training data for steel judged to be unsatisfactory in material inspection is also used. At this time, data cleaning is performed to reduce the training data for steel judged to be satisfactory in material inspection, thereby increasing the proportion of training data for steel judged to be unsatisfactory.

This allows for highly accurate prediction of steel material properties. Furthermore, since material data measured from samples taken at the longitudinal ends of the steel material is used, training data can be easily generated.

<Different example>
In the above embodiment, the case in which material prediction processing and learning processing are implemented in a single device was described as an example, but it is not limited to this, and the system may be configured by separating it into a prediction device that performs material prediction processing and a learning device that performs learning processing. In this case, the learning device includes a measurement data acquisition unit 32, a material data acquisition unit 34, a learning unit 36, and a prediction model storage unit 38. The prediction device includes a measurement data acquisition unit 32, a prediction unit 40, and a display unit 42.

Furthermore, while this embodiment describes the case where sensors are installed at the downstream section of the heating furnace 11, the downstream section of the roughing mill 14, the upstream and downstream sections of the finishing mill 15, the downstream section of the cooling device 16, and the winding section of the winding device 17, the installation locations are not limited to these. Sensors may be installed at other locations.

Furthermore, the measurement data used as input to the neural network is not limited to the examples described above. Similarly, the material data used as output to the neural network is not limited to the examples described above. In particular, similar learning and prediction are possible for changes in toughness values with respect to test temperature obtained from Charpy impact tests and CTOD (Crack Tip Opening Displacement) tests, which are often used as toughness evaluation tests.

The embodiments of the present invention described above can be realized by a computer executing a program. Furthermore, a computer-readable recording medium on which the program is recorded, and a computer program product such as the program itself, can also be applied as embodiments of the present invention. Examples of recording media that can be used include flexible disks, hard disks, optical disks, magneto-optical disks, CD-ROMs, magnetic tapes, non-volatile memory cards, ROMs, and the like.

10 Material prediction device 11 Heating furnace 13 Width direction rolling mill 14 Rough rolling mill 15 Finish rolling mill 16 Cooling device 17 Winding devices 20, 22, 23, 24, 25, 26 Sensor 32 Measurement data acquisition unit 34 Material data acquisition unit 36 Learning unit 38 Prediction model storage unit 40 Prediction unit 42 Display unit

Claims (8)

A predictive device for predicting the toughness of steel materials manufactured on a production line,
The acquisition unit acquires multiple measurement data, including the temperature and size of the steel material measured by multiple sensors in the manufacturing line, as well as the chemical composition of the steel material.
It includes a prediction unit that uses a pre-trained neural network to predict the toughness of the steel material from the plurality of measurement data and the chemical composition acquired by the acquisition unit,
The neural network is a neural network for predicting parameters relating to a graph representing the toughness of the steel material based on the plurality of measurement data and the chemical composition of the steel material manufactured on the manufacturing line.
The parameters for the graph representing the toughness of the steel material include the temperature at which the ductile fracture surface ratio in the Drop Weight Tear Test (DWTT) becomes 50% , and the shape parameters of the regression curve of the fracture transition temperature in the DWTT. A learning device for training a neural network to predict the toughness of steel materials manufactured on a production line,
An acquisition unit that acquires multiple measurement data, including the temperature and size of the steel material measured by multiple sensors in the manufacturing line, as well as the chemical composition , for each of the multiple steel materials,
The system includes a learning unit that learns a neural network based on the plurality of measurement data and the chemical composition acquired by the acquisition unit, and the measured toughness of the steel material,
The neural network is a neural network for predicting parameters relating to a graph representing the toughness of steel manufactured on the production line, based on the multiple measurement data and the chemical composition of the steel.
The parameters for the graph representing the toughness of the steel material include a learning device that includes the temperature at which the ductile fracture surface ratio in the Drop Weight Tear Test (DWTT) becomes 50% , and the shape parameters of the regression curve of the fracture transition temperature in the DWTT. The learning device according to claim 2, wherein the toughness of the steel material is measured from a sample taken from at least one of the leading and trailing ends of the steel material. A predictive device for predicting the toughness of steel materials manufactured on a production line,
The acquisition unit acquires multiple measurement data , including the temperature and size of the steel material measured by multiple sensors in the manufacturing line, and the chemical composition of the steel material.
A prediction unit that uses the neural network pre-trained by the learning device described in claim 2 to predict parameters relating to a graph representing the toughness of the steel material from the plurality of measurement data and the chemical composition of the steel material acquired by the acquisition unit,
A prediction device that includes this. The aforementioned manufacturing line is a hot rolling line,
The prediction device according to claim 1, wherein the steel material is a hot-rolled steel sheet. The prediction device according to any one of claims 1, 4 to 5, wherein the prediction unit further obtains a curve of the graph representing the toughness of the steel material from parameters relating to the predicted graph representing the toughness of the steel material, and predicts the ductile fracture surface ratio at a predetermined temperature. A prediction program for predicting the toughness of steel materials manufactured on a production line,
Computers,
The system includes an acquisition unit that acquires multiple measurement data, including the temperature and size of the steel material measured by multiple sensors in the manufacturing line, as well as its chemical composition , and a prediction program that functions as a prediction unit that uses a pre-trained neural network to predict the toughness of the steel material from the multiple measurement data and chemical composition acquired by the acquisition unit.
The neural network is a neural network for predicting parameters relating to a graph representing the toughness of the steel material based on the plurality of measurement data and the chemical composition of the steel material manufactured on the manufacturing line.
The parameters for the graph representing the toughness of the steel material include a prediction program that includes the temperature at which the ductile fracture surface ratio in the Drop Weight Tear Test (DWTT) reaches 50% , and the shape parameters of the regression curve of the fracture transition temperature in the DWTT. A learning program for training a neural network to predict the toughness of steel materials manufactured on a production line,
Computers,
A learning program for a learning unit that functions as a learning unit that learns a neural network based on the multiple measurement data and chemical composition of each of the multiple steel materials, including the temperature and size of the steel material measured by multiple sensors in the manufacturing line, and the measured toughness of the steel material, for each of the multiple steel materials,
The neural network is a neural network for predicting parameters relating to a graph representing the toughness of steel manufactured on the production line, based on the multiple measurement data and the chemical composition of the steel.
The parameters for the graph representing the toughness of the steel material include a learning program that includes the temperature at which the ductile fracture surface ratio in the Drop Weight Tear Test (DWTT) reaches 50% , and the shape parameters of the regression curve of the fracture transition temperature in the DWTT.