Apparatus for predicting the surface quality of hot-rolled steel strips and method for manufacturing hot-rolled steel strips

Abstract

The present invention provides a hot-rolled steel strip surface quality prediction device and a hot-rolled steel strip manufacturing method that can predict the surface quality of a hot-rolled steel strip with high accuracy, even when the surface quality or finishing rolling conditions change along the longitudinal direction of the hot-rolled steel strip. [Solution] The hot-rolled steel strip surface quality prediction device 10 comprises a performance data storage unit 20, a training data creation unit 30, a quality prediction model generation unit 40, and a surface quality prediction unit 50. The performance data storage unit 20 acquires and stores performance data of the finishing rolling conditions in the finishing rolling process PP3 in the hot rolling process P1 for each hot-rolled steel strip S to be manufactured, along with longitudinal position information, in multiple performance collection unit lengths L1 divided along the longitudinal direction of the hot-rolled steel strip S. It also acquires and stores performance data of the surface quality, which is the inspection result in the inspection process P3, for each hot-rolled steel strip S that has been manufactured, along with longitudinal position information, in multiple performance collection unit lengths L2 divided along the longitudinal direction of the hot-rolled steel strip S.

Description

[Technical Field]

[0001] This invention relates to a device for predicting the surface quality of hot-rolled steel strips and a method for manufacturing hot-rolled steel strips. [Background technology]

[0002] Hot-rolled steel strip is manufactured through a hot rolling process. The hot rolling process includes a rough rolling process in which the material to be rolled is roughly rolled, a finish rolling process in which the material roughly rolled in the rough rolling process is finish rolled, and a cooling process in which the material finish rolled in the finish rolling process is cooled.

[0003] In the hot-rolled steel strip produced through the hot-rolling process, multiple defects typically occur on the surface. Some of these defects originate in the hot-rolling process, but are difficult to detect during inspection at that stage. A problem arising from this is that similar defects can continue to be produced before they are detected in inspections at the next stage after the hot-rolling process. One solution to this problem is to predict the surface quality of the hot-rolled steel strip at the hot-rolling stage and evaluate the quality of the finishing rolling conditions during that process. By evaluating the finishing rolling conditions during the hot-rolling process, it becomes possible to modify the finishing rolling conditions to prevent defects from occurring on the surface.

[0004] Conventional methods for predicting surface quality include treating defect occurrence as a physical phenomenon and using a physical model created based on knowledge of its mechanism, and using a regression model obtained by performing regression analysis on manufacturing conditions that are thought to affect quality and surface quality.

[0005] Furthermore, in cases where a physical model cannot be constructed because the defect generation mechanism cannot be elucidated, or when regression analysis cannot achieve the necessary accuracy due to the occurrence of defects caused by a combination of factors, there are also methods other than regression to correlate manufacturing conditions with surface quality.

[0006] Conventionally, a quality prediction device, such as the one shown in Patent Document 1, is known for predicting product quality while taking into account changes in product manufacturing conditions and the manufacturing environment.

[0007] The quality prediction device shown in Patent Document 1 includes: an acquisition data identification unit that identifies the product to be quality predicted and the product for learning; a prediction target data acquisition unit that acquires operational performance data of the product to be quality predicted; a learning data generation unit that acquires operational performance data as explanatory variables and quality performance data as the objective variable for the product for learning, and generates learning data; and a learning data selection unit that calculates the similarity of the operational performance data of the product for learning with that of the product to be quality predicted, and selects a predetermined number of learning data in order of decreasing similarity. The quality prediction device also includes: a prediction model generation unit that generates a quality prediction model based on the learning data selected by the learning data selection unit; and a prediction processing unit that uses the quality prediction model to predict the quality of the product to be quality predicted based on the operational performance data of the product to be quality predicted.

[0008] Furthermore, a quality prediction device, such as the one shown in Patent Document 2, is known to predict quality with high accuracy from a large number of operating variables.

[0009] The quality prediction device described in Patent Document 2 divides the operational variable space, which uses operational data as a basis vector, into multiple local regions based on process operation data and quality data. It calculates an activity function based on the operational data that calculates the contribution rate of the local relation equations representing the relationship between quality and operational variables in each local region to the whole. Considering this activity function, it selects only the operational variables that have a high correlation with quality in each local region and calculates local relation equations that represent the relationship between the selected operational variables and quality. Furthermore, the quality prediction device derives a mathematical model representing the relationship between operational variables and quality as a superposition of local regions having local relation equations and activity functions, and selects the minimum error mathematical model from among the mathematical models of multiple division patterns. Furthermore, if the error of the minimum error mathematical model is larger than a given convergence criterion variable, the quality prediction device further subdivides the operational variable space and repeats each step, displaying the converged mathematical model as the analysis result. [Prior art documents] [Patent Documents]

[0010] [Patent Document 1] Japanese Patent Publication No. 2019-74969 [Patent Document 2] Japanese Patent Publication No. 2012-27683 [Overview of the project] [Problems that the invention aims to solve]

[0011] However, the conventional quality prediction devices described in Patent Documents 1 and 2 had the following problems. In other words, among defects that are difficult to detect during inspection in the hot rolling process, surface quality such as the presence or absence of defects and the extent of defects may change along the longitudinal direction of the hot-rolled steel strip. Also, if the hot-rolled steel strip is long, there may be variations in the finishing rolling conditions in the finishing rolling process along the longitudinal direction of the hot-rolled steel strip. In this case, because the surface quality and finishing rolling conditions change along the longitudinal direction of the hot-rolled steel strip, analysis that ignores the positional information of the hot-rolled steel strip may not yield accurate quality prediction results. For example, if an analysis is performed on a hot-rolled steel strip where defects occur locally, by correlating the finishing rolling conditions of a defect-free area with the quality inspection result indicating the presence of defects, a problem may arise in which the quality prediction result indicates that defects will occur under finishing rolling conditions where defects should not occur.

[0012] Here, the acquisition of operational performance data and quality performance data in the quality prediction device shown in Patent Document 1 is, for example, obtained from a portion of the longitudinal direction of the plated steel sheet, which is the product to be predicted and the product used for learning, and is not obtained taking into account the longitudinal position information of the plated steel sheet. The same applies to the quality prediction device shown in Patent Document 2. For this reason, the quality prediction devices shown in Patent Documents 1 and 2 cannot predict with high accuracy the surface quality of hot-rolled steel strip, where the surface quality and finishing rolling conditions change in the longitudinal direction of the hot-rolled steel strip.

[0013] Therefore, the present invention has been made to solve this conventional problem, and its objective is to provide a hot-rolled steel strip surface quality prediction device and a hot-rolled steel strip manufacturing method that can predict the surface quality of a hot-rolled steel strip with high accuracy, even when the surface quality or finishing rolling conditions change in the longitudinal direction of the hot-rolled steel strip. [Means for solving the problem]

[0014] To solve the above problems, a surface quality prediction device for hot-rolled steel strips according to one aspect of the present invention is a surface quality prediction device for hot-rolled steel strips manufactured through a hot-rolling process and surface-inspected in an inspection process after the hot-rolling process, comprising: a finish-rolling condition acquisition unit that acquires actual data of the finish-rolling conditions in the finish-rolling process in the hot-rolling process, along with longitudinal position information, in multiple performance collection unit lengths divided along the longitudinal direction of the hot-rolled steel strip; and a finish-rolling condition performance storage unit that stores the actual data of the finish-rolling conditions acquired by the finish-rolling condition acquisition unit; and a surface quality acquisition unit that acquires actual data of the surface quality, which is the inspection result in the inspection process, along with longitudinal position information, in multiple performance collection unit lengths divided along the longitudinal direction of the hot-rolled steel strip; and a surface quality performance storage unit that stores the actual data of the surface quality acquired by the surface quality acquisition unit. The gist of the system is that it comprises: a performance data storage unit; a training data creation unit that creates paired data by linking the performance data of the finish rolling conditions stored in the performance data storage unit's finish rolling condition performance storage unit with the performance data of the surface quality stored in the performance data storage unit's surface quality performance storage unit, and creates and stores training data, which is a collection of paired data, by performing this paired data creation for multiple hot-rolled steel strips to be manufactured; a quality prediction model generation unit that analyzes the training data stored in the training data creation unit using machine learning to generate and store a surface quality prediction model; and a surface quality prediction unit that inputs the planned finish rolling conditions or actual finish rolling conditions of a newly manufactured hot-rolled steel strip into the surface quality prediction model stored in the quality prediction model generation unit, and predicts the surface quality at multiple prediction points divided along the longitudinal direction of the newly manufactured hot-rolled steel strip.

[0015] Furthermore, another embodiment of the present invention relates to a method for manufacturing hot-rolled steel strips, which includes a surface quality prediction step in which the surface quality of a newly manufactured hot-rolled steel strip is predicted at multiple prediction points divided along the longitudinal direction of the hot-rolled steel strip using the aforementioned hot-rolled steel strip surface quality prediction device. [Effects of the Invention]

[0016] According to the hot-rolled steel strip surface quality prediction device and hot-rolled steel strip manufacturing method of the present invention, it is possible to provide a hot-rolled steel strip surface quality prediction device and hot-rolled steel strip manufacturing method that can predict the surface quality of a hot-rolled steel strip with high accuracy, even when the surface quality or finishing rolling conditions change in the longitudinal direction of the hot-rolled steel strip. [Brief explanation of the drawing]

[0017] [Figure 1] This figure illustrates the manufacturing process of pickled steel strips to which a surface quality prediction device according to one embodiment of the present invention is applied. [Figure 2] This figure illustrates a portion of the hot rolling equipment used to carry out the rough rolling and finish rolling processes shown in Figure 1. [Figure 3] This is a functional block diagram showing the schematic configuration of a hot-rolled steel strip surface quality prediction device according to one embodiment of the present invention. [Figure 4] This diagram illustrates the actual data on finishing rolling conditions. [Figure 5] This diagram illustrates the actual data on surface quality. [Figure 6] This shows actual data for finish rolling conditions, actual data for surface quality, and paired data created by linking the actual data for finish rolling conditions and the actual data for surface quality. (a) shows the actual data for finish rolling conditions, (b) shows the paired data, and (c) shows the actual data for surface quality. [Figure 7] This diagram illustrates an example of calculating representative values ​​for each data collection unit length from actual surface quality data. [Figure 8]This is a flowchart illustrating the processing flow in the training data creation section. [Figure 9] This diagram illustrates the classification of multiple training data sets (training data classification) stored in the training data storage section of the training data creation section. [Figure 10] This is a flowchart illustrating the processing flow in the surface quality prediction model generation unit. [Figure 11] This is a flowchart illustrating the processing flow in the surface quality prediction unit. [Figure 12] This is a flowchart illustrating the processing flow in step S23 in Figure 11. [Figure 13] This is a diagram illustrating an example of display on a prediction result presentation device. [Figure 14] This is a front view, seen from the axial direction of the conveyor roll, showing the schematic configuration of a surface inspection device installed in an inspection line that performs surface inspection of pickled materials during the inspection process. [Figure 15] Figure 14 is a right side view of the surface inspection apparatus shown. [Figure 16] This is a functional block diagram of the image processing unit that constitutes the surface inspection system. [Figure 17] This is a flowchart illustrating the processing flow in an image processing device. [Figure 18] The following are examples of specular reflection and diffuse reflection images of a type of scale defect called a rough surface, with (a) showing an example of a specular reflection image of the rough surface and (b) showing an example of a diffuse reflection image of the rough surface. [Figure 19] The following are examples of specular reflection and diffuse reflection images of surface roughness, a type of scale defect. (a) is a figure showing an example of a specular reflection image of the surface roughness, and (b) is a figure showing an example of a diffuse reflection image of the surface roughness. [Figure 20]The following are examples of specular reflection and diffuse reflection images of scale delamination, a type of scale defect. (a) is a figure showing an example of a specular reflection image of the scale delamination, and (b) is a figure showing an example of a diffuse reflection image of the scale delamination. [Modes for carrying out the invention]

[0018] Embodiments of the present invention will be described below with reference to the drawings. The embodiments shown below are illustrative examples of devices and methods for realizing the technical concept of the present invention, and the technical concept of the present invention is not limited to the following embodiments in terms of the material, shape, structure, arrangement, etc. of the components.

[0019] Furthermore, drawings are schematic representations. Therefore, it should be noted that the relationship and ratios between thickness and planar dimensions may differ from those in reality, and there may be differences in dimensional relationships and ratios between drawings themselves.

[0020] Figure 1 shows the manufacturing process of pickled steel strip for external sale to which the surface quality prediction device according to one embodiment of the present invention is applied. Figure 2 shows a part of the hot rolling equipment for carrying out the rough rolling process and finish rolling process in Figure 1. Pickled steel strips for sale are manufactured by surface-treating a hot-rolled steel strip S (see Figure 2) produced in the hot-rolling process P1 shown in Figure 1 in the downstream pickling process P2, and then performing a surface inspection of the hot-rolled steel strip S in the inspection process P3. The hot-rolling process P1 is carried out by the hot-rolling equipment 1 and includes a heating process PP1 in which a slab (not shown) is heated to a predetermined temperature in a heating furnace (not shown), a rough-rolling process PP2 in which the heated slab is roughly rolled into a sheet bar B of a predetermined thickness in a rough-rolling mill 2, a finish-rolling process PP3 in which the roughly-rolled sheet bar B is finish-rolled into a hot-rolled steel strip S of a predetermined thickness in a finish-rolling mill 3, a cooling process PP4 in which the finish-rolled hot-rolled steel strip S is cooled by a cooling device (not shown), and a winding process PP5 in which the cooled hot-rolled steel strip S is wound up by a winding device (not shown).

[0021] Here, the finishing rolling mill 3 is connected to a finishing rolling control device 4 that controls the finishing rolling process in the finishing rolling mill 3. The finishing rolling control device 4 obtains information such as the steel type and size of the hot-rolled steel strip S that will become the product from the higher-level computer 5, and based on this information, determines the finishing rolling conditions in the finishing rolling process PP3 (finishing rolling mill 3), and controls the finishing rolling mill 3 to perform the finishing rolling under the determined finishing rolling conditions. The finishing rolling conditions include the rolling load, rolling speed, temperature of the material being rolled, composition of the material being rolled, mechanical properties of the material being rolled, amount of cooling water, and the time it takes for the material to pass between stands in the finishing rolling mill 3.

[0022] The problem with pickled materials sold externally is high-temperature scale defects. While the hot rolling process P1 is the cause of these high-temperature scale defects, they become apparent in the pickling process P2, which removes surface scale. Because the process that causes the scale defects and the process that makes them apparent are different, the lead time between the cause and the appearance of the defects delays preventative measures, leading to the occurrence of similar defects.

[0023] One solution to this problem is to identify the correspondence between the manufacturing conditions in the causative process (finish rolling conditions in the finish rolling process) and the surface quality, create a surface quality prediction model, and then use this created surface quality prediction model to predict the surface quality of newly manufactured hot-rolled steel strip S.

[0024] However, surface quality, such as the presence or absence of defects and the extent of defects, may vary along the longitudinal direction of the hot-rolled steel strip S. Furthermore, if the hot-rolled steel strip S is long, there may be variations in the finishing rolling conditions during the finishing rolling process along its longitudinal direction. In this case, because surface quality and finishing rolling conditions change along the longitudinal direction of the hot-rolled steel strip S, an analysis that ignores the positional information of the hot-rolled steel strip S may not yield accurate quality predictions.

[0025] Therefore, in this embodiment, the surface quality prediction device 10 of the hot-rolled steel strip, shown in the functional block diagram in Figure 3, collects and stores the actual finishing rolling conditions and surface quality of the hot-rolled steel strip S as longitudinal data, and, while considering the longitudinal position, finds the correspondence between the finishing rolling conditions and surface quality, and divides multiple prediction points t1~t along the longitudinal direction of the newly manufactured hot-rolled steel strip S into multiple sections. n (See Figure 13) Surface quality is predicted using this method.

[0026] To explain the surface quality prediction device 10 of this hot-rolled steel strip in detail, this surface quality prediction device 10 is a surface quality prediction device for hot-rolled steel strip (pickled material for external sale) S that is manufactured through a hot-rolling process P1 and whose surface is inspected in an inspection process P3 after the hot-rolling process P1. The surface quality prediction device 10 includes a performance data storage unit 20, a training data creation unit 30, a quality prediction model generation unit 40, a surface quality prediction unit 50, and a prediction result presentation device 60.

[0027] The surface quality prediction device 10 is a computer system equipped with a processing unit, and executes the functions of the actual data storage unit 20, the training data creation unit 30, the quality prediction model generation unit 40, and the surface quality prediction unit 50 on software according to the instructions of the installed program. The performance data storage unit 20 includes a finish rolling condition acquisition unit 21 and a finish rolling condition performance storage unit 22, and a surface quality acquisition unit 23 and a surface quality performance storage unit 24.

[0028] The finish rolling condition acquisition unit 21 acquires, for each hot-rolled steel strip S being manufactured, actual data of the finish rolling conditions in the finish rolling process PP3 during the hot rolling process P1 (see Figures 4 and 6(a)), along with longitudinal position information, in multiple actual data collection unit lengths L1 (see Figure 6(a)) divided along the longitudinal direction of the hot-rolled steel strip S. As mentioned above, the finish rolling conditions include the rolling load, rolling speed, temperature of the rolled material, composition of the rolled material, mechanical properties of the rolled material, amount of cooling water, and the time it takes for the rolled material to pass between stands in the finish rolling mill 3. Figure 4 shows an example of actual data for the finish rolling conditions. In Figure 4, there are indicators 1 to N as finish rolling conditions, and each of the indicators 1 to N refers to a finish rolling condition, such as the rolling load and temperature of the rolled material mentioned above. As shown by Indicator 1 in Figure 4 (for example, the temperature of the rolled material), some of the actual finishing rolling conditions vary along the longitudinal direction of the rolled material, while others can be considered constant along the longitudinal direction of the rolled material, as shown by Indicator a (for example, the composition of the rolled material). The actual data for the finishing rolling conditions is expressed by the following equation (1).

[0029]

number

[0030] In equation (1), ID is a product identification ID, and the coil (hot-rolled steel strip) number is used. S is the front and back sides (front and back) of the coil (hot-rolled steel strip). x is the longitudinal position of the coil (hot-rolled steel strip) from which the operational indicators were collected. The actual data for the finishing rolling conditions is calculated by the finishing rolling control device 4 and the calculated value is input to the finishing rolling condition acquisition unit 21, or the actual data for the finishing rolling conditions is detected by a sensor (not shown) and the detected value is input to the finishing rolling condition acquisition unit 21.

[0031] Furthermore, the finishing rolling condition record storage unit 22 stores the actual data of finishing rolling conditions acquired by the finishing rolling condition acquisition unit 21 for each hot-rolled steel strip S that is manufactured.

[0032] In this embodiment, the finishing rolling condition acquisition unit 21 acquires actual data on finishing rolling conditions for each hot-rolled steel strip S to be manufactured, using multiple data collection unit lengths L1 that are divided along the longitudinal direction of the hot-rolled steel strip S. Each data collection unit length L1 is determined according to a predetermined rule. In this embodiment, for example, each data collection unit length L1 is approximately 1 m, converted to the length of the sheet bar B. The finishing rolling condition data storage unit 22 then calculates and stores a representative value for each data collection unit length L1 from the actual data on finishing rolling conditions acquired by the finishing rolling condition acquisition unit 21. In this embodiment, the representative value is the average value of the actual data on finishing rolling conditions for each data collection unit length L1.

[0033] Furthermore, the surface quality acquisition unit 23 acquires, for each manufactured hot-rolled steel strip S, actual surface quality data (see Figures 5 and 6(c)), which is the inspection result in the inspection process P3, along with longitudinal position information in multiple actual data collection unit lengths L2 (see Figure 6(c)) divided along the longitudinal direction of the hot-rolled steel strip S. In this embodiment, surface quality is information on the presence or absence of scale defects, as shown in Figure 5. In Figure 5, the front surface of the hot-rolled steel strip S shows a defect in the center of the hot-rolled steel strip S in the longitudinal direction, no defects on the leading edge (right end in Figure 5) side of the hot-rolled steel strip S from the center, and no defects on the tail end (left end in Figure 5) side of the hot-rolled steel strip S from the center. Furthermore, on the underside of the hot-rolled steel strip S, a defect is found at the leading edge relative to the longitudinal center of the hot-rolled steel strip S. However, there are no defects on the leading edge side of the defective area of ​​the hot-rolled steel strip S, and there are no defects on the trailing end side of the defective area of ​​the hot-rolled steel strip S. The actual surface quality data is expressed by the following equation (2).

[0034]

number

[0035] In equation (2), ID is a product identification ID, and the coil (hot-rolled steel strip) number is used. S is the front and back sides (front and back) of the coil (hot-rolled steel strip). y is the longitudinal position of the coil (hot-rolled steel strip) from which surface quality data was collected. Furthermore, the actual surface quality data is input to the surface quality acquisition unit 23 from defect data measured by a surface defect meter (not shown) in inspection process P3.

[0036] Furthermore, the surface quality performance data I acquired by the surface quality acquisition unit 23 is stored in the surface quality performance data storage unit 24. ID This is stored for each hot-rolled steel strip S that is manufactured. In the surface quality acquisition unit 23, for each hot-rolled steel strip S manufactured, actual surface quality data is acquired in multiple actual data collection unit lengths L2, which are divided along the longitudinal direction of the hot-rolled steel strip S. Each actual data collection unit length L2 is determined according to a predetermined rule. In this embodiment, for example, each actual data collection unit length L2 is set to approximately 1m, which is the length of the sheet bar B.

[0037] Then, the surface quality performance data storage unit 24 calculates and stores representative values ​​for each performance data collection unit length L2 from the surface quality performance data acquired by the surface quality acquisition unit 23. The method for calculating these representative values ​​of surface quality performance data will be explained with reference to Figure 7. Figure 7 shows an example of calculating representative values ​​for each performance data collection unit length L2 from surface quality performance data. In Figure 7, the leftmost section is an enlarged view of the area indicated by A in Figure 6(c). The leftmost section of Figure 7 shows the raw data of surface quality measured by a surface defect meter. This raw data is expressed by priority and defect type, with lower priority numbers indicating higher priority.

[0038] First, in the leftmost part of Figure 7, the defects are represented in order from top to bottom along the longitudinal direction of the coil (hot-rolled steel strip S) as follows: priority "00001" and defect type "KA", priority "00010" and defect type "KB", priority "00010" and defect type "KB", priority "00010" and defect type "KB", and priority "00100" and defect type "KC". Then, as shown in the middle of Figure 7, the portion of Figure 7 containing the raw data for the hot-rolled steel strip S on the far left is divided by the actual data collection unit length L2.

[0039] Then, as shown on the far right of Figure 7, in the raw data separated by the performance collection unit length L2, the one with the highest priority is calculated as the representative value for the performance collection unit length L2. On the far right of Figure 7, the defect type with the highest priority (priority "00001") among the defects included in the performance collection unit length L2 is KA, so the representative value is KA.

[0040] Furthermore, the training data creation unit 30 includes a training data creation processing unit 31 and a training data storage unit 32. The training data creation processing unit 31 creates paired data represented by equation (3) below (see Figure 6(b)) by linking the actual data of finishing rolling conditions stored in the actual data storage unit 22 of the actual data storage unit 20 with the actual data of surface quality stored in the actual data storage unit 24 of the actual data storage unit 20, and by performing this paired data creation for multiple hot-rolled steel strips S to be manufactured, it creates training data represented by equations (4) and (5) below, which are sets of paired data.

[0041]

number

[0042] In equation (3), ID is a product identification ID, and the coil (hot-rolled steel strip) number is used. S is the front and back sides of the coil (hot-rolled steel strip), with U being the front side and L being the back side. z is the longitudinal position of the coil (hot-rolled steel strip) from which operational indicators and surface quality data were collected.

[0043]

number

[0044]

number

[0045] Here, as shown in Figure 6(b), when creating the paired data represented by equation (3) above, the integrated unit length for each pair of actual data of finishing rolling conditions and actual data of surface quality is l1, and multiple pairs of data at this integrated unit length l1 are created along the longitudinal direction of the hot-rolled steel strip S. Then, the training data creation processing unit 31 classifies the training data represented by equations (4) and (5) above into multiple training data (training data classification 1, ..., training data classification c, ..., training data classification M, see Figure 9) based on the finishing rolling conditions according to predefined rules.

[0046] Furthermore, the training data storage unit 32 stores multiple training data (training data classification 1, ..., training data classification c, ..., training data classification M) classified by the training data creation processing unit 31.

[0047] Next, the processing flow in the training data creation unit 30 will be explained with reference to the flowchart shown in Figure 8. First, in step S1, the training data creation processing unit 31 of the training data creation unit 30 obtains the coil number of the coil (hot-rolled steel strip S) for which surface inspection has been completed in the inspection process. Next, in step S2, the training data creation processing unit 31 reads the actual data of the finishing rolling conditions stored in the finishing rolling condition record storage unit 22, using the coil No. obtained in step S1 as the key (see Figure 6(a)).

[0048] Next, in step S3, the training data creation processing unit 31 reads the surface quality performance data stored in the surface quality performance storage unit 24 using the coil No. obtained in step S1 as a key (see Figure 6(c)). Next, in step S4, the training data creation processing unit 31 creates paired data represented by equation (3) above by linking the actual data for finishing rolling conditions and the actual data for surface quality, while taking into account the reversal of the leading and trailing ends of the hot-rolled steel strip S due to the change in the winding process PP5 and the pickling process P2, and the cut end lengths LL1 and LL2 of the leading and trailing ends of the hot-rolled steel strip S described later, while aligning the longitudinal positions of the actual data for finishing rolling conditions and the actual data for surface quality (see Figure 6(b)).

[0049] Here, the training data creation processing unit 31 selects which data to use in the actual data for finishing rolling conditions and the actual data for surface quality when creating the paired data. Specifically, since the properties of the leading and trailing ends of the hot-rolled steel strip S are unstable, as shown in Figure 6(c), the leading end of the hot-rolled steel strip S is pre-cut by a cut end length LL2, and the trailing end of the hot-rolled steel strip S is pre-cut by a cut end length LL1, before the surface inspection is performed in the inspection process P3. As shown in Figure 6(a), the actual data for finishing rolling conditions contains data corresponding to these cut end lengths LL1 and LL2, so when creating the paired data, the training data creation processing unit 31 discards the data corresponding to these cut end lengths LL1 and LL2.

[0050] Furthermore, the length of each integrated unit when creating the paired data (each prediction point t1~t, described later) n The predicted unit length (equivalent to the length of the predicted unit length) l1 is different from the actual data collection unit length L1 where the actual data of the finishing rolling conditions is stored and the actual data collection unit length L1 where the actual data of the surface quality is stored, and is longer than those actual data collection unit lengths L1 and L2. Here, as shown in Figures 6(a) and (b), the actual data of the finishing rolling conditions P at longitudinal position i-1 i-1 Actual data P of the finishing rolling conditions at longitudinal position i+1 i+1 For example, the results that only partially wrap the data are compared with the data M k When linking to this, a weighted average will be taken according to the wrap lengths ll1 and ll2. Also, as shown in Figures 6(b) and (c), the actual surface quality data I at longitudinal position j-1 j-1and the performance data I of the finish rolling conditions at the longitudinal position j + 1 j+1 For the performance where only a part is wrapped as in the case of the pair data M k When associating them, a weighted average according to the wrap lengths ll3 and ll4 shall be taken.[[ID=�]]

[0051] Next, in step S5, the training data creation processing unit 31 creates the training data represented by the aforementioned equations (4) and (5), which is a set of pair data, by performing the creation of the pair data executed in steps S1 to S4 for a plurality of manufactured hot-rolled steel strips S.

[0052] Next, in step S6, the training data creation processing unit 31 classifies the training data created in step S5 into a plurality of training data according to the finish rolling conditions according to a pre-specified rule. In the example shown in FIG. 9, the training data is classified into training data classification 1, ···, training data classification c, ···, training data classification M. Here, the meaning of "according to the finish rolling conditions according to a pre-specified rule" means, for example, according to a pre-determined rule, for each plate thickness, for each component, for each tensile strength, etc. of the finish rolling conditions. For example, the training data within a predetermined plate thickness range is classified as, for example, training data classification 1, and the training data outside the predetermined plate thickness range is classified as training data classification 2.

[0053] Next, in step S7, the training data storage unit 32 stores the plurality of training data (training data classification 1, ···, training data classification c, ···, training data classification M) classified in step S6 as shown in FIG. 9. Thereby, the processing in the training data creation unit 30 is terminated. Next, the quality prediction model generation unit 40 includes a quality prediction model generation processing unit 41 and a quality prediction model storage unit 42.

[0054] The quality prediction model generation processing unit 41 analyzes each of the multiple training data (training data classification 1, ..., training data classification c, ..., training data classification M) stored in the training data storage unit 32 using machine learning to generate surface quality prediction models F1, ..., F c , , , F M Generates.

[0055] Furthermore, the quality prediction model storage unit 42 stores the surface quality prediction models F1, ..., F for each classification generated by the quality prediction model generation processing unit 41. c , , , F M Save it.

[0056] Next, the processing flow in the quality prediction model generation unit 40 will be explained with reference to the flowchart shown in Figure 10. First, in step S11, the quality prediction model generation processing unit 41 of the quality prediction model generation unit 40 acquires multiple training data (training data classification 1, ..., training data classification c, ..., training data classification M) stored in the training data storage unit 32, and analyzes each of these training data classifications using machine learning to generate surface quality prediction models F1, ..., F c , , , F M Generates.

[0057] While an ensemble tree model is used as the machine learning technique, other machine learning techniques such as neural networks or tree structures may also be used. Next, in step S2, the quality prediction model storage unit 42 of the quality prediction model generation unit 40 stores the surface quality prediction models F1, ..., F generated in step S11. c , , , F M Save it. This completes the processing in the quality prediction model generation unit 40.

[0058] Furthermore, the surface quality prediction unit 50 includes a surface quality prediction processing unit 51. The surface quality prediction processing unit 51 generates surface quality prediction models F1, ..., F for each classification of the training data (training data classification 1, ..., training data classification c, ..., training data classification M) stored in the quality prediction model storage unit 42. c , , , F M Surface quality prediction models F1, ..., F used from c , , , F M Select the surface quality prediction model F1, ..., F c , , , F M The system inputs the planned or actual finishing rolling conditions for the newly manufactured hot-rolled steel strip S, and then generates predictions for multiple sections t1~t along the longitudinal direction of the newly manufactured hot-rolled steel strip S. n Predict the surface quality at multiple prediction points t1~t along the longitudinal direction of the hot-rolled steel strip S. n Number of divisions and predicted locations t1~t n The predicted unit length is the same as the number of divisions of multiple pairs of data in the longitudinal direction of the hot-rolled steel strip S and the integrated unit length l1 of each pair of data, as shown in Figure 13, for each prediction location t1~t n The predicted unit length is represented by l1.

[0059] Next, the processing flow in the surface quality prediction unit 50 will be explained with reference to the flowchart shown in Figure 11. First, in step S21, the surface quality prediction processing unit 51 of the surface quality prediction unit 50 acquires the planned finish rolling conditions or actual finish rolling conditions for the newly manufactured hot-rolled steel strip S.

[0060] Here, the timing at which the surface quality prediction processing unit 51 predicts the surface quality of the newly manufactured hot-rolled steel strip S is as shown in Figure 2, at two times: when the rolled material is roughly rolled in the roughing mill 2 to become a sheet bar B, and when the sheet bar B is finish-rolled in the finishing mill 3 to become a hot-rolled steel strip S.

[0061] When surface prediction processing is performed at the time the sheet bar becomes B, the surface quality prediction processing unit 51 obtains the planned finish rolling conditions for the newly manufactured hot-rolled steel strip S from the finish rolling control device 4. The finish rolling control device 4 obtains information such as the steel type and size of the newly manufactured hot-rolled steel strip S that will become the product from the higher-level computer 5, and calculates the finish rolling conditions for the finish rolling mill 3 based on this information. The calculated finish rolling conditions become the aforementioned "planned finish rolling conditions for the newly manufactured hot-rolled steel strip S".

[0062] The planned finishing rolling conditions are acquired along with longitudinal position information for multiple unit lengths L1 (see Figure 6(a)) which are divided along the longitudinal direction of the hot-rolled steel strip S. Furthermore, when surface prediction processing is performed at the time the hot-rolled steel strip S is produced, the surface quality prediction processing unit 51 obtains the actual finishing rolling conditions for the newly produced hot-rolled steel strip S from the finishing rolling control device 4. The finishing rolling control device 4 calculates the finishing rolling conditions actually performed by the finishing rolling mill 3. These calculated finishing rolling conditions become the aforementioned "actual finishing rolling conditions for the newly produced hot-rolled steel strip S".

[0063] The actual results of these finishing rolling conditions are acquired along with longitudinal position information for multiple unit lengths L1 (see Figure 6(a)) which are divided along the longitudinal direction of the hot-rolled steel strip S. The surface quality prediction processing unit 51 may also detect the actual finishing rolling conditions of the newly manufactured hot-rolled steel strip S using a sensor (not shown) installed on the exit side of the finishing rolling mill 3, and acquire the detected value from the sensor.

[0064] Next, in step S22, the surface quality prediction processing unit 51 generates surface quality prediction models F1, ..., F for each classification of the training data (training data classification 1, ..., training data classification c, ..., training data classification M) stored in the quality prediction model storage unit 42. c , , , F M Surface quality prediction models F1, ..., F used from c , , , F M Select this option. Next, in step S23, the surface quality prediction processing unit 51 selects the surface quality prediction model F1, ..., F in step S22. c , , , F M Then, the planned or actual finishing rolling conditions for the newly manufactured hot-rolled steel strip S, obtained in step S21, are input, and multiple predicted locations t1~t are divided along the longitudinal direction of the newly manufactured hot-rolled steel strip S. n To predict surface quality.

[0065] Here, the processing flow in step S23 will be explained in detail with reference to the flowchart shown in Figure 12. First, in step S231, the surface quality prediction processing unit 51 selects the surface quality prediction model F1, ..., F in step S22. c , , , F M In step S21, the planned or actual finishing rolling conditions for the newly manufactured hot-rolled steel strip S are entered.

[0066] Next, in step S232, the surface quality prediction processing unit 51 divides the newly manufactured hot-rolled steel strip S into multiple prediction locations t1~t along the longitudinal direction. n The probability p of defect occurrence in each case is calculated. Next, in step S233, the surface quality prediction processing unit 51 calculates the multiple prediction locations t1~t calculated in step S232. n For each of the predicted locations t1~t, the probability of defect occurrence p is greater than a predetermined threshold. n Determine whether or not it exists.

[0067] Then, if the result of the judgment in step S233 is YES, the process proceeds to step S234; if the result is NO, the process proceeds to step S235. In step S234, the predicted locations t1~t where the probability of defect occurrence p calculated in step S232 is greater than a predetermined threshold are determined. n Given this, the surface quality prediction processing unit 51 determines the corresponding predicted locations t1~t where the probability of defect occurrence p is greater than the threshold.n The surface quality of the area is predicted to be defective. In step S234, the surface quality prediction processing unit 51 determines the corresponding predicted area t1~t where the probability of defect occurrence p is not greater than the threshold. n Other prediction points t1~t n We predict there will be no defects.

[0068] On the other hand, in step S235, multiple prediction points t1~t calculated in step S232 are used. n Since the probability p of defect occurrence at each of the locations is not greater than a predetermined threshold, the surface quality prediction processing unit 51 predicts that the surface quality of all prediction locations is defect-free. This terminates the process in step S23.

[0069] Once step S23 is completed, proceed to step S24. In step S24, the surface quality prediction processing unit 51 outputs the prediction result from step S23 to the prediction result presentation device 60. This completes the processing in the surface quality prediction unit 50. The prediction result presentation device 60 displays multiple prediction points t1~t along the longitudinal direction of the newly manufactured hot-rolled steel strip S, which were predicted by the surface quality prediction unit 50. n This presents the surface quality prediction results to the surface quality prediction user (not shown).

[0070] As shown in Figure 3, the prediction result presentation device 60 includes a display device 61 that displays the distribution of defects in the longitudinal direction of the hot-rolled steel strip S, and a speaker 62 that communicates whether or not defects have occurred.

[0071] Figure 13 shows an example of the display on the display device 61 of the prediction result presentation device 60. The display device 61 displays a diagram in color that shows the predicted locations where defect D is predicted to occur. In Figure 13, multiple predicted locations t1~t are shown along the longitudinal direction from the leading edge Sa to the trailing edge Sb on the front surface of the hot-rolled steel strip S. n It is divided into parts, and the predicted location t n-2Defect D is predicted to occur at the location indicated, and multiple predicted locations t1~t along the longitudinal direction from the leading edge Sa to the trailing edge Sb on the reverse side of the hot-rolled steel strip S are also predicted. n It is divided into parts, and the predicted location t n-2 Examples of situations where defect D is predicted to occur are shown in that section.

[0072] As described above, the hot-rolled steel strip surface quality prediction device 10 according to this embodiment includes a performance data storage unit 20, a training data creation unit 30, a quality prediction model generation unit 40, and a surface quality prediction unit 50. The performance data storage unit 20 includes a finish rolling condition acquisition unit 21 that acquires performance data of the finish rolling conditions in the finish rolling process PP3 in the hot rolling process P1, along with longitudinal position information, in multiple performance collection unit lengths L1 divided along the longitudinal direction of the hot rolling steel strip S, for each hot-rolled steel strip S that is manufactured, and a finish rolling condition performance storage unit 22 that stores the performance data of the finish rolling conditions acquired by the finish rolling condition acquisition unit 21. It also includes a surface quality acquisition unit 23 that acquires performance data of the surface quality, which is the inspection result in the inspection process P3, along with longitudinal position information, in multiple performance collection unit lengths L2 divided along the longitudinal direction of the hot-rolled steel strip S, for each hot-rolled steel strip S that is manufactured, and a surface quality performance storage unit 24 that stores the performance data of the surface quality acquired by the surface quality acquisition unit 23.

[0073] Furthermore, the training data creation unit 30 creates paired data by linking the actual data of finishing rolling conditions stored in the finishing rolling condition actual data storage unit 22 of the actual data storage unit 20 with the actual data of surface quality stored in the surface quality actual data storage unit 24 of the actual data storage unit 20. This paired data creation is performed for multiple hot-rolled steel strips S to be manufactured, thereby creating training data, which is a collection of paired data. The training data created is then classified into multiple training data sets (training data classification 1, ..., training data classification c, ..., training data classification M) based on the finishing rolling conditions according to predefined rules and stored.

[0074] Furthermore, the quality prediction model generation unit 40 analyzes each of the multiple training data (training data classification 1, ..., training data classification c, ..., training data classification M) stored in the training data creation unit 30 using machine learning to generate surface quality prediction models F1, ..., F c , , , F M Generate and save it. Then, the surface quality prediction unit 50 generates surface quality prediction models F1, ..., F for each classification of the multiple training data (training data classification 1, ..., training data classification c, ..., training data classification M) stored in the quality prediction model generation unit 40. c , , , F M Select a surface quality prediction model to use, input the planned or actual finishing rolling conditions for the newly manufactured hot-rolled steel strip S into the selected surface quality prediction model, and then generate predictions for multiple predicted locations t1~t along the longitudinal direction of the newly manufactured hot-rolled steel strip S. n To predict surface quality.

[0075] This makes it possible to predict the surface quality of the hot-rolled steel strip S with high accuracy, even if the surface quality or finishing rolling conditions change along the longitudinal direction of the hot-rolled steel strip S. As mentioned above, the training data creation unit 30 classifies the training data created based on the finishing rolling conditions according to predefined rules into multiple training data sets (training data classification 1, ..., training data classification c, ..., training data classification M) and saves them. The quality prediction model generation unit 40 then uses machine learning to analyze each classification of the multiple training data sets (training data classification 1, ..., training data classification c, ..., training data classification M) and generates surface quality prediction models F1, ..., F c , , , F M The surface quality prediction unit 50 generates and saves the following. Furthermore, the surface quality prediction unit 50 generates surface quality prediction models F1, ..., F for each classification of the multiple training data (training data classification 1, ..., training data classification c, ..., training data classification M) saved by the quality prediction model generation unit 40. c , , , F MSelect a surface quality prediction model to use, input the planned or actual finishing rolling conditions for the newly manufactured hot-rolled steel strip S into the selected surface quality prediction model, and then generate predictions for multiple predicted locations t1~t along the longitudinal direction of the newly manufactured hot-rolled steel strip S. n To predict surface quality.

[0076] This allows for the creation of surface quality prediction models F1, ..., F suitable for use based on the finishing rolling conditions. c , , , F M Using the selected surface quality prediction model, multiple prediction points t1~t are divided along the longitudinal direction of the newly manufactured hot-rolled steel strip S. n Since surface quality can be predicted using this method, the burden of calculation processing on the surface quality prediction device 10, which has calculation processing capabilities, can be reduced, and the surface quality of the hot-rolled steel strip S can be predicted, thereby improving the accuracy of the surface quality prediction.

[0077] Furthermore, according to the surface quality prediction device 10 of this embodiment, in the finishing rolling condition acquisition unit 21, for each hot-rolled steel strip S to be manufactured, actual data of the finishing rolling conditions is acquired in multiple actual data collection unit lengths L1 which are divided along the longitudinal direction of the hot-rolled steel strip S. Each actual data collection unit length L1 is determined according to a predetermined rule. In addition, the finishing rolling condition actual data storage unit 22 calculates and stores representative values ​​for each actual data collection unit length L1 from the actual data of the finishing rolling conditions acquired in the finishing rolling condition acquisition unit 21. This allows for accurate collection of actual data on finishing rolling conditions in multiple data collection unit lengths L1, which are divided along the longitudinal direction of the hot-rolled steel strip S, when predicting surface quality.

[0078] Furthermore, according to the surface quality prediction device 10 of this embodiment, the surface quality acquisition unit 23 acquires actual surface quality data for each hot-rolled steel strip S to be manufactured, using multiple actual data collection unit lengths L2 that are divided along the longitudinal direction of the hot-rolled steel strip S. Each actual data collection unit length L2 is determined according to a predetermined rule. In addition, the surface quality data storage unit 24 calculates and stores representative values ​​for each actual data collection unit length L2 from the actual surface quality data acquired by the surface quality acquisition unit 23. This allows for accurate collection of actual surface quality data in multiple data collection unit lengths L2, which are divided along the longitudinal direction of the hot-rolled steel strip S, when predicting surface quality.

[0079] Furthermore, according to the hot-rolled steel strip surface quality prediction device 10 of this embodiment, the surface quality prediction unit 50 predicts multiple prediction locations t1~t along the longitudinal direction of the newly manufactured hot-rolled steel strip S. n The system includes a prediction result presentation device 60 that presents the predicted surface quality results to the surface quality prediction user. This allows for the creation of multiple prediction points t1~t along the longitudinal direction of the newly manufactured hot-rolled steel strip S. n The surface quality prediction results can be obtained by the surface quality prediction user by referring to the prediction result display device 60, and the surface quality prediction user can take appropriate action, such as changing the finishing rolling conditions to obtain the desired surface quality.

[0080] Furthermore, according to the hot-rolled steel strip surface quality prediction device 10 of this embodiment, the prediction result presentation device 60 includes a display device 61 that displays the distribution of defects in the longitudinal direction of the hot-rolled steel strip S, and a speaker 62 that communicates whether or not defects have occurred. This allows the surface quality prediction user to divide the newly manufactured hot-rolled steel strip S into multiple prediction points t1~t along the longitudinal direction. n The distribution of defects can be determined by the display device 61, and the presence or absence of defects can be determined by the sound from the speaker 62.

[0081] Furthermore, according to the hot-rolled steel strip surface quality prediction device 10 of this embodiment, the hot-rolled steel strip S is a pickled material that is manufactured through a hot-rolling process P1 and a pickling process P2, and is surface-inspected in an inspection process P3 after the pickling process P2. This allows for the identification of high-temperature scale defects, which are a particular problem in pickled steel materials, by dividing the newly manufactured hot-rolled steel strip S into multiple predicted locations t1~t in the hot-rolling process P1, which precedes the pickling process P2, the process in which scale defects become apparent, along the longitudinal direction of the strip. n This will allow us to predict the surface quality.

[0082] Furthermore, according to the manufacturing method of hot-rolled steel strip according to this embodiment, the surface quality prediction device 10 of the hot-rolled steel strip is used to divide the newly manufactured hot-rolled steel strip S into multiple prediction locations t1~t n It includes a surface quality prediction process that predicts the surface quality. As a result, in the hot rolling process P1 before surface inspection in inspection process P3 for newly manufactured hot-rolled steel strip S, multiple predicted locations t1~t are divided along the longitudinal direction of the hot-rolled steel strip S. n This allows for prediction of surface quality, enabling users of the surface quality prediction system to adjust the finishing rolling conditions to achieve the desired surface quality. This avoids the risk of continuously producing hot-rolled steel strips with similar surface quality defects between the hot-rolling process P1 and the inspection process P3 where surface quality defects are discovered.

[0083] Next, a surface inspection device 100 for performing surface inspection of pickled steel strip S, which is provided in the inspection line that performs surface inspection of pickled steel strip S for external sale SS in inspection process P3, will be described with reference to Figures 14 to 20. Figure 14 shows a schematic configuration of a surface inspection device 100 for performing surface inspection of pickled external sales material SS.

[0084] The surface inspection device 100 shown in Figure 14 is installed in an inspection line that performs surface inspection in inspection process P3, which is carried out after the pickling process P2. It determines the type of scale defect (for example, rough surface D1 (see Figure 18), surface roughness D2 (see Figure 19), scale peeling D3 (see Figure 20)) on the surface Ssa of the pickled external sales material SS that is conveyed in the direction of arrow A by the conveyor roll 101. Rough surface D1, a type of scale defect, is an uneven defect that occurs on the surface Ssa of the pickled external sales material SS and is caused by the manufacturing conditions of the material. Rough surface D1 results in localized exposure of the iron itself. Surface roughness D2 is an uneven defect that occurs on the surface Ssa of the pickled external sales material SS and is caused by the condition of the rolling roll. Surface roughness D2 results in localized high surface roughness. Scale peeling D3 is an uneven defect that occurs on the surface Ssa of the pickled external sales material SS, in which scale is peeled off from the surface Ssa.

[0085] The surface inspection apparatus 100 includes an illumination device 110, an imaging device 120, and an image processing device 130. The lighting device 110 includes specular reflection lighting 111 and oblique lighting 112. As shown in Figure 15, the specular reflection illumination 111 is a rod-shaped light source extending along the width direction (xx direction) of the pickled outer material SS, and specularly reflected light L over the entire width direction (xx direction) of the pickled outer material SS surface SSa of the pickled outer material SS. 111 The specular reflection illumination 111 has a length that covers the entire width direction (xx direction) of the pickled outer coating material SS so that it can be irradiated. 111 It is installed relative to the pickled external sales material SS such that its direction is parallel to the longitudinal direction (yy direction) of the pickled external sales material SS.

[0086] Furthermore, as shown in Figure 14, the specularly reflected light L from the specularly reflected light illuminator 111 is also reflected. 111 The horizontal line HL on the surface SSa, which is the intersection line of the horizontal plane HS and the surface SSa of the pickled outer material SS, is directed downstream in the transport direction at a predetermined irradiation angle (light L 111 (The angle between the horizontal plane HS and the horizontal plane)θ 111It is installed at an angle. The specular reflection illumination 111 employs rod-shaped blue LED illumination, and by frontal reflection, it emphasizes highly specular features, making it possible to identify scale defects (e.g., rough surface D1, surface roughness D2, scale peeling D3) with high identifiability.

[0087] On the other hand, as shown in Figure 15, the oblique light illumination 112 is composed of a first oblique light illumination 112a and a second oblique light illumination 112b, which are divided into two parts in the width direction (xx direction) of the pickled outer material SS.

[0088] The first oblique light illumination 112a is a rod-shaped light source extending along the width direction (xx direction) of the pickled external material SS, and diffusely reflected light L on one side of the width direction (xx direction) of the surface SSa of the pickled external material SS. 112a The length of the first oblique light illumination 112a covers half of the width direction (xx direction) of the pickled outer material SS so that it can be irradiated.

[0089] Furthermore, the second oblique light illumination 112b is a rod-shaped light source extending along the width direction (xx direction) of the pickled external material SS, and diffusely reflected light L on the other side of the width direction (xx direction) of the surface SSa of the pickled external material SS. 112b The second oblique light illuminator 112b has a length that covers half of the width direction (xx direction) of the pickled material SS so that it can be irradiated. The second oblique light illuminator 112b partially overlaps with the first oblique light illuminator 112a in the width direction (xx direction) of the pickled material SS. The second oblique light illuminator 112b is a rod-shaped green LED light. The combined length of the width direction (xx direction) of the first oblique light illuminator 112a and the width direction (xx direction) of the second oblique light illuminator 112b is longer than the width in the width direction (xx direction) of the pickled material SS.

[0090] Furthermore, the first oblique light illuminator 112a and the second oblique light illuminator 112b are installed separately in the longitudinal direction (yy direction) of the steel plate so as not to interfere with each other.

[0091] Specifically, as shown in Figure 14, the first oblique light illuminator 112a is installed on the upper side (upstream side in the conveying direction) in the longitudinal direction (yy direction) of the horizontal plane HS through which the central axis of the conveying roll 101 passes. The second oblique light illuminator 112b is installed on the lower side (downstream side in the conveying direction) in the longitudinal direction (yy direction) of the horizontal plane HS through which the central axis of the conveying roll 101 passes. As shown in Figure 14, the light L from the first oblique light illuminator 112a... 112a The horizontal line HL on the surface SSa is the intersection line of the horizontal plane HS and the surface SSa of the pickled outer coating SS, and the light is directed at a predetermined angle (light L 112a (The angle between the horizontal plane HS and the horizontal plane)θ 112a It is installed so as to be tilted. Also, as shown in Figure 14, the second oblique light illumination 112b receives light L from the second oblique light illumination 112b. 112b The horizontal line HL on the surface SSa is the intersection line of the horizontal plane HS and the surface SSa of the pickled outer coating SS, and the light is directed at a predetermined angle (light L 112b (The angle between the horizontal plane HS and the horizontal plane)θ 112b It is installed so as to be sloped.

[0092] Furthermore, as shown in Figure 15, the first oblique light illumination 112a and the second oblique light illumination 112b each emit light L 112a , L 112b The system is installed so that the light is irradiated from the inside in the width direction of the pickled external sales material SS toward the outside in the width direction of the pickled external sales material SS.

[0093] Specifically, as shown in Figure 15, the first oblique light illumination 112a emits light L from the first oblique light illumination 112a. 112a From the inside in the width direction to the outside in the width direction of the pickled outer material SS, towards the horizontal line HL on the surface Ssa of the pickled outer material SS, at a predetermined oblique angle (light L 112a (Angle of oblique light relative to line VL in the longitudinal direction (yy direction)) Δ 112a It is installed so that it is irradiated by [the light source].

[0094] Furthermore, as shown in Figure 15, the second oblique light illumination 112b emits light L from the second oblique light illumination 112b. 112bis installed so as to irradiate at a predetermined oblique light angle (the angle of obliquely irradiating with respect to the line VL in the longitudinal direction (y - y direction) of the light L) Δ from the inner side in the width direction of the pickled externally sold material SS toward the outer side in the width direction, toward the horizontal line HL on the surface SSa of the pickled externally sold material SS. 112b with respect to the line VL in the longitudinal direction (y - y direction) of the light L 112b is installed so as to irradiate at a predetermined oblique light angle (the angle of obliquely irradiating with respect to the line VL in the longitudinal direction (y - y direction) of the light L) Δ from the inner side in the width direction of the pickled externally sold material SS toward the outer side in the width direction, toward the horizontal line HL on the surface SSa of the pickled externally sold material SS. As the first oblique light illumination 112a, as described above, red LED illumination is adopted, and as the second oblique light illumination 112b, green LED illumination is adopted, and the wavelength of the light L 112a from the first oblique light illumination 112a and the wavelength of the light L 112b from the second oblique light illumination 112b are made different.

[0095] Thus, the reason for adopting the oblique light illumination 112 is to emphasize the unevenness of the scale defect by diffuse reflection and to highly discriminatively determine the scale defect (for example, rough surface D1, pitted surface D2, scale peeling D3). By adopting the oblique light illumination 112, even when it is difficult to detect the scale defect by specular reflection, the unevenness is emphasized by the oblique light illumination 112 to detect the scale defect, and the detection accuracy of the scale defect can be improved. On the other hand, by adopting the specular reflection illumination 111, even when it is difficult to detect the scale defect by diffuse reflection, there may be cases where the specular property is emphasized by the specular reflection illumination 111 to detect the scale defect.

[0096] Also, the reason for dividing the oblique light illumination 112 into two in the width direction (x - x direction) of the pickled externally sold material SS and constituting it with the first oblique light illumination 112a and the second oblique light illumination 112b is that when it is constituted by a single oblique light illumination 112, since it is necessary to irradiate the entire width in the width direction of the pickled externally sold material SS, an installation space is required by the amount that the oblique light illumination 112 protrudes from the end in the width direction (x - x direction) of the pickled externally sold material SS. Therefore, it is to eliminate this extra protruding installation space.

[0097] Note that it is not always necessary to divide the oblique light illumination 112 into two in the width direction (x - x direction) of the pickled externally sold material SS, and it may be constituted alone.

[0098] In addition, the imaging device 120 includes a plurality (four in this embodiment) of front-reflection illumination imaging devices 121, and a plurality (two in this embodiment) of first oblique illumination imaging devices 122a and a plurality (two in this embodiment) of second oblique illumination imaging devices 122b as oblique illumination imaging devices.

[0099] The plurality of front-reflection illumination imaging devices 121 image the surface SSa of the pickled and externally sold material SS irradiated with the front-reflected light L by the front-reflection illumination 111. As shown in FIG. 15, they are arranged at predetermined intervals along the width direction (x-x direction) of the pickled and externally sold material SS over the entire width of the surface SSa of the pickled and externally sold material SS. 11 Each front-reflection illumination imaging device 122 includes a first filter 123 that transmits the wavelength (blue) of the light L that is irradiated from the front-reflection illumination 111 and front-reflected by the surface SSa of the pickled and externally sold material SS. 111

[0100] And as shown in FIG. 14, each front-reflection illumination imaging device 121 is installed such that the camera angle θ formed by the central axis CL of each front-reflection illumination imaging device 121 and the horizontal plane HS 121 is of a predetermined magnitude. 121

[0101]

[0102] The reason why the imaging device 120 includes the front-reflection illumination imaging device 121 is that the front-reflection imaging images (for example, the front-reflection imaging image G1 of rough skin (see FIG. 18(a)), the front-reflection imaging image G10 of surface roughness (see FIG. 19(a)), the front-reflection imaging image G100 of rough skin (see FIG. 20(a))) of the surface SSa of the pickled and externally sold material SS irradiated with the front-reflected light L by the front-reflection illumination 111 are imaging images captured with the specularity emphasized, and thus scale defects can be detected with high accuracy. 111

[0103] In addition, the plurality of first oblique illumination imaging devices 122a image the diffusely reflected light L by the first oblique illumination 112a of the oblique illumination 112 112aThe system images the surface Ssa of the pickled externally sold material SS that has been irradiated with light, and is arranged at predetermined intervals along the width direction (xx direction) on the side where the first oblique light illumination 112a is installed in the width direction (xx direction) of the surface Ssa of the pickled externally sold material SS. Each first oblique light illumination imaging device 122a captures the light L irradiated from the first oblique light illumination 112a and diffusely reflected from the surface Ssa of the pickled externally sold material SS. 112a It is equipped with a second filter 124a that transmits the wavelength (red).

[0103] Furthermore, the multiple second oblique light illumination imaging devices 122b receive diffusely reflected light L from the second oblique light illumination 112b of the oblique light illumination 112. 112b The system images the surface Ssa of the pickled external material SS that has been irradiated with light, and is arranged at predetermined intervals along the width direction (xx direction) on the side where the second oblique light illumination 112b is installed in the width direction (xx direction) of the surface Ssa of the pickled external material SS. Each second oblique light illumination imaging device 122b captures the light L irradiated from the second oblique light illumination 112b and diffusely reflected from the surface Ssa of the pickled external material SS. 112b It is equipped with a third filter 124b that transmits the wavelength (green).

[0104] Then, as shown in Figure 14, each first oblique light illumination imaging device 122a has a central axis CL 122a The camera is positioned so that the angle between it and the horizontal plane HS is 0°.

[0105] Furthermore, as shown in Figure 14, each second oblique light illumination imaging device 122b has a central axis CL 122b The camera is positioned so that the angle between it and the horizontal plane HS is 0°.

[0106] Thus, the reason why the imaging device 120 is equipped with multiple first oblique illumination imaging devices 122a and multiple second oblique illumination imaging devices 122b as imaging devices for oblique illumination is that the diffusely reflected light L is illuminated by the first oblique illumination 112a. 112a Light L diffusely reflected by the surface SSa of the pickled outer material SS irradiated with light and the second oblique light illumination 112b 112bDiffuse reflection images of the surface SSa of pickled external material SS irradiated with light (for example, diffuse reflection image G2 of rough surface (see Figure 18(b)), diffuse reflection image G20 of rough surface (see Figure 19(b)), diffuse reflection image G200 of scale peeling (see Figure 20(b))) are images that emphasize the irregularities of scale defects, thereby enabling highly accurate detection of scale defects. By employing diffuse reflection images, even when it is difficult to detect scale defects using specular reflection images, the irregularities can be emphasized by the diffuse reflection images to detect the scale defects, thereby improving the accuracy of scale defect detection. On the other hand, by employing specular reflection images, even when it is difficult to detect scale defects using diffuse reflection images, the specular surface can be emphasized by the specular reflection images to detect the scale defects.

[0107] Furthermore, the image processing device 130 processes specular reflection images captured by multiple specular reflection imaging devices 121 and diffuse reflection images captured by multiple first oblique illumination imaging devices 122a and multiple second oblique illumination imaging devices 122b, which serve as oblique illumination imaging devices, to determine the type of scale defect (for example, rough surface D1, surface roughness D2, scale peeling D3) on the surface SSa of the pickled external material SS. The image processing device 130 is connected to multiple specular reflection imaging devices 121, multiple first oblique illumination imaging devices 122a and multiple second oblique illumination imaging devices 122b.

[0108] As shown in Figure 16, the image processing device 130 includes an image acquisition unit 131, a defect type determination unit 132, and an output unit 133. The image processing apparatus 130 is a computer system having computational processing capabilities to realize the functions of the image acquisition unit 131, the defect type determination unit 132, and the output unit 133 by executing programs on computer software. This computer system is configured with ROM, RAM, CPU, etc., and realizes the aforementioned functions on software by executing various dedicated programs pre-stored in the ROM, etc.

[0109] The functions of the image acquisition unit 131, the defect type determination unit 132, and the output unit 133, which constitute the image processing apparatus 130, and their processing flow will be explained with reference to Figure 7. First, in step S101, the image acquisition unit 131 of the image processing device 130 acquires multiple specular reflection images captured by multiple specular reflection illumination imaging devices 121 (for example, specular reflection image G1 of rough skin (see Figure 18(a)), specular reflection image G10 of rough surface (see Figure 19(a)), and specular reflection image G100 of rough skin (see Figure 20(a))). The image acquisition unit 131 also acquires multiple diffuse reflection images captured by multiple first oblique light illumination imaging devices 122a and multiple second oblique light illumination imaging devices 122b (for example, diffuse reflection image G2 of rough skin (see Figure 18(b)), diffuse reflection image G20 of rough surface (see Figure 19(b)), and diffuse reflection image G200 of scale peeling (see Figure 20(b))).

[0110] Next, in step S102, the defect type determination unit 132 inputs each of the multiple specular reflection images and multiple diffuse reflection images acquired by the image acquisition unit 131 to be determined into the scale defect determination model and determines the type of scale defect (for example, rough surface D1, rough surface D2, scale peeling D3).

[0111] Here, the scale defect determination model is obtained by machine learning multiple training datasets, where past specular reflection images from the specular reflection imaging device 121 and past diffuse reflection images from the first oblique illumination imaging device 122a and the second oblique illumination imaging device 122b are used as input data, and the past determination results of the scale defect type for this input data are used as output data. The machine learning method is a neural network, and the scale defect determination model is a determination model constructed using a neural network.

[0112] Information from this scale defect determination model is input to the defect type determination unit 132 from the input device 134 connected to the image processing device 130.

[0113] Here, the defect type determination unit 132 inputs specular reflection images (for example, specular reflection image G1 of rough surface (see Figure 18(a)), specular reflection image G10 of rough surface (see Figure 19(a)), specular reflection image G100 of rough surface (see Figure 20(a))) and diffuse reflection images (for example, diffuse reflection image G2 of rough surface (see Figure 18(b)), diffuse reflection image G20 of rough surface (see Figure 19(b)), diffuse reflection image G200 of scale peeling (see Figure 20(b)))) that cover a length of 1m to 2m in the longitudinal direction and the entire width of the pickled external material SS into the scale defect determination model to determine the type of scale defect. The defect type determination unit 132 then performs this determination process over the entire length of the pickled external material SS.

[0114] The defect type determination unit 132 then sends the determination result of the scale defect type to the output unit 133. In this case, if there are no defects in either the specular reflection image or the diffuse reflection image, the output unit 133 receives a result indicating that the scale defect does not fall under any of the scale defect types.

[0115] Finally, in step S103, the output unit 133 outputs the determination result from the defect type determination unit 132 to the surface quality acquisition unit 23 via the output device 135.

[0116] Specifically, the output unit 133 determines the result that does not fall under any of the scale defects as "no defects," and outputs this "no defects" determination result and the determination result of the scale defect type (for example, rough surface D1, rough surface D2, scale peeling D3) determined by the defect type determination unit 132 to the surface quality acquisition unit 23 via the output device 135. In this process, the output unit 133 outputs the results of determining the type of scale defect (e.g., rough surface D1, surface roughness D2, scale peeling D3) based on specular reflection and diffuse reflection images that cover a length of 1m to 2m in the longitudinal direction of the pickled external material SS and the entire width in the width direction of the pickled external material SS, to the surface quality acquisition unit 23 via the output device 135 as surface quality data for each of the multiple performance collection unit lengths L2 divided along the longitudinal direction of the pickled external material SS.

[0117] Thus, the surface inspection device 100 uses specularly reflected light L on the surface SSa of the pickled outer material SS. 111 A specular reflectance illumination 111 that irradiates light L, and light L illuminated by the specular reflectance illumination 111. 111 The system includes a specular reflection illumination imaging device 121 that images the surface SSa of the pickled external sales material SS that has been irradiated with light.

[0118] Furthermore, the surface inspection device 100 uses diffusely reflected light L on the surface SSa of the pickled outer material SS. 112a ,L 112b The oblique light illumination 112 emits light L by first oblique light illumination 112a and second oblique light illumination 112b of the oblique light illumination 112, and the first oblique light illumination 112a and second oblique light illumination 112b of the oblique light illumination 112 emits light L 112a ,L 112b The system includes a first oblique light illumination imaging device 122a and a second oblique light illumination imaging device 122b, which are oblique light illumination imaging devices that image the surface SSa of the pickled external sales material SS that has been irradiated with light.

[0119] Furthermore, the surface inspection apparatus 100 includes an image processing device 130 that processes specular reflection images G1, G10, G100 taken by the specular reflection illumination imaging device 121 and diffuse reflection images G2, G20, G200 taken by the first oblique light illumination imaging device 122a and the second oblique light illumination imaging device 122b, respectively, to determine the type of scale defect (for example, rough surface D1, surface roughness D2, scale peeling D3) on the surface SSa of the pickled external material SS.

[0120] In this surface inspection apparatus 100, the type of scale defect determined by the image processing apparatus 130 is used as the actual surface quality data, which is the inspection result in inspection process P3. As a result, the surface quality prediction device 10 equipped with the surface inspection device 100 can accurately determine the type of scale defect on the surface SSa of the pickled outer material SS, and consequently, the actual surface quality data obtained from the inspection process P3 can be made highly accurate.

[0121] Furthermore, according to the surface quality prediction device 10 for hot-rolled steel strip equipped with this surface inspection device 100, the image processing device 130 determines the type of scale defect by taking past specular reflection images from the specular reflection imaging device 121 and past diffuse reflection images from the first oblique light illumination imaging device 122a and the second oblique light illumination imaging device 122b as input data, and output data the past determination results of the type of scale defect for this input data, as input to a scale defect determination model generated by machine learning. The specular reflection images G1, G10, G100 from the specular reflection imaging device 121 and the diffuse reflection images G2, G20, G200 from the first oblique light illumination imaging device 122a and the second oblique light illumination imaging device 122b as input data.

[0122] This allows for more accurate determination of the type of scale defect on the surface SSa of the pickled external material SS.

[0123] Furthermore, according to the surface quality prediction device 10 for hot-rolled steel strip equipped with a surface inspection device 100, the image processing device 130 inputs specular reflection images G1, G10, G100 and diffuse reflection images G2, G20, G200, each covering a length of 1m to 2m in the longitudinal direction and the entire width of the pickled steel strip SS into a scale defect determination model to determine the type of scale defect. This determination process is then performed over the entire length of the pickled steel strip SS.

[0124] This allows for more accurate determination of the type of scale defects on the surface SSa of the pickled external material SS. Furthermore, by inputting specular reflection images G1, G10, G100 and diffuse reflection images G2, G20, G200, each covering a longitudinal length of 1m to 2m and the entire width of the pickled externally sold material SS into a scale defect determination model to determine the type of scale defect, the yield loss due to surface roughness D1 on the surface SSa of the pickled externally sold material SS was reduced from 1.0% in the conventional method to 0.2%.

[0125] Conventional methods do not use the image processing device 130, which is characterized by the present invention, to determine the type of scale defect on the surface Sa of the pickled externally sold material SS. Instead, the type of scale defect on the surface Sa of the pickled externally sold material SS is determined by visual inspection by an inspector, and the results are reflected in the manufacturing conditions.

[0126] Furthermore, according to the surface quality prediction device 10 equipped with a surface inspection device 100, the types of scale defects are rough surface D1, surface roughness D2, and scale peeling D3. The image processing device 130 inputs specular reflection images G1, G10, G100 and diffuse reflection images G2, G20, G200, each covering a length of 1m to 2m in the longitudinal direction and the entire width of the pickled outer material SS into a scale defect determination model to determine rough surface D1, surface roughness D2, and scale peeling D3 on the surface SSa of the pickled outer material SS, and performs this determination process over the entire length of the pickled outer material SS.

[0127] This allows for highly accurate detection of surface roughness D1, surface roughness D2, and scale delamination D3 on the surface SSa of the pickled external material SS, achieving a recognition rate of approximately 86% for surface roughness D1, approximately 93% for surface roughness D2, and approximately 99% for scale delamination D3.

[0128] Furthermore, according to the surface quality prediction device 10 equipped with a surface inspection device 100, the image processing device 130 outputs the determination results of the type of scale defect (e.g., rough surface D1, surface roughness D2, scale peeling D3) based on specular reflection and diffuse reflection images that cover a length of 1m to 2m in the longitudinal direction and the entire width in the width direction of the pickled outer sales material SS to the surface quality acquisition unit 23 as surface quality data for each of the multiple actual data collection unit lengths L2 divided along the longitudinal direction of the pickled outer sales material SS.

[0129] As a result, the surface quality acquisition unit 23 receives the results of determining the type of scale defects from specular reflection and diffuse reflection images, which cover a length of 1m to 2m in the longitudinal direction of the pickled externally sold material SS and the entire width in the width direction of the pickled externally sold material SS, as surface quality data for each of the multiple actual data collection unit lengths L2 divided along the longitudinal direction of the pickled externally sold material SS. As a result, even if the surface quality or finishing rolling conditions change in the longitudinal direction of the pickled externally sold material SS, the surface quality of the pickled externally sold material SS can be predicted with high accuracy.

[0130] Although embodiments of the present invention have been described above, the present invention is not limited thereto and can be modified and improved in various ways. For example, the training data creation unit 30 classifies the training data created based on the finishing rolling conditions according to predetermined rules into multiple training data sets (training data classification 1, ..., training data classification c, ..., training data classification M) and saves them. However, the created training data may also be saved as is without classification.

[0131] Furthermore, the quality prediction model generation unit 40 analyzes each of the multiple training data (training data classification 1, ..., training data classification c, ..., training data classification M) stored in the training data creation unit 30 using machine learning to generate surface quality prediction models F1, ..., F c , , , F MAlthough it is designed to generate and save the data, if the training data created by the training data creation unit 30 is saved as is without classification, the training data saved by the training data creation unit 30 will be analyzed using machine learning to create a surface quality prediction model F. c You can also generate and save it.

[0132] Furthermore, the surface quality prediction unit 50 generates surface quality prediction models F1, ..., F for each classification of the multiple training data (training data classification 1, ..., training data classification c, ..., training data classification M) stored in the quality prediction model generation unit 40. c , , , F M Select a surface quality prediction model to use, input the planned or actual finishing rolling conditions for the newly manufactured hot-rolled steel strip S into the selected surface quality prediction model, and then generate predictions for multiple predicted locations t1~t along the longitudinal direction of the newly manufactured hot-rolled steel strip S. n The system is designed to predict surface quality. However, the quality prediction model generation unit 40 analyzes the training data saved by the training data creation unit 30 using machine learning, and generates the surface quality prediction model F c When a surface quality prediction model F is generated and saved, the surface quality prediction unit 50 uses the surface quality prediction model F saved by the quality prediction model generation unit 40. c The system inputs the planned or actual finishing rolling conditions for the newly manufactured hot-rolled steel strip S, and then generates predictions for multiple sections t1~t along the longitudinal direction of the newly manufactured hot-rolled steel strip S. n You may also perform surface quality predictions.

[0133] In this way, the training data creation unit 30 creates paired data by linking the actual data of finishing rolling conditions saved in the finishing rolling condition record storage unit 22 with the actual data of surface quality saved in the surface quality record storage unit 24, and by creating this paired data for multiple hot-rolled steel strips S to be manufactured, it creates and saves training data, which is a collection of paired data. Furthermore, the quality prediction model generation unit 40 analyzes the training data saved in the training data creation unit 30 using machine learning and generates a surface quality prediction model F cIt generates and saves the surface quality prediction model F saved by the quality prediction model generation unit 40. c The system inputs the planned or actual finishing rolling conditions for the newly manufactured hot-rolled steel strip S, and then generates predictions for multiple sections t1~t along the longitudinal direction of the newly manufactured hot-rolled steel strip S. n To predict surface quality.

[0134] This makes it possible to predict the surface quality of the hot-rolled steel strip S with high accuracy, even if the surface quality or finishing rolling conditions change along the longitudinal direction of the hot-rolled steel strip S.

[0135] Furthermore, in the prediction process of the surface quality prediction processing unit 51, the surface quality prediction processing unit 51 divides the newly manufactured hot-rolled steel strip S into multiple prediction locations t1~t n The probability of defect occurrence p in each of the above is calculated (step S232), and the multiple predicted locations t1~t are calculated. n For each of the predicted locations t1~t, the probability of defect occurrence p is greater than a predetermined threshold. n The presence or absence of [unclear] is determined (step S233) and the surface quality is predicted.

[0136] However, in the prediction process in the surface quality prediction processing unit 51, multiple prediction points t1~t are divided along the longitudinal direction of the newly manufactured hot-rolled steel strip S. n The probability of defect occurrence p for each of these can be calculated, and the surface quality can be predicted solely from these calculation results.

[0137] Furthermore, in this embodiment, surface quality is defined as information regarding the presence or absence of scale defects. However, surface quality may also include information such as the size and degree of scale defects if present, and the location of scale defects in the width direction of the surface of the hot-rolled steel strip S.

[0138] Furthermore, in this embodiment, the hot-rolled steel strip S is manufactured through a hot-rolling process P1 and a pickling process P2, and is a pickled material that undergoes surface inspection in an inspection process P3 after the pickling process P2. However, it may also be a hot-rolled steel strip manufactured through a hot-rolling process P1 without a pickling process P2, and that undergoes surface inspection in an inspection process P3 after the hot-rolling process P1. Furthermore, the type of scale defect determined by the surface inspection device 100 may be a concave scale defect occurring on the surface SSa of the pickled external sales material SS, other than surface roughness D1, surface roughness D2, and scale peeling D3.

[0139] Furthermore, when determining the types of scale defects, for example, skin roughness D1, surface roughness D2, and scale delamination D3, it is also possible to further subdivide each of skin roughness D1, surface roughness D2, and scale delamination D3 to determine the types of defects.

[0140] Furthermore, the surface inspection device 100 is equipped with two oblique light illuminations 112, a first oblique light illumination 112a and a second oblique light illumination 112b. However, the oblique light illumination 112 may consist of one or more illuminations that irradiate the surface SSa of the pickled outer material SS with diffusely reflected light.

[0141] Furthermore, as imaging devices for oblique light illumination, there are two types, the first oblique light illumination imaging device 122a and the second oblique light illumination imaging device 122b, corresponding to the first oblique light illumination 112a and the second oblique light illumination 112b that constitute the oblique light illumination 112. However, if the oblique light illumination 112 is composed of one light, one type may be used, and if the oblique light illumination 112 is composed of three or more lights, three or more types may be used.

[0142] Furthermore, although multiple specular reflection illumination imaging devices 121, first oblique light illumination imaging devices 122a, and second oblique light illumination imaging devices 122b are installed, there are no limit to the number of each; one of each may be installed. [Explanation of Symbols]

[0143] 1. Hot rolling equipment 2 Roughing mill 3. Finishing Rolling Mill 4. Finishing Rolling Control Device 5 Upper level calculator 10. Surface quality prediction device for hot-rolled steel strip 20 Performance Data Storage Section 21. Finishing Rolling Condition Acquisition Section 22. Record Storage Section for Finish Rolling Conditions 23 Surface quality acquisition section 24 Surface Quality Records Storage Department 30 Training Data Creation Department 31 Training data creation processing unit 32 Training data storage unit 40 Quality Prediction Model Generation Unit 41. Quality Prediction Model Generation Processing Unit 42 Quality Prediction Model Storage Unit 50 Surface quality prediction section 51 Surface Quality Prediction Processing Unit 60 Prediction result presentation device 61 Display device 62 speakers 100 Surface inspection device 101 Conveyor Roll 110 Lighting device 111 Specular Reflection Illumination 112 Oblique lighting 112a 1st oblique illumination 112b Second oblique illumination 120 Imaging device 121 Imaging device for specular reflection illumination 122a First oblique light illumination imaging device (oblique light illumination imaging device) 122b Second oblique light illumination imaging device (oblique light illumination imaging device) 123 First filter 123 124a Second filter 124b Third filter 130 Image Processing Device 131 Image acquisition unit 132 Defect type determination unit 133 Output section 134 Input device 135 Output device B Seat Bar D Defect D1 Skin irritation D2 Surface Roughness D3 Scale Detachment G1 Frontal reflection imaging of skin irritation G2 Diffuse reflection imaging of skin irritation G10 Frontal reflection image of a rough surface G20 Diffuse Reflectance Imaging of Rough Surface Front reflection image of G100 scale peeling G200 scale peeling diffuse reflectance imaging P1 Hot rolling process PP1 Heating process PP2 rough rolling process PP3 Finishing Rolling Process PP4 cooling process PP5 Winding process P2 Pickling process P3 Inspection Process S hot rolled steel strip Sa tip Sb tail end SS pickled external sales material SSa surface t1~t n Predicted location

Claims

1. A surface quality prediction device for hot-rolled steel strip manufactured through a hot-rolling process and subjected to surface inspection in an inspection process following the hot-rolling process, A performance data storage unit having, for each hot-rolled steel strip manufactured, a performance data storage unit having, a performance data storage unit having, a performance data storage unit having, a performance data storage unit having, a performance data storage unit having, a performance data of surface quality, which is the inspection result in the inspection process, for each hot-rolled steel strip manufactured, which is, a performance data of surface quality, which is the inspection result in the inspection process, which is, a performance data of surface quality, which is, a performance data of surface quality, which is, a performance data of surface quality, which is, a performance data of surface quality, which is, a performance data of surface quality, which is, a performance data of surface quality, which is, a performance data storage unit having A training data creation unit creates paired data by linking the actual data of the finishing rolling conditions stored in the actual finishing rolling conditions storage unit and the actual data of the surface quality stored in the actual surface quality storage unit of the actual data storage unit, and creates and stores training data, which is a collection of paired data, by performing this paired data creation for multiple hot-rolled steel strips to be manufactured. A quality prediction model generation unit analyzes the training data stored in the training data creation unit using machine learning, generates a surface quality prediction model, and saves it. A surface quality prediction device for hot-rolled steel strips, characterized by comprising: a surface quality prediction unit that inputs the planned or actual finishing rolling conditions for a newly manufactured hot-rolled steel strip into the surface quality prediction model stored in the quality prediction model generation unit, and performs surface quality predictions for multiple prediction locations divided along the longitudinal direction of the newly manufactured hot-rolled steel strip.

2. The surface quality prediction device for a hot-rolled steel strip according to claim 1, characterized in that, in the finishing rolling condition acquisition unit, for each hot-rolled steel strip to be manufactured, actual data of the finishing rolling conditions is acquired in multiple actual data collection unit lengths divided along the longitudinal direction of the hot-rolled steel strip, with each actual data collection unit length determined according to a predetermined rule, and in the finishing rolling condition data storage unit, a representative value for each actual data collection unit length is calculated from the actual data of the finishing rolling conditions acquired in the finishing rolling condition acquisition unit and stored.

3. The surface quality prediction device for a hot-rolled steel strip according to claim 1, characterized in that, in the surface quality acquisition unit, for each hot-rolled steel strip to be manufactured, the actual surface quality data is acquired in multiple actual data collection unit lengths divided along the longitudinal direction of the hot-rolled steel strip, with each actual data collection unit length determined according to a predetermined rule, and in the surface quality data storage unit, a representative value for each actual data collection unit length is calculated from the actual surface quality data acquired in the surface quality acquisition unit and stored.

4. The hot-rolled steel strip surface quality prediction device according to claim 1, characterized in that the training data creation unit selects and discards data to be used in the actual data of the finishing rolling conditions and the actual data of the surface quality when creating the paired data.

5. The hot-rolled steel strip surface quality prediction apparatus according to claim 1, characterized in that the training data creation unit classifies the training data into multiple training data sets based on finish rolling conditions according to predetermined rules and stores them, the quality prediction model generation unit generates and stores a surface quality prediction model for each classification of the training data by analyzing it using machine learning, and the surface quality prediction unit selects a surface quality prediction model to be used from the surface quality prediction models for each classification of the training data stored in the quality prediction model generation unit, inputs the planned finish rolling conditions or actual finish rolling conditions for the newly manufactured hot-rolled steel strip into the selected surface quality prediction model, and predicts the surface quality at multiple prediction locations divided along the longitudinal direction of the newly manufactured hot-rolled steel strip.

6. The surface quality prediction device for a hot-rolled steel strip according to claim 1, further comprising a prediction result presentation device that presents to the surface quality prediction user the prediction results for surface quality at multiple prediction points divided along the longitudinal direction of the newly manufactured hot-rolled steel strip, as predicted by the surface quality prediction unit.

7. The hot-rolled steel strip surface quality prediction device according to claim 6, characterized in that the prediction result presentation device comprises a display device that displays the distribution of defects in the longitudinal direction of the hot-rolled steel strip and a speaker that communicates whether or not defects have occurred.

8. The hot-rolled steel strip surface quality prediction device according to claim 1, characterized in that the hot-rolled steel strip is manufactured through a hot-rolling process and a pickling process, and is a pickled material that is surface-inspected in an inspection process after the pickling process.

9. In the inspection process described above, the inspection line for performing surface inspection of the pickled material sold as hot-rolled steel strip is equipped with a surface inspection device for performing surface inspection of the pickled material sold. The surface inspection device, Specular reflection illumination is provided, which irradiates the surface of the pickled outer material with specularly reflected light. A specular reflection illumination imaging device for imaging the surface of the pickled outer material illuminated by specular reflection illumination, Oblique lighting is provided, which irradiates the surface of the pickled outer material with diffusely reflected light, An imaging device for oblique light illumination that images the surface of the pickled material irradiated with light by the oblique light illumination, The system includes an image processing device that processes specular reflection images captured by the specular reflection illumination imaging device and diffuse reflection images captured by the oblique light illumination imaging device to determine the type of scale defects on the surface of the pickled outer material, The surface quality prediction device for hot-rolled steel strip according to claim 8, characterized in that the type of scale defect determined by the image processing device is used as actual surface quality data, which is the result of the inspection process.

10. The hot-rolled steel strip surface quality prediction device according to claim 9, characterized in that the determination of the type of scale defect by the image processing device is performed by inputting the specular reflection image taken by the specular reflection imaging device and the diffuse reflection image taken by the oblique illumination imaging device, respectively, into a scale defect determination model generated by machine learning using multiple training datasets, each of which is used as input data, and which is used as output data, to determine the type of scale defect.

11. The surface quality prediction device for hot-rolled steel strip according to claim 10, wherein the image processing device inputs the specular reflection image and the diffuse reflection image, each of which has a field of view of a length of 1 m to 2 m in the longitudinal direction and the entire width in the width direction of the pickled outer material, to the scale defect determination model to determine the type of scale defect, and performs this determination process over the entire length of the pickled outer material.

12. The surface quality prediction device for hot-rolled steel strip according to claim 11, wherein the types of scale defects are rough surface, surface roughness, and scale peeling, and the image processing device inputs the specular reflection image and the diffuse reflection image, each of which has a field of view of a length of 1 m to 2 m in the longitudinal direction of the pickled outer material and the entire width in the width direction of the pickled outer material, into the scale defect determination model to determine whether there is rough surface, surface roughness, and scale peeling on the surface of the pickled outer material, and this determination process is performed over the entire length of the pickled outer material.

13. The surface quality prediction device for hot-rolled steel strip according to claim 11, characterized in that the image processing device outputs to the surface quality acquisition unit the result of determining the type of scale defect based on the specular reflection image and the diffuse reflection image, which have a field of view of a length of 1 m to 2 m in the longitudinal direction of the pickled outer material and the entire width in the width direction of the pickled outer material, as surface quality data for each of the multiple performance collection unit lengths divided along the longitudinal direction of the pickled outer material.

14. A method for manufacturing a hot-rolled steel strip, characterized by including a surface quality prediction step of predicting the surface quality at multiple prediction points divided along the longitudinal direction of a newly manufactured hot-rolled steel strip using a hot-rolled steel strip surface quality prediction device according to any one of claims 1 to 13.

Images