Rolling condition setting method for cold rolling mill, cold rolling method, steel sheet manufacturing method, rolling condition setting device for cold rolling mill, and cold rolling mill

Abstract

In a rolling condition setting method for a cold rolling mill, a prediction model is generated using first multi-dimensional data, which is obtained by converting past rolling performance data into multi-dimensional data, as an explanatory variable, and using a control amount of a turning roll and a reduction position of the cold rolling mill as target variables. The past rolling performance data includes three-dimensional steel plate information including a plate path of a material to be rolled on the input side of the cold rolling mill. The rolling condition setting method for the cold-rolling mill includes a step of estimating at least one of a control amount of the turning roll and a reduction position of the cold-rolling mill by inputting second multi-dimensional data, which is generated from three-dimensional steel plate information outside a plate path including a material to be rolled on an input side of the cold-rolling mill, into the prediction model.

Description

Technical Field

[0001] The present invention relates to a rolling condition setting method for a cold rolling mill, a cold rolling method, a method for manufacturing a steel plate, a rolling condition setting device for a cold rolling mill, and a cold rolling mill. Background Art

[0002] Generally speaking, when cold rolling rolled materials such as cold-rolled thin steel plates, it is desired to make the shape (or flatness) of the rolled materials good while maintaining the thickness accuracy in the length direction and width direction of the rolled materials well, so as to perform cold rolling in a state where the passability of the rolled materials is stabilized. On the other hand, for the purpose of reducing fuel consumption based on lightweighting, etc., the demand for difficult-to-roll materials such as thin hard materials with high loads and thin plate thickness before rolling is increasing. When cold rolling such difficult-to-roll materials, in order to suppress the rolling load, the difficult-to-roll materials are thinned in the hot rolling of the previous process and then transported to the cold rolling process.

[0003] In recent years, most of the control factors of cold rolling mills are automatically controlled by actuators installed in the cold rolling mills, and the opportunities for operators to set the control factors of cold rolling mills are decreasing. However, when cold rolling difficult-to-roll materials such as those described above, the coil may be joined to the next coil in a state where the front and rear ends of the coil are bent due to shape defects during hot rolling.

[0004] In the case of a sudden change in the length of the coil due to bending or shape defects, the rolling load, the roll gap of the cold rolling mill, the work roll bending device, the displacement of the intermediate rolls, the roll expansion caused by thermal bulge, etc., which are relative to the roll deflection correction, cannot be absorbed by automatic control in many cases. In addition, the above rolling load also includes the slip rate and torque calculated by the attachment.

[0005] In such a case, the operator sets the rolling pass schedule and the shape control actuator in a way that satisfies the equipment constraints of the cold rolling mill and does not hinder productivity. Therefore, in recent years, the operating speed and thus the productivity of the cold rolling mill are easily affected by the operator's experience and subjective differences.

[0006] Based on this background, Patent Document 1 proposes a method of using a neural network to learn past operating conditions and using the learning results to set the rolling mill of a cold rolling mill. In addition, Patent Document 2 proposes a method of performing shape control by machine learning using image information that contributes to the shape after cold rolling.

[0007] Prior art literature

[0008] Patent Literature

[0009] Patent Document 1: Japanese Patent No. 6705519

[0010] Patent Document 2: Japanese Patent Application Publication No. 2021-30280 Summary of the invention

[0011] Problems to be solved by the invention

[0012] However, in the method described in Patent Document 1, even if the cold rolling mill is in the optimal operating condition at the time of mill setting, the shape of the rolled material on the output side of the cold rolling mill also changes greatly when the plate convexity changes along the length direction. Therefore, there is a possibility of limiting the rolling speed due to poor shape and causing the rolled material to break in the worst case. On the other hand, in the method described in Patent Document 2, it is difficult to quantitatively capture the state of coil bending and poor shape and reflect it on the shape control actuator.

[0013] The present invention has been completed in view of the above-mentioned problems, and its purpose is to provide a rolling condition setting method and rolling condition setting device for a cold rolling mill, which can set rolling conditions that ensure the stability of cold rolling and perform cold rolling with good productivity even when cold rolling is performed on a difficult-to-roll material with a high load and a thin plate thickness before rolling. In addition, another purpose of the present invention is to provide a cold rolling method and a cold rolling mill, which can ensure the stability of cold rolling and perform cold rolling with good productivity even when cold rolling is performed on a difficult-to-roll material with a high load and a thin plate thickness before rolling. In addition, another purpose of the present invention is to provide a method for manufacturing steel plates that can manufacture steel plates with a good yield.

[0014] Means for solving problems

[0015] In order to solve the above-mentioned problems and achieve the purpose, the rolling condition setting method of a cold rolling mill of the present invention uses a prediction model that predicts the state of the rolling object material after cold rolling to set the target rolling conditions of the cold rolling mill when cold rolling the rolling object material, wherein the prediction model is a model generated by using first multidimensional data as an explanatory variable and using the control amount of the steering roller and the pressing position of the cold rolling mill as target variables, the first multidimensional data is data obtained by converting past rolling performance data into multidimensional data, the past rolling performance data includes three-dimensional steel plate information, and the three-dimensional steel plate information includes the plate track outside the rolling object material on the input side of the cold rolling mill. The rolling condition setting method of the cold rolling mill includes the following steps: by inputting second multidimensional data generated based on the three-dimensional steel plate information including the plate track outside the rolling object material on the input side of the cold rolling mill into the prediction model, at least one of the control amount of the steering roller and the pressing position of the cold rolling mill is estimated.

[0016] In order to solve the above problems and achieve the purpose, the cold rolling method of the present invention includes the step of cold rolling a rolling target material using the target rolling conditions of the cold rolling mill changed by the above-mentioned method for setting rolling conditions of the cold rolling mill.

[0017] In order to solve the above problems and achieve the purpose, a method for manufacturing a steel plate according to the present invention includes the step of manufacturing a steel plate using the above cold rolling method.

[0018] In order to solve the above-mentioned problems and achieve the purpose, the rolling condition setting device of the cold rolling mill of the present invention uses a prediction model that predicts the state of the rolling object material after cold rolling to set the target rolling conditions of the cold rolling mill when cold rolling the rolling object material, wherein the prediction model is a model generated by using first multidimensional data as an explanatory variable and using the control amount of the steering roller and the pressing position of the cold rolling mill as target variables, the first multidimensional data is data obtained by converting past rolling performance data into multidimensional data, the past rolling performance data includes three-dimensional steel plate information, and the three-dimensional steel plate information includes the plate track outside the rolling object material on the input side of the cold rolling mill, and the rolling condition setting device of the cold rolling mill has the following mechanism: by inputting the second multidimensional data generated based on the three-dimensional steel plate information including the plate track outside the rolling object material on the input side of the cold rolling mill into the prediction model, at least one of the control amount of the steering roller and the pressing position of the cold rolling mill is estimated.

[0019] In order to solve the above-mentioned problems and achieve the purpose, a cold rolling mill of the present invention includes the rolling condition setting device of the above-mentioned cold rolling mill.

[0020] Effects of the Invention

[0021] According to the rolling condition setting method and rolling condition setting device of the cold rolling mill of the present invention, it is possible to set rolling conditions that ensure the stability of cold rolling and perform cold rolling with good productivity even when cold rolling is performed on a difficult-to-roll material with a high load and a thin plate thickness before rolling. In addition, according to the cold rolling method and cold rolling mill of the present invention, it is possible to ensure the stability of cold rolling and perform cold rolling with good productivity when cold rolling is performed on a difficult-to-roll material with a high load and a thin plate thickness before rolling. In addition, according to the manufacturing method of the steel plate of the present invention, the steel plate can be manufactured with a good yield. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram showing the structure of a cold rolling mill according to one embodiment of the present invention.

[0023] Figure 2 Yes means Figure 1 A block diagram of the structure of the arithmetic unit shown.

[0024] Figure 3This is a diagram showing an example of multi-dimensional array information.

[0025] Figure 4 It is a diagram showing a configuration example of an actuator prediction model.

[0026] Figure 5 This is a flowchart showing the flow of processing for converting multi-dimensional array information into one-dimensional information.

[0027] Figure 6 This is a flowchart showing the processing flow of the prediction model execution unit. DETAILED DESCRIPTION

[0028] Hereinafter, with reference to the accompanying drawings, a rolling condition setting method for a cold rolling mill, a cold rolling method, a method for manufacturing a steel plate, a rolling condition setting device for a cold rolling mill, and a cold rolling mill as one embodiment of the present invention will be described. In addition, the embodiments shown below illustrate devices and methods for concretizing the technical ideas of the present invention, and the materials, shapes, structures, configurations, etc. of the constituent parts are not limited to the embodiments shown below. In addition, the accompanying drawings are schematic. Therefore, it should be noted that the relationship and ratio between the thickness and the plane size are different from the actual situation, and the drawings also include parts with different relationship and ratio of each other's dimensions.

[0029] 〔Structure of cold rolling mill〕

[0030] First, refer to Figure 1 The structure of a cold rolling mill as one embodiment of the present invention is described. In addition, in this specification, "cold rolling" is sometimes abbreviated as "rolling", and in this specification, "cold rolling" and "rolling" have the same meaning. In addition, in the following description, a steel plate is cited as an example of a rolled material (rolling object material) rolled by a cold rolling mill. However, the rolled material is not limited to a steel plate, and other metal plates such as an aluminum plate can also be applied.

[0031] Figure 1 Schematic diagram showing the structure of a cold rolling mill according to an embodiment of the present invention. Figure 1 As shown in FIG. 1 , a cold rolling mill 1 as one embodiment of the present invention is a cold rolling mill 1 which ... as shown in FIG. Figure 1 The paper side is the left side) toward the output side (towards Figure 1 The cold rolling mill 1 is a cold rolling mill having five rolling stands, namely, the first rolling stand to the fifth rolling stand (#1STD to #5STD), in sequence. In the cold rolling mill 1, tension rollers, differential rollers, plate thickness gauges, and shape gauges (not shown) are appropriately provided between adjacent rolling stands. The structure of the rolling stand, the conveying device of the steel plate S, etc. are not particularly limited, and known technologies can be appropriately applied.

[0032] exist Figure 1In the illustrated embodiment, the steel sheet S rolled by the hot rolling line (not shown) is discharged from the pay-off drum 2 and passes through the input-side looper 3 and the steering roller 4. Then, the steel sheet S is cold-rolled by the cold rolling mill 1 and then coiled by the coiler 5. The length of the input-side looper 3 and the number of the steering rollers 4 are not particularly limited.

[0033] 〔Actuator prediction model〕

[0034] Next, refer to Figure 1 to Figure 6 , an actuator prediction model as one embodiment of the present invention is described.

[0035] The functions associated with the actuator prediction model as one embodiment of the present invention are described by Figure 1 The rolling control device 100, the computing unit 200 and the steel plate three-dimensional information measuring device 300 shown in FIG. Figure 2 The operation monitoring device 400 shown is implemented.

[0036] The rolling control device 100 controls the rolling conditions of the cold rolling mill 1 based on the control signal from the calculation unit 200 .

[0037] Figure 2 Yes means Figure 1 The block diagram of the structure of the computing unit 200 is shown in FIG. Figure 2 As shown, the computing unit 200 includes a computing device 210 , an input device 220 , a storage device 230 , and an output device 240 .

[0038] The computing device 210 is wiredly connected to the input device 220, the storage device 230, and the output device 240 via the bus 250. However, the computing device 210, the input device 220, the storage device 230, and the output device 240 are not limited to this connection method, and may be wirelessly connected or may be connected in a combination of wired and wireless connections.

[0039] The input device 220 functions as an input port for inputting various information. For example, control information of the cold rolling mill 1 by the rolling control device 100 is input to the input device 220. In addition, the input device 220 also inputs the rolling input side steel plate information (including the three-dimensional steel plate information outside the plate track of the steel plate S on the input side of the cold rolling mill 1 (for example, steel plate coordinates, curvature, steepness, etc.)) measured by the steel plate three-dimensional information measuring device 300 and the information from the operation monitoring device 400.

[0040] The operation monitoring device 400 is installed in the production line of the steel plate S, and is composed of, for example, an input device (such as a keyboard, a mouse, etc.) for the operator to perform various settings, a display device (such as a liquid crystal display, etc.) for monitoring the rolling status, etc. The information from the operation monitoring device 400 includes execution instruction information of the actuator prediction model. In addition, the information from the operation monitoring device 400 includes information related to the steel plate S as the rolling object (pre-process conditions, steel type, size), and cold rolling condition information (numerical information, text information, image information) set by a process computer or an operator before cold rolling.

[0041] The storage device 230 is a device composed of, for example, a hard disk drive, a semiconductor drive, an optical drive, etc., and stores information required in this system (information required to realize the functions of the prediction model generation unit 214 and the prediction model execution unit 215 described later).

[0042] The information required to realize the function of the prediction model generation unit 214 includes, for example, the three-dimensional steel plate information on the rolling input side measured by the steel plate three-dimensional information measuring device 300, the required characteristics of the steel plate S (steel type, plate thickness, plate width, etc. of the product), and the equipment constraints of the cold rolling mill 1. In addition, the information required to realize the function of the prediction model generation unit 214 includes the rolling information of the steel plate S after passing through the welding point (including coil information, shape actuator position), and the properties of the coolant used in the rolling stand. In addition, the information required to realize the function of the prediction model generation unit 214 includes explanatory variables related to cold rolling such as rolling conditions (including target rolling speed), and information representing target variables such as the cylinder position of the steering roll 4 and the press position information in the first rolling stand.

[0043] Examples of information required to realize the function of the prediction model execution unit 215 include an actuator prediction model for each rolling state of the steel sheet S generated by the prediction model generation unit 214 and various information input to the actuator prediction model.

[0044] The output device 240 functions as an output port for outputting a control signal from the calculation device 210 to the rolling control device 100 .

[0045] The operation monitoring device 400 includes any display device such as a liquid crystal display, an organic display, etc. The operation monitoring device 400 receives various information indicating the operation status of the cold rolling mill 1 from the rolling control device 100, and displays the received information on an operation screen (operation screen) for an operator to monitor the operation status of the cold rolling mill 1.

[0046] The computing device 210 includes a RAM (Random Access Memory) 211 , a ROM (Read Only Memory) 212 , and a computing processing unit 213 .

[0047] The ROM 212 stores a prediction model generation program 212 a and a prediction model execution program 212 b as computer programs.

[0048] The calculation processing unit 213 has a calculation processing function, and is connected to the RAM 211 and the ROM 212 via the bus 250 .

[0049] The RAM 211 , the ROM 212 , and the arithmetic processing unit 213 are connected to the input device 220 , the storage device 230 , and the output device 240 via the bus 250 .

[0050] The calculation processing unit 213 includes a prediction model generation unit 214 and a prediction model execution unit 215 as functional blocks.

[0051] The prediction model generation unit 214 is a processing unit that generates an actuator prediction model. The actuator prediction model is a model generated by using the first multidimensional data as an explanatory variable and the control amount of the steering roll 4 and the pressing position of the cold rolling mill 1 as target variables. The first multidimensional data is data obtained by converting past rolling performance data into multidimensional data. In addition, the "past rolling performance data" includes three-dimensional steel plate information outside the plate track of the steel plate S on the input side of the cold rolling mill 1.

[0052] In the prediction model generation unit 214, a machine learning method is used to associate three-dimensional steel plate information outside the plate pass of the steel plate S before rolling in the past rolling performance of the cold rolling mill 1 with the actuator quantity (cylinder quantity of the steering roll 4 and the pressing position of the first rolling stand).

[0053] For example, by changing the cylinder volume of the turning roller 4 in advance according to the shape and curvature of the steel sheet S, the winding position of the steel sheet S relative to the turning roller 4 can be adjusted, so that the centering can be performed even for the sudden meandering of the steel sheet S. Similarly, by controlling the pressing position (leveling) of the first rolling stand in a manner to correct the one-sided elongation and curvature of the steel sheet S, the one-sided elongation and curvature on the rolling output side can be reduced.

[0054] In this embodiment, a neural network model is used as an actuator prediction model based on a machine learning method. However, the machine learning method is not limited to a neural network, and other well-known machine learning methods may also be used.

[0055] The prediction model generation unit 214 includes a learning data acquisition unit 214a, a pre-processing unit 214b, a first data conversion unit 214c, a model generation unit 214d, and a result storage unit 214e. When the prediction model generation unit 214 receives an instruction to generate an actuator prediction model from the operation monitoring device 400, it executes the prediction model generation program 212a stored in the ROM 212. Thus, the prediction model generation unit 214 functions as the learning data acquisition unit 214a, the pre-processing unit 214b, the first data conversion unit 214c, the model generation unit 214d, and the result storage unit 214e. In addition, the actuator prediction model is updated each time the prediction model generation unit 214 is executed.

[0056] The learning data acquisition unit 214a acquires a plurality of learning data consisting of input performance data (explanatory variables) and output performance data (target variables) as a preliminary process for generating an actuator prediction model. The input performance data includes three-dimensional steel plate information including the outside of the plate track from the steel plate three-dimensional information measuring device 300 in the past rolling performance data. In addition, the output performance data includes the actuator amount (cylinder amount of the steering roll 4 and the pressing position of the first rolling stand) in the past rolling performance data.

[0057] The learning data acquisition unit 214a obtains the above-mentioned input performance data and output performance data from the storage device 230 to create learning data. Each learning data is composed of a set of input performance data and output performance data. In addition, the created learning data is stored in the storage device 230. In addition, the learning data acquisition unit 214a may not store the learning data in the storage device 230, but supply the learning data to the pre-processing unit 214b and the model generation unit 214d.

[0058] The input performance data includes multidimensional array information obtained by linking explanatory variables in the time direction. In this embodiment, as the multidimensional array information, for example, Figure 3 Information as shown.

[0059] Figure 3The example of the case where there are multiple measurement points of the steel plate three-dimensional information measuring device 300 in the width direction including the plate track outside the steel plate S. In this case, the learning data acquisition unit 214a uses the measurement point group continuously measured in the longitudinal direction of the steel plate S as input performance data. By also taking in the information outside the plate track of the steel plate S (equivalent to the part outside the plate edge of the steel plate S), not only the shape defect of the coil itself can be detected, but also the bending relative to the longitudinal direction of the coil becomes input data. In addition, the measurement method of the steel plate three-dimensional information measuring device 300 is not particularly limited, and can be contact type or non-contact type (two-dimensional laser, 3D scanner, gamma ray, X-ray, etc.). In addition, the data after averaging the measurement point group or the data after processing such as spline smoothing can also be used as input performance data. In addition, the information outside the plate track taken into the input performance data is preferably less than 10% of the plate width of the steel plate S on each side.

[0060] Here, it is also assumed that there is no past rolling performance data stored in the storage device 230 (for example, rolling conditions and steel grade conditions for which there is no past performance) and there is a small amount of samples. In this case, the learning data acquisition unit 214a requires the operator to perform one or more cold rolling without using the actuator prediction model. In addition, the more learning data stored in the storage device 230, the higher the prediction accuracy of the actuator prediction model. Therefore, when the number of learning data is less than a preset threshold, the learning data acquisition unit 214a may also require the operator to perform cold rolling without using the actuator prediction model until the number of data reaches the threshold.

[0061] The pre-processing unit 214b processes the learning data acquired by the learning data acquisition unit 214a for generating the actuator prediction model. Specifically, the pre-processing unit 214b standardizes the value range of the input performance data between 0 and 1 as needed in order to allow the neural network model to read the rolling performance data constituting the learning data.

[0062] The input performance data is multi-dimensional information. Therefore, the first data conversion unit 214c uses a convolutional neural network to compress the input performance data into one-dimensional information while retaining the feature quantity (see Figure 4 ). The input performance data is converted into one-dimensional information. Figure 4 The input layer 501 of the neural network model shown is combined.

[0063] Here, refer to Figure 5 , a processing example of the first data conversion unit 214c is described. Figure 5 is a flow chart showing the process of converting multi-dimensional array information into one-dimensional information. Figure 5As shown, the process of converting multidimensional array information into one-dimensional information, that is, the storage method of multidimensional array information has a structure in which the input and output of multiple filters are connected in multiple stages. That is, the process of converting multidimensional array information into one-dimensional information includes, from the input side, a first convolution step S1, a first pooling step S2, a second convolution step S3, a second pooling step S4, and a full connection step S5.

[0064] In the first convolution step S1, the first data conversion unit 214c takes the multi-dimensional array information of 64×64 as input, and outputs a 64×64 first feature map through convolution operation. The first feature map indicates what kind of local features exist in which part of the input array. In the convolution operation, for example, a filter of 3×3 pixels in width and 32 channels is set, the filter application interval is set to 1, and the length of the periphery filled with 0 is set to 1.

[0065] In the first pooling step S2, the first data conversion unit 214c takes the first feature map output by the first convolution step S1 as input, and sets the maximum value within the horizontal 3×vertical 3 pixels of the first feature map as a new pixel. The first data conversion unit 214c performs such operation while staggering pixels and covering the entire map. Thus, in the first pooling step S2, the first data conversion unit 214c outputs a second feature map after compressing the first feature map.

[0066] In the second convolution step S3, the first data conversion unit 214c takes the second feature map as input and outputs the third feature map through convolution operation. In the convolution operation, for example, a filter with 3 pixels in width and 3 pixels in height and 32 channels is set, the filter application interval is set to 1, and the length of the peripheral filling (filling) with 0 is set to 1.

[0067] In the second pooling step S4, the first data conversion unit 214c uses the third feature map output by the second convolution step S3 as input, and sets the maximum value within the horizontal 3×vertical 3 pixels of the third feature map as a new pixel. The first data conversion unit 214c performs such an operation while staggering the pixels and covering the entire map. Thus, in the second pooling step S4, the first data conversion unit 214c outputs a fourth feature map after compressing the third feature map.

[0068] In the fully connected step S5, the first data conversion unit 214c arranges the information of the fourth feature map output by the second pooling step S4 into a row. Then, the 100 neurons output from the fully connected step S5 become Figure 4The input layer 501 of the neural network model shown. In addition, the convolution method and the number of output neurons are not limited to the above. In addition, as a convolutional neural network method, known models such as GoogleNet, VGG16, MOBILENET, EFFICIENTNET, etc. can also be used.

[0069] Return to Figure 2 The model generation unit 214d performs machine learning using a plurality of learning data (including information converted by the first data conversion unit 214c) obtained by the pre-processing unit 214b. Thus, the model generation unit 214d generates an actuator prediction model including three-dimensional steel plate information on the rolling input side as input performance data and actuator quantity as output performance data.

[0070] In this embodiment, since a neural network is used as a machine learning method, the model generation unit 214d generates a neural network model as an actuator prediction model. That is, the model generation unit 214d generates a neural network model as an actuator prediction model by associating input performance data and output performance data in the learning data processed in a manner for generating the actuator prediction model. The neural network model is represented by a function, for example.

[0071] Specifically, the model generation unit 214d sets the hyperparameters used by the neural network model and performs learning based on the neural network model using the hyperparameters. As an optimization calculation of the hyperparameters, the model generation unit 214d first generates a neural network model with several hyperparameters in the hyperparameters changed in stages for the learning data, and selects the hyperparameters that have the highest prediction accuracy for the verification data.

[0072] As hyperparameters, the number of hidden layers, the number of neurons in each hidden layer, the filtering rate in each hidden layer (blocking the transmission of neurons with a certain probability), the activation function in each hidden layer, and the number of outputs are usually set, but they are not limited to this. In addition, the optimization method of hyperparameters is not particularly limited, and grid search that changes parameters in stages, random search that randomly selects parameters, or search based on Bayesian optimization can be used.

[0073] In addition, although the model generation unit 214d is incorporated as a part of the computing device 210, the structure is not limited to this. For example, the actuator prediction models may be generated and stored in advance, and they may be read as appropriate.

[0074] like Figure 4 As shown, the neural network model as the actuator prediction model in this embodiment includes an input layer 501, an intermediate layer 502, and an output layer 503 in order from the input side.

[0075] exist Figure 3The multi-dimensional array information generated in the process is dimensionally compressed while retaining the feature quantity by the learning data acquisition unit 214a using a convolutional neural network, and is stored in the input layer 501 as one-dimensional information.

[0076] The middle layer 502 is composed of a plurality of hidden layers, and a plurality of neurons are arranged in each hidden layer. The number of hidden layers formed in the middle layer 502 and the number of neurons arranged in each hidden layer are not particularly limited. In the middle layer 502, the transmission from a certain neuron to the neuron of the next hidden layer is performed through an activation function together with the weighting of the variable based on the weight coefficient. As the activation function, a known function such as a sigmoid function, a hyperbolic tangent function, or a ramp function can be used.

[0077] The output layer 503 combines the information of the neurons transmitted by the intermediate layer 502 and outputs it as the final actuator quantity (actuator information). The number of outputs configured in the output layer 503 is not particularly limited. Based on the output results, the three-dimensional steel sheet information of the steel sheet S before cold rolling in the past, and the actuator performance at this time (the cylinder quantity of the steering roll 4 and the pressing position of the first rolling stand), the weight coefficients in the neural network model are gradually optimized, thereby performing learning.

[0078] After learning the weight coefficients of the neural network model, the model generation unit 214d inputs the evaluation data (three-dimensional steel plate information of the steel plate S to be rolled using the actuator prediction model) into the neural network model that has learned the weight coefficients. Thus, the model generation unit 214d obtains an estimation result for the evaluation data.

[0079] Return to Figure 2 The result storage unit 214e stores the learning data, the evaluation data, and the parameters (weight coefficients) of the neural network model in the storage device 230. In addition, the result storage unit 214e stores the output results of the neural network model for the learning data and the output results of the neural network model for the evaluation data in the storage device 230.

[0080] The prediction model execution unit 215 predicts the actuator position corresponding to the three-dimensional steel sheet information of the steel sheet S to be rolled using the actuator prediction model generated by the prediction model generation unit 214 before cold rolling of the steel sheet S. Then, the prediction model execution unit 215 determines the actuator position in the steel sheet S to be rolled.

[0081] In order to perform the above-mentioned processing, the prediction model execution unit 215 includes an information reading unit 215a, a second data conversion unit 215b, an actuator prediction unit 215c, an actuator determination unit 215d, and a result output unit 215e. Here, when the prediction model execution unit 215 receives a signal notifying that cold rolling is being performed from the rolling control device 100 via the input device 220, it executes the prediction model execution program 212b stored in the ROM 212. Thus, the prediction model execution unit 215 functions as the information reading unit 215a, the second data conversion unit 215b, the actuator prediction unit 215c, the actuator determination unit 215d, and the result output unit 215e.

[0082] The information reading unit 215 a reads, from the storage device 230 , the rolling conditions of the steel plate S to be rolled, which are set by the process computer and the operator via the operation monitoring device 400 .

[0083] The second data conversion unit 215b performs a process of convolving the multidimensional array information and the one-dimensional information, which will be the input data to the actuator prediction model. The process of the second data conversion unit 215b is the same as the process of the first data conversion unit 214c, so the detailed description of the process is omitted. In addition, the first data conversion unit 214c and the second data conversion unit 215b can also be sub-routined as one processing unit.

[0084] The actuator prediction unit 215c inputs the one-dimensional information convolved by the second data conversion unit 215b to the actuator prediction model, and predicts the actuator amount (cylinder amount of the steering roll 4 and the pressing position of the first rolling stand) in the steel plate S to be rolled. That is, the actuator prediction unit 215c inputs the second multidimensional data generated based on the three-dimensional steel plate information outside the plate track of the steel plate S on the input side of the cold rolling mill 1 to the actuator prediction model. Thus, at least one of the control amount of the steering roll 4 and the pressing position of the cold rolling mill 1 is estimated.

[0085] The actuator determination unit 215d performs processing for determining whether the predicted actuator amount of the steel sheet S is within a separately set equipment restriction threshold.

[0086] The result output unit 215e operates when the predicted actuator amount of the steel plate S is within the preset equipment restriction threshold. Then, the result output unit 215e outputs the operating conditions (shape control actuator amount) for the determined three-dimensional steel plate shape (three-dimensional steel plate information) of the steel plate S as the rolling object, and performs feedforward control.

[0087] Next, refer to Figure 6 The processing of the prediction model execution unit 215 will be described.

[0088] Figure 6: is a flowchart showing the processing flow of the prediction model execution unit 215. Figure 6 As shown, when the actuator prediction model is executed, first, as a process of step S11, the information reading unit 215a reads a neural network model as an actuator prediction model corresponding to the required characteristics of the steel plate S to be rolled from the storage device 230.

[0089] Next, as a process of step S12, the information reading unit 215a reads the required equipment restriction threshold stored in the storage device 230 from the host computer via the input device 220. Next, as a process of step S13, the information reading unit 215a reads the three-dimensional steel plate shape of the steel plate S as the rolling object stored in the storage device 230 from the host computer via the input device 220.

[0090] Next, as a process of step S14, the actuator prediction unit 215c uses the neural network model as the actuator prediction model read in the process of step S11 to predict the actuator amount. In the process of step S14, specifically, the neural network model is used to obtain the actuator amount for the steel sheet S being cold rolled, corresponding to the input performance data obtained by multi-dimensionally arraying the three-dimensional steel sheet shape of the steel sheet S as the rolling target read in the process of step S13. In addition, the prediction result based on the neural network model is given to the actuator prediction unit 215c. Figure 4 The output layer 503 outputs the neural network model shown.

[0091] Then, when the predicted actuator amount is within the equipment restriction threshold, the result output unit 215e transmits the changed information on the actuator to the rolling control device 100 via the output device 240, and the prediction model execution unit 215 ends a series of processing.

[0092] As described above, in this embodiment, the prediction model generation unit 214 generates an actuator prediction model based on a machine learning method that associates past three-dimensional steel plate shape performance outside the plate track on the rolling input side of the steel plate S with past actuator quantities corresponding to the three-dimensional steel plate shape performance.

[0093] In addition, the prediction model execution unit 215 obtains the three-dimensional steel plate shape of the steel plate S to be rolled using the generated actuator prediction model during the cold rolling of the steel plate S. Then, when the actuator prediction result for the obtained three-dimensional steel plate shape is within the equipment constraint threshold, the prediction model execution unit 215 determines the actuator amount for the three-dimensional steel plate shape of the steel plate S to be rolled. Thus, shape control that is not dependent on the operator's experience, subjective, and satisfies various constraints in the rolling operation is implemented, and the occurrence of faults such as shape defects and breakage during cold rolling can be suppressed, while maintaining productivity.

[0094] Furthermore, in this embodiment, as explanatory variables used in the actuator of the steel sheet S during cold rolling, numerical information extracted from the three-dimensional steel sheet shape performance including the plate track is linked and multi-dimensional array information is used as input data. Therefore, while considering the bending and shape defective parts in the coil length direction near the coil joint part, it is possible to identify the factors that contribute greatly to the shape defect during cold rolling on the neural network model.

[0095] 〔Cold rolling method〕

[0096] In the cold rolling method as one embodiment of the present invention, the steel sheet S is cold rolled using the target rolling conditions of the cold rolling mill 1 changed by the above-mentioned rolling condition setting method. As a result, even when cold rolling is performed on a difficult-to-roll material with a high load and a thin plate thickness before rolling, the stability of the cold rolling can be ensured, and the cold rolling can be performed with good productivity.

[0097] [Method for manufacturing steel plate]

[0098] In the method for manufacturing a steel sheet according to one embodiment of the present invention, the above-mentioned cold rolling method is used to manufacture the steel sheet S. Thereby, the steel sheet S can be manufactured with a good yield.

[0099] [Variation]

[0100] The above is an explanation of the embodiment of the present invention, but the present invention is not limited thereto, and various changes and improvements can be made. For example, in the present embodiment, the determination of the actuator conditions of the steel plate S based on the actuator prediction model is performed over the entire length of the coil, but it can also be performed in a part. In addition, the cold rolling mill 1 is not limited to a 4-stage type, and can also be a multiple rolling mill such as a 2-stage type (2Hi) and a 6-stage type (6Hi), and the number of rolling stands is not particularly limited. In addition, it can also be a multi-roll rolling mill or a Sendzimir rolling mill.

[0101] In addition, when the calculation unit 200 calculates an abnormal control amount exceeding the upper and lower limit values ​​of the change, or when the control amount cannot be calculated, the rolling control device 100 cannot perform control based on the instruction from the calculation unit 200. Therefore, the rolling control device 100 may not perform this implementation when it is determined that the control amount from the calculation unit 200 is abnormal or the control amount is not supplied from the calculation unit 200. In addition, when the measurement result from the steel plate three-dimensional information measurement device 300 exceeds the past learning range, in order to prevent a fracture failure, the roll may be released to decelerate and allow the plate to pass.

[0102] In addition, in the present embodiment, the installation position of the steel plate three-dimensional information measuring device 300 is the upstream side of the steering roll 4, but it can also be installed on the downstream side. In this case, the target variable is only the pressing position of the cold rolling mill 1. In addition, when the feedback control of the pressing position based on the output side shape meter of the first rolling stand fully functions, the target variable can also be set to only the cylinder amount of the steering roll 4.

[0103] In addition, when feedback control is performed using a profile meter installed between rolling stands of the cold rolling mill 1, control oscillations of the present invention may occur. Therefore, the rolling control device 100 may have a mechanism capable of adjusting gain to prevent the control amount from the calculation unit 200 from becoming excessive.

[0104] In addition, Figure 2 In the illustrated configuration example, the output device 240 is not connected to the operation monitoring device 400, but the two may be connected in a communicative manner. Thus, the processing results of the prediction model execution unit 215 (particularly, the shape prediction information of the steel plate S during rolling based on the actuator prediction unit 215c and the changed rolling conditions determined by the actuator determination unit 215d) can be displayed on the operation screen of the operation monitoring device 400.

[0105] In addition, Figure 2 In the illustrated structural example, the prediction model is created and executed in the same computing device 210, which is connected to the rolling control device 100, the steel plate three-dimensional information measuring device 300, and the operation monitoring device 400 via the bus 250, but the present invention is not limited to this structure. For example, the prediction model may be created by a device other than the computing device 210, and the created prediction model may be stored in the ROM 212 and executed by the prediction model execution unit 215.

[0106] Example

[0107] Hereinafter, the present invention will be described based on examples.

[0108] use Figure 1 The cold rolling mill composed of a total of 5 rolling stands of the embodiment shown in the figure conducted the following experiment: a raw material steel plate for electromagnetic steel plate containing 2.5-3.0 mass % Si with a base material thickness of 2.0-2.6 mm and a plate width of 1000 mm was used as a rolling material, and cold rolled to a finishing thickness of 0.300 mm. As a prior learning, first, learning based on a neural network model was implemented using learning data (about 3000 pieces of past steel plate rolling performance data). Thus, the past three-dimensional shape performance of the steel plate before rolling was associated with the past actuator performance, and a neural network model for actuator prediction was created.

[0109] In the invention example, as the past performance data before rolling of the steel plate, the three-dimensional steel plate shape including the plate track is measured by a 3D scanner, and the input performance data is in the form of multi-dimensional array information. As the past actuator information, the cylinder position of the steering roll set after the three-dimensional steel plate shape measuring device and the performance of the pressing position in the first rolling stand are learned. After the leading steel strip and the trailing steel strip are joined at the input side of the cold rolling mill, at the stage when the rolling control device 100 is turned on, the actuator amount in cold rolling based on the generated neural network model is predicted.

[0110] In the comparative example, the following experiment was conducted in the same manner as in the inventive example: a raw material steel sheet for electromagnetic steel sheet containing 2.5-3.0 mass % Si with a base material thickness of 2.0-2.6 mm and a plate width of 1000 mm was used as a rolling material, and cold rolled to a finishing thickness of 0.300 mm. In the comparative example, as the previous performance data of the steel sheet before rolling, a plurality of laser displacement meters were arranged in the steel sheet width direction, and the steel sheet displacement of only the plate track was associated with the actuator amount, and a neural network model for prediction of the actuator was generated.

[0111] The number of fractures of steel plates after rolling 100 coils of the invention example and the comparative example is shown in Table 1. As shown in Table 1, in the comparative example, the bending of the coil was not sufficiently learned. Therefore, when the bending near the joint of the coil changed greatly, the operation amount of the actuator could not keep up, and the coil was separated from the rolling center of the cold rolling mill, resulting in fracture failure caused by unilateral elongation or edge overtension.

[0112]

[0113] From the above, it is confirmed that the cold rolling method and cold rolling mill of the present invention are preferably used to appropriately predict the shape of the steel plate before rolling and appropriately control the operation of the subsequent actuator. In addition, it is confirmed that by applying the present invention, not only can the occurrence of faults such as shape defects and plate breakage during cold rolling be suppressed, but also the productivity and quality of the rolling process and the next process can be greatly improved.

[0114] The above describes the embodiment to which the invention completed by the inventors is applied, but the present invention is not limited to the description and drawings that form a part of the disclosure of the present invention based on the embodiment. That is, other embodiments, examples, and application technologies completed by those skilled in the art based on the embodiment are all included in the scope of the present invention.

[0115] Description of symbols

[0116] 1 Cold rolling mill

[0117] 2 Pay-off reel

[0118] 3 Input side looper

[0119] 4 Steering rollers

[0120] 5 Winding machine

[0121] 100 Rolling control device

[0122] 200 Arithmetic Units

[0123] 210 Computing Device

[0124] 211 RAM(Random Access Memory)

[0125] 212 ROM(Read Only Memory:Read Only Memory)

[0126] 212a Prediction Model Generator

[0127] 212b Prediction Model Execution Program

[0128] 213 Calculation Processing Unit

[0129] 214 Prediction Model Generation Department

[0130] 214a Learning data acquisition unit

[0131] 214b Pre-processing unit

[0132] 214c First data conversion unit

[0133] 214d Model Generation Department

[0134] 214e Result storage unit

[0135] 215 Prediction Model Execution Department

[0136] 215a Information reading unit

[0137] 215b Second data conversion unit

[0138] 215c Actuator prediction unit

[0139] 215d Actuator determination unit

[0140] 215e Result output unit

[0141] 220 Input device

[0142] 230 Storage Device

[0143] 240 Output device

[0144] 300 Steel Plate Three-Dimensional Information Measurement Device

[0145] 400 Operation Monitoring Device

[0146] 501 Input layer

[0147] 502 Middle Layer

[0148] 503 Output layer

[0149] S steel plate.

Claims

1. A method for setting rolling conditions of a cold rolling mill, wherein a prediction model for predicting a state of a rolling target material after cold rolling is used to set target rolling conditions of the cold rolling mill when the rolling target material is cold rolled, wherein: The prediction model is a model generated by using first multidimensional data as explanatory variables and the control amount of the steering roll and the pressing position of the cold rolling mill as target variables, wherein the first multidimensional data is data obtained by converting past rolling performance data into multidimensional data, wherein the past rolling performance data includes three-dimensional steel plate information, and the three-dimensional steel plate information includes the plate track of the rolling object material at the input side of the cold rolling mill, The rolling condition setting method of the cold rolling mill includes the following steps: by inputting second multidimensional data generated based on three-dimensional steel plate information outside the plate track of the rolling object material including the input side of the cold rolling mill into the prediction model, at least one of the control amount of the steering roller and the pressing position of the cold rolling mill is estimated.

2. A cold rolling method, wherein: The method comprises the step of cold rolling a rolling target material using the target rolling conditions of the cold rolling mill changed by the method for setting rolling conditions of the cold rolling mill according to claim 1.

3. A method for manufacturing a steel plate, wherein: The method comprises the step of manufacturing a steel sheet using the cold rolling method according to claim 2.

4. A rolling condition setting device for a cold rolling mill, which sets target rolling conditions of the cold rolling mill when cold rolling the rolling object material using a prediction model for predicting the state of the rolling object material after cold rolling, wherein: The prediction model is a model generated by using first multidimensional data as explanatory variables and the control amount of the steering roll and the pressing position of the cold rolling mill as target variables, wherein the first multidimensional data is data obtained by converting past rolling performance data into multidimensional data, wherein the past rolling performance data includes three-dimensional steel plate information, and the three-dimensional steel plate information includes the plate track of the rolling object material at the input side of the cold rolling mill, The rolling condition setting device of the cold rolling mill has the following mechanism: by inputting the second multidimensional data generated based on the three-dimensional steel plate information outside the plate track of the rolling object material including the input side of the cold rolling mill into the prediction model, at least one of the control amount of the steering roller and the pressing position of the cold rolling mill is estimated.

5. A cold rolling mill, wherein: A rolling condition setting device for a cold rolling mill according to claim 4.

Images