JP7484845B2 - Quality prediction model creation method, quality prediction method, operation condition presentation method, quality prediction model creation device, quality prediction device, and operation condition presentation device - Google Patents

Description

The present invention relates to a method for creating a quality prediction model, a quality prediction method, a method for presenting operating conditions, a device for creating a quality prediction model, a quality prediction device, and a device for presenting operating conditions.

Patent Document 1 discloses a method for predicting the quality of steel products, in which data on operating conditions during multi-process manufacturing of the product is linked to data on quality judgment in the final process, taking into account the quality judgment position on the product, and then the quality is predicted using a model trained on the data.

JP 2019-74969 A

However, in the method of Patent Document 1, the data to be learned is limited to a specific quality judgment position on the product. As a result, for example, it is not possible to learn information about the surrounding area of the quality judgment position on the product, making it difficult to consider the impact on quality that accompanies changes in operating conditions, including the surrounding information. Therefore, in the method of Patent Document 1, there is still room for improvement in the accuracy of product quality prediction.

The present invention has been made in consideration of the above, and aims to provide a quality prediction model creation method, a quality prediction method, an operating condition presentation method, a quality prediction model creation device, a quality prediction device, and an operating condition presentation device that are capable of predicting product quality with high accuracy.

In order to solve the above-mentioned problems and achieve the objective, the quality prediction model creation method of the present invention is a method for creating a quality prediction model for a product manufactured through multiple processes, and includes a data creation step of creating two-dimensional data for each specific position in a specific direction of a final product after the final process by linking one or more operating conditions in each process with the quality of the final product, and a model creation step of creating a quality prediction model by machine learning that predicts the quality for each specific position in the specific direction using the two-dimensional data as input data and a product defect occurrence probability score as output data.

In addition, in the method for creating a quality prediction model according to the present invention, in the above invention, the quality of the final product is obtained for each predetermined length, and the data creation step creates the two-dimensional data by linking one or more operating conditions in each process with the quality of the final product for each predetermined unit length in the length direction of the final product for the quality for each predetermined length.

In addition, in the method for creating a quality prediction model according to the present invention, in the above invention, the data creation step determines the order of the operating conditions in each process so as to improve the prediction accuracy of the quality prediction model, and creates the two-dimensional data by linking the determined order to the quality of the final product.

In addition, in the method for creating a quality prediction model according to the present invention, in the above invention, the model creation step creates a model using a convolutional neural network.

The method for creating a quality prediction model according to the present invention is the above invention in which the product is a steel product manufactured through a plurality of processes, and the plurality of processes include one or more of ironmaking, steelmaking, hot rolling, cold rolling, annealing, and surface treatment.

In addition, in the method for creating a quality prediction model according to the present invention, in the above invention, the operating conditions are preset operating conditions set as standard conditions in each process, or actual operating conditions measured in each process.

In addition, in the method for creating a quality prediction model according to the present invention, in the above invention, the actual operating conditions are the average, maximum or minimum values of the measured data.

In order to solve the above-mentioned problems and achieve the objective, the quality prediction method according to the present invention includes a quality prediction step of predicting a defect occurrence probability score of a product by inputting two-dimensional data that links one or more operating conditions in each process with the quality of the final product for each specific position in a specific direction of the final product after the final process into the quality prediction model created by the above-mentioned quality prediction model creation method.

In order to solve the above-mentioned problems and achieve the object, the operating condition presentation method according to the present invention includes a quality prediction step of predicting the defect occurrence probability score of the product when manufactured under the actual operating conditions and the set operating conditions by inputting the actual operating conditions up to the intermediate manufacturing process of the product and the set operating conditions set as standard conditions for the subsequent processes for each specific position in a specific direction of the final product after the final process into the quality prediction model created by the above-mentioned quality prediction model creation method, an operating condition search step of searching for set operating conditions that will make the defect occurrence probability score equal to or less than a predetermined value, and an operating condition presentation step of presenting the searched set operating conditions.

In the operating condition presentation method according to the present invention, in the above invention, the operating condition search step searches for the set operating conditions that make the defect occurrence probability score equal to or less than the predetermined value by inputting the set operating conditions, among the set operating conditions, to the quality prediction model together with the actual operating conditions while changing the set operating conditions that make the defect occurrence probability score exceed a predetermined value.

In order to solve the above-mentioned problems and achieve the objective, the quality prediction model creation device of the present invention is a quality prediction model creation device for a product manufactured through multiple processes, and includes: a data creation means that creates two-dimensional data for each specific position in a specific direction of a final product after the final process by linking one or more operating conditions in each process with the quality of the final product; and a model creation means that uses machine learning to create a prediction model that predicts the quality for each specific position in the specific direction, using the two-dimensional data as input data and a product defect occurrence probability score as output data.

In order to solve the above-mentioned problems and achieve the objective, the quality prediction device according to the present invention includes a quality prediction means for predicting a defect occurrence probability score of a product by inputting two-dimensional data that links one or more operating conditions in each process with the quality of the final product for each specific position in a specific direction of the final product after the final process into the quality prediction model created by the above-mentioned quality prediction model creation device.

In order to solve the above-mentioned problems and achieve the object, the operating condition presentation device according to the present invention includes a quality prediction means for inputting actual operating conditions up to an intermediate process in the manufacture of the product and set operating conditions set as standard conditions for the subsequent processes for each specific position in a specific direction of the final product after the final process into the quality prediction model created by the above-mentioned quality prediction model creation device, thereby predicting the defect occurrence probability score of the product when manufactured under the actual operating conditions and the set operating conditions, an operating condition search means for searching for set operating conditions that will make the defect occurrence probability score equal to or less than a predetermined value, and an operating condition presentation means for presenting the searched set operating conditions.

The quality prediction model creation method, quality prediction method, operating condition presentation method, quality prediction model creation device, quality prediction device, and operating condition presentation device of the present invention can create a quality prediction model that can predict product quality with high accuracy by learning two-dimensional data that links one or more operating conditions in each process to the quality of the final product for each specific position in a specific direction of the final product as input data.

FIG. 1 is a diagram showing a schematic configuration of an information processing device which realizes a quality prediction model creation device, a quality prediction device, and an operation condition presentation device according to an embodiment of the present invention. FIG. 2 is a diagram showing two-dimensional data created by linking operation condition data and quality data in the method for creating a quality prediction model according to the embodiment of the present invention. FIG. 3 is an explanatory diagram for explaining an overview of a method for creating a quality prediction model according to an embodiment of the present invention. FIG. 4 is a flowchart showing the steps of a method for creating a quality prediction model according to an embodiment of the present invention.

The quality prediction model creation method, quality prediction method, operating condition presentation method, quality prediction model creation device, quality prediction device, and operating condition presentation device according to the embodiments of the present invention will be described with reference to the drawings.

[Information processing device]
1 shows the configuration of an information processing device 1 that realizes a quality prediction model creation device (hereinafter referred to as a "model creation device"), a quality prediction device, and an operation condition presentation device according to an embodiment. As shown in FIG. 1, the information processing device 1 includes an input unit 10, a storage unit 20, a calculation unit 30, and a display unit 40.

The model creation device according to the embodiment is realized by the components of the information processing device 1, excluding the quality prediction unit 33, the influence calculation unit 34, and the operation condition presentation unit 35 of the calculation unit 30. The quality prediction device according to the embodiment is realized by the components of the information processing device 1, excluding the influence calculation unit 34 and the operation condition presentation unit 35 of the calculation unit 30. The operation condition presentation device according to the embodiment is realized by each of the components of the information processing device 1. Each of the components of the information processing device 1 will be described below.

The input unit 10 is an input means for the calculation unit 30, and is realized by an input device such as a keyboard, a mouse pointer, or a numeric keypad. The input unit 10 inputs information necessary for various processes in the calculation unit 30.

The memory unit 20 is composed of recording media such as an EPROM (Erasable Programmable ROM), a hard disk drive (HDD), and removable media. Examples of removable media include a universal serial bus (USB) memory, and disc recording media such as a compact disc (CD), a digital versatile disc (DVD), and a Blu-ray (registered trademark) disc (BD). The memory unit 20 can store an operating system (OS), various programs, various tables, various databases, and the like. The memory unit 20 also stores an operation database (DB) 21.

Operation DB21 stores searchable operational data for products that have been manufactured in the past. In this embodiment, the "product" refers to a steel product (e.g., coil) that is manufactured through multiple processes. For example, in the case of a steel product, the "multiple processes" mentioned above include one or more of the following processes: iron-making process, steel-making process, hot rolling process, cold rolling process, annealing process, and surface treatment process.

Examples of "operation data" include set operating conditions (set manufacturing conditions) data that are set in advance as standard conditions for each process, and actual operating conditions (actual manufacturing conditions) data measured by measuring devices in multiple processes when products were manufactured in the past.

The "actual operating condition data" may be, for example, the average, maximum, or minimum value of the data measured by a measuring device in each process. In this embodiment, the combined data of the set operating condition data and the actual operating condition data is called "operating condition data."

For example, in the case of steel products, the set operating condition data and actual operating condition data include data on product thickness, product width, product length, steel components, casting speed, etc. in the steelmaking process, data on product thickness, product width, product length, product temperature, rolling speed, etc. in the hot rolling process, data on product thickness, product width, product length, rolling speed, etc. in the cold rolling process, data on product thickness, product width, product length, temperature in the surface treatment furnace, surface treatment speed, etc. in the surface treatment process. In addition, the quality data includes data on the results of surface defect inspection at the end of the surface treatment process.

The calculation unit 30 is realized by a processor such as a CPU (Central Processing Unit) and a memory (main storage unit) such as a RAM (Random Access Memory) and a ROM (Read Only Memory).

The calculation unit 30 loads a program into the working area of the main memory and executes it, and controls each component part through the execution of the program to realize functions that meet a predetermined purpose. Through the execution of the program, the calculation unit 30 functions as a data creation unit 31, a model creation unit 32, a quality prediction unit 33, an impact calculation unit 34, and an operation condition presentation unit 35. Note that while FIG. 1 shows an example in which the functions of each part are realized by, for example, one computer, the means for realizing the functions of each part is not particularly limited, and the functions of each part may be realized by, for example, multiple computers.

The data creation unit 31 creates input data for the model by linking the operating condition data with the quality data. First, the data creation unit 31 collects from the operation DB 21 set operating condition data that is set in advance in the manufacturing facilities and manufacturing equipment during the multi-process manufacturing process, actual operating condition data measured by multiple measuring devices installed at each process, and quality data of the product actually inspected after the final process. Note that the set operating condition data, actual operating condition data, and quality data may be input from the input unit 10 as necessary.

The operation condition data X during the manufacture of a certain product has elements such as those shown in the following formula (1), for example.

However, "x _mn " in the above formula (1) indicates the mth (1≦m≦M) value among the first to Mth divisions of a predetermined unit length (e.g., 1 m) as manufacturing condition information for a specific position (area) in a specific direction of the product (e.g., the length direction of the product). Also, "x _mn " indicates the nth (1≦n≦N) value among the N types of sets arranged from the first to Nth, such as setting values or output values (i.e., set operating condition data) of a large number of manufacturing facilities or manufacturing devices that exist throughout multiple processes for manufacturing a product, or measurement values obtained by measuring the operating state (i.e., actual operating condition data).

Moreover, quality data Y of a certain product has elements such as those shown in the following formula (2).
Y = [y] ... (2)

However, the above "y" indicates the quality inspection judgment result of the final product after process K, i.e., the final process, when K processes from the first process to the Kth process are passed when manufacturing a product. Here, the quality may be a plurality of different quality evaluations. In other words, a plurality of outputs can be considered as y _i for each quality category. In addition, the quality data of the final product is acquired every predetermined length (e.g., 100 m).

The data creation unit 31 creates input data to be input when creating a quality prediction model, which will be described later. This input data is two-dimensional data that links one or more pieces of operating condition data for each process with quality data of the final product inspected after the final process for each specific position in a specific direction of the final product after the final process, and is composed of multiple two-dimensional matrices, for example, as shown in FIG. 2.

The two-dimensional data shown in FIG. 2 is composed of data D1, D2, and D3, which are two-dimensional matrices. The columns of the data D1, D2, and D3 are arranged with quality data Y and operation condition data for each process. The rows of the data D1, D2, and D3 are arranged with data relating to a specific position (specific area) in a specific direction of the product.

The aforementioned "data relating to a specific position in a specific direction of the product" may be, for example, a number or coordinates indicating a position when the final product is divided in a specific direction (e.g., the length direction). The two-dimensional data as exemplified in FIG. 2 is used as input data when creating a quality prediction model, which will be described later. The data creation unit 31 may temporarily store the created two-dimensional data in the storage unit 20.

The data D1, D2, and D3 shown in FIG. 2 may be data relating to the same product (e.g., a coil) or may be data relating to different products. When the data D1, D2, and D3 are data relating to the same product, the data D1 is, for example, data in which the operation condition data for each process up to the product positions "1 to 100" is arranged in the row direction. The data D2 is data in which the operation condition data for each process up to the product positions "101 to 200" is arranged in the row direction. The data D3 is data in which the operation condition data for each process up to the product positions "201 to 300" is arranged in the row direction.

The operation condition data in the length direction of the final product after the final process, which is used by the data creation unit 31 when creating the two-dimensional data, is collected for each predetermined unit length, and M in the above formula (1) is set to a predetermined length. As a result, two-dimensional data is created in a matrix format for each Mm in the length direction of the final product, in other words, for each M data rows. Furthermore, the above-mentioned "unit length" can be set to, for example, 1 m, and the above-mentioned "predetermined length" can be set to, for example, 100 m. In this case, for example, the length corresponding to A in Figure 2 is 1 m, and the length corresponding to B is 100 m.

The values in the "unit length" are treated as the average value of the observed data. In addition, some of the items may include items that take the maximum value or items that take the minimum value. A feature of the "unit length" is that the unit length of the steel plate manufactured in the final process has a minimum unit limit that corresponds to the data collection period in the length direction of the quality data required to create a quality prediction model. In addition, the "unit length" is preferably a unit longer than the data collection period corresponding to the unit length of the final process for the operational condition data in the length direction obtained in the intermediate process, and is set individually according to the data collection period of the target of prediction.

The "predetermined length Mm" is preferably a small value in order to specify the position within the coil as precisely as possible. On the other hand, to achieve high quality prediction performance, a minimum length is required that reveals the characteristics of fluctuations in the operating condition data. Therefore, the "predetermined length Mm" has an appropriate value according to the requirements for position specification and the characteristics of the input data, and is set individually according to the target and purpose of the prediction.

In this way, the data creation unit 31 creates two-dimensional data by linking the operating condition data and quality data for each process for each predetermined unit length (e.g., 1 m) in the length direction of the final product for the quality of each specified length (e.g., 100 m).

The two-dimensional data created by the data creation unit 31 can be considered to contain, as latent information, information on fluctuations in operating conditions for a specific direction of the product and information on the relationships between operating conditions that are nearby in the data structure. Therefore, by converting the input data for the quality prediction model into two dimensions and increasing the amount of information on operating conditions for a specific direction of the product compared to normal one-dimensional data, it becomes easier to capture patterns of operational fluctuations related to the quality of the product.

The data creation unit 31 normally creates two-dimensional data using the operating condition data sequence arranged in order from the upstream to downstream of the manufacturing process, but in order to improve the prediction accuracy of the quality prediction model described below, it may also determine the order of the operating condition data for each process and link it to the quality data of the final product in the determined order to create two-dimensional data.

When the order of multiple operating condition data strings in two-dimensional data (see Figure 2) has a relationship to the quality of a product, the order of the operating condition data strings can make it easier for the quality prediction model feature discrimination layer, which reacts to specific two-dimensional patterns and transmits information, to discern patterns of information relating to the relationships between operating condition data.

Therefore, when creating the two-dimensional data, the data creation unit 31 may search and rearrange the order of the operating condition data string, which is initially arranged in the order of the manufacturing processes, using, for example, an optimization algorithm so as to maximize the prediction accuracy, or may rearrange the order based on knowledge of the relationship between each piece of operating condition data regarding quality. In this way, by optimizing the order of the operating condition data when creating the two-dimensional data, it is possible to further improve the prediction accuracy of product quality.

The model creation unit 32 uses the two-dimensional data created by the data creation unit 31 to create a quality prediction model that predicts the quality of a product for any required conditions. This quality prediction model is composed of an input part, a feature discrimination part, a prediction judgment part, and an output part.

As shown in Figure 2 above, the input section inputs two-dimensional data having a two-dimensional matrix structure with the position coordinates of the product in a specific direction (e.g., the length direction of the product) in the row direction and each operating condition in the column direction.

In the feature discrimination section, a model is trained to capture patterns of two-dimensional features in the input data and discriminate features. This forms a feature discrimination layer that reacts to specific two-dimensional patterns and transmits information. This feature discrimination layer is formed in multiple layers, and information is transmitted from one layer to the next in sequence. Information is also transmitted from the layer connected to the input part that passes on information to the layer connected to the prediction and judgment part that receives information.

The prediction and judgment part outputs quality judgment prediction results for the input of two-dimensional data, based on the complex patterns of multiple operating conditions that affect quality.

In this embodiment, a quality prediction model is constructed using a Convolutional Neural Network (hereinafter referred to as "CNN"), which is a type of algorithm with a neural network structure. CNN is a machine learning algorithm that mimics the network structure of nerve cells (neurons) in the human brain, called the neural circuit network, and its learning mechanism. CNN is composed of an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer.

In a CNN, a filter called a kernel is applied to each partial region of the input data in the convolutional layer to convolve the data and extract features to construct a feature map (also called an "element"). There may be multiple filters, in which case multiple feature maps will be constructed. When training a CNN, the filter parameters are learned automatically.

In the pooling layer, the size of each feature map obtained by convolution is reduced. The convolutional layers and pooling layers may be configured with the same number of layers, alternating. In the fully connected layer, classification is performed to classify the combination of feature maps into specific prediction results. For example, in the case of steel products, the aforementioned "specific prediction results" include the presence or absence of quality defects such as surface defects.

In this way, by applying CNN to create a quality prediction model, it is possible to extract characteristic patterns from input data and obtain classifications that are identified with high accuracy based on those patterns as output data. In this embodiment, complex patterns of multiple operating conditions that affect quality are identified using a CNN (one-dimensional CNN) that outputs the classification probability of the occurrence or absence of quality defects for each one-dimensional input of an operating condition. In this case, a threshold for the classification probability is set so that when the classification probability of the occurrence of quality defects is less than an arbitrary value p (0≦p≦1), the data is classified as "no quality defect," and when the classification probability is equal to or greater than an arbitrary value p, the data is classified as "quality defect."

Figure 3 shows an overview of a CNN model that predicts quality for two-dimensional data input. For ease of explanation, the arrangement of data in the columns and rows in the "Input" section of the figure has been changed from that in Figure 2. That is, in this figure, specific directional positions of products are arranged in the columns, and operating condition data for each process are arranged in the rows. Also, the areas outlined in bold in the "Input" section of the figure are units of data input to the CNN, and correspond to, for example, data D1, D2, and D3 in Figure 2.

The model creation unit 32 uses the two-dimensional data created by the data creation unit 31 as input data and the defect occurrence probability score for each quality category of the product as output data to create a quality prediction model using machine learning (CNN) to predict the quality at each specific position in a specific direction. Note that the "defect occurrence probability score for each quality category" refers to the classification probability regarding the presence or absence of multiple types of defects, such as "the probability of occurrence of defect A is 70%" and "the probability of occurrence of defect B is 30%."

The quality prediction unit 33 predicts the defect occurrence probability score of the target product by inputting two-dimensional data that links one or more operational condition data in each process with the quality data of the final product for each specific position in a specific direction of the final product after the final process. That is, the quality prediction unit 33 inputs two-dimensional structure data consisting of the coordinates of a specific position in a specific direction of the product whose quality is to be predicted and multiple operational conditions, as shown in FIG. 2, to the quality prediction model created by the model creation unit 32, and outputs the defect occurrence probability score of the target product.

Then, the quality prediction unit 33 displays the defect occurrence probability score, which is the quality prediction result, on the display unit 40. In this case, the quality prediction unit 33 may display a circle mark in the case of a prediction result of no quality defect, and a cross mark in the case of a prediction result of a quality defect, on the display unit 40 where information on the target product can be viewed, or may display a statement such as "The target product is predicted to have a quality defect" only in the case of a prediction result of a quality defect.

The quality prediction unit 33 may input, as input data, the actual operating conditions up to the intermediate manufacturing process of the product and the set operating conditions set as standard conditions for the subsequent processes for each specific position in a specific direction of the final product after the final process into the quality prediction model created by the model creation unit 32, thereby predicting the defect occurrence probability score of the target product when manufactured under the actual operating conditions and the set operating conditions.

In this case, when K processes from the first process to the Kth process are carried out as processes for manufacturing a product, the quality prediction unit 33 makes a prediction using a quality prediction model using as input data the actual operating condition data from the first process 1 to the kth process k (1≦k≦K) and the set operating condition data from the k+1th process (k+1) to the Kth process K.

The influence calculation unit 34 calculates the relative influence (relative influence evaluation value) of the input data on the machine learning algorithm. Specifically, when the defect occurrence probability score predicted by the quality prediction unit 33 exceeds a predetermined value, the influence calculation unit 34 calculates the relative influence of each set operating condition on the defect occurrence probability score.

The influence calculation unit 34 applies a mathematical method to calculate the relative influence of each set operating condition to the quality prediction model that predicts the quality of the product. For example, Grad-CAM (Gradient-weighted Class Activation Mapping), Guided Grad-CAM, etc. can be used as this mathematical method. In this embodiment, Grad-CAM (Gradient-weighted Class Activation Mapping) described in Reference 1 below is used as this mathematical method.

Reference 1: Ramprasaath R. Selvaraju, et al. "Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization", 2017, IEEE International Conference on Computer Vision (ICCV)

The influence calculation unit 34 calculates the relative influence of each set operational condition on the defect occurrence probability score by the following procedure.
(1) In a trained CNN, the output of a specific classification class judgment (the defect occurrence probability score of a specific quality category) is differentiated with respect to the value of one point (one pixel) of one feature map in the final convolutional layer, and a differential coefficient that represents the rate of change in the defect occurrence probability score when the pixel value is slightly changed is calculated.
(2) A weighting factor for one feature map corresponding to a particular classification class is calculated by averaging the derivatives of all pixels.
(3) All feature maps in the final convolutional layer are multiplied by a weighting factor and averaged, resulting in a superposition that emphasizes feature maps that are important for classifying a particular class.
(4) Calculate the relative influence, which takes real values ranging from 0 to 1, through an activation function ReLU (ramp function).

Here, in the above calculation, the effect on the output of a specific classification class is evaluated in relation to the final convolutional layer, which is obtained from the input layer via multiple convolutional layers and pooling layers. Therefore, the size of the final convolutional layer is a reduced size of the input layer. Therefore, by expanding it by the amount of the reduced size, it is possible to clearly establish a correspondence with the input.

The operating condition presentation unit 35 searches for set operating conditions that will result in a defect occurrence probability score that is equal to or less than a predetermined value based on the relative impact calculated by the impact calculation unit 34, and presents the set operating conditions that have been found via the display unit 40.

The operating condition presentation unit 35 performs quality prediction by inputting the set operating conditions, together with the actual operating conditions, into the quality prediction model while changing the set operating conditions whose relative influences have been calculated by the influence calculation unit 34 and whose defect occurrence probability score exceeds a predetermined value. This searches for set operating conditions that will result in a defect occurrence probability score that is equal to or less than the predetermined value.

The operation condition presentation unit 35 searches for and presents the set operation conditions in the following procedure.
(1) From the relative influence calculated by the influence calculation unit 34, only the items of the set operation conditions from the k+1th process (k+1) to the Kth process K are extracted.
(2) The relative influence of each item extracted in (1) above is rearranged in descending order of value.
(3) The possibility of changing the settings is determined in any manner, starting from the top of the items sorted in (2) above. Note that "possibility of changing the settings" refers to whether the value of the set operating condition can be manipulated (changed). For example, a set operating condition that is intended to simply measure the state of a certain piece of equipment and cannot be manipulated is determined to be "non-changeable."
(4) When a changeable setting item is determined for the first time in (3) above, the setting item is determined as an operable setting operating condition.
(5) The operable set operating conditions determined in (4) above are changed from their current values to different values, while the other set operating conditions are left unchanged at their current values. These set operating conditions and the actual operating conditions up to process k are input into a quality prediction model, and a quality prediction result is output.
(6) The quality prediction result output in (5) above is compared with a preset quality control range.
(7) In the above (6), if the quality prediction result falls within the quality control range, the value changed in the above (5) is presented as a setting operation condition via the display unit 40.
(8) In the above (6), if the quality prediction result is not within the quality control range, return to the above (5), change the operable set operating condition value to another value, and repeat the following (6). Note that when repeating the above (5) and (6), an optimization method such as an iterative method may be used.

In the above steps (2) to (5), only one set operating condition with the greatest relative influence (and which can be changed) is changed to a different value, but multiple set operating conditions with a relative influence greater than a preset value (and which can be changed) may each be changed to a different value.

The display unit 40 is realized by a display device such as an LCD display or a CRT display. Based on the display signal input from the calculation unit 30, the display unit 40 displays, for example, the quality prediction result by the quality prediction unit 33, the calculation result of the relative impact by the impact calculation unit 34, the search result of the set operating conditions by the operating condition presentation unit 35, etc., in the form of characters, figures, etc.

[Model creation method]
A method for creating a quality prediction model according to an embodiment (hereinafter, referred to as a "model creation method") will be described with reference to Fig. 4. The model creation method includes a data collection step (step S1), a data creation step (step S2), and a model creation step (step S3).

(Data collection step)
In the data collection step, the data creation unit 31 collects operation condition data and quality data for a plurality of processes from the operation DB 21 (step S1).

(Data creation step)
In the data creation step, the data creation unit 31 creates two-dimensional data by linking the operating condition data and quality data in each process for each specific position in a specific direction of the final product (step S2).

(Model creation step)
In the model creation step, the model creation unit 32 uses the two-dimensional data created in step S2 as an input, the defect occurrence probability score of the target product as an output, and creates a quality prediction model that predicts the quality at each specific position in a specific direction, for example, by using a CNN. With the above, the processing of the model creation method is completed.

[Quality prediction method]
A quality prediction method according to an embodiment will now be described. In the quality prediction method, a quality prediction step is performed in addition to the steps of the model creation method (a data collection step, a data creation step, and a model creation step).

The data collection step, data creation step, and model creation step are the same as those in the model creation method described above, and therefore will not be described here. In addition, the data collection step, data creation step, and model creation step may be performed in advance at a different timing from the other steps.

(Quality prediction step)
In the quality prediction step, the quality prediction unit 33 predicts the defect occurrence probability score of the target product by inputting two-dimensional data in which the operation condition data and the quality data in each process are linked to each other for each specific position in a specific direction of the final product to the quality prediction model. With the above, the processing of the quality prediction method is completed.

[How to present operating conditions]
The method for proposing operating conditions according to the embodiment will be described below. In the method for proposing operating conditions, in addition to the steps of the quality prediction method (data collection step, data creation step, model creation step, and quality prediction step), an influence degree calculation step and a set operating condition search step are performed.

The data collection step, data creation step, model creation step, and quality prediction step are the same as those in the quality prediction method described above, and therefore will not be described here. In addition, the data collection step, data creation step, model creation step, and quality prediction step may be performed in advance at a different timing from the other steps.

(Influence calculation step)
In the influence calculation step, the influence calculation unit 34 judges whether the quality prediction model predicts the occurrence of a quality defect, i.e., whether a quality defect will occur under the actual operating conditions and the set operating conditions inputted to the quality prediction model. If it is judged that a quality defect will occur, the influence calculation unit 34 calculates the relative influence of each set operating condition on the defect occurrence probability score based on the trained quality prediction model.

(Step for searching for set operating conditions)
In the set operating condition search step, the operating condition presentation unit 35 predicts the defect occurrence probability score using the quality prediction model while changing the values of the set operating conditions that have a relatively large influence among the set operating conditions input to the quality prediction model. Next, the operating condition presentation unit 35 presents the set operating conditions when the defect occurrence probability score is predicted to be within the target range via the display unit 40. This completes the processing of the operating condition presentation method.

In addition, in the method for presenting operating conditions, after the set operating conditions search step, a product manufacturing step may be further performed in which a subsequent process is carried out to manufacture steel products based on the operating conditions presented in the set operating conditions search step.

〔Example〕
An example of the present invention will be described. In this example, the quality prediction method according to the embodiment was applied to data on steel products. Specifically, in the manufacture of steel sheets, surface defects occurring on the surface of a steel sheet after passing through a surface treatment process were predicted based on actual data of the product that had passed through a steelmaking process, a hot rolling process, a cold rolling process, and a surface treatment process.

In the data collection step, the following data were collected: operational condition data in the steelmaking process (product thickness, product width, product length, steel composition, casting speed, etc.); operational condition data in the hot rolling process (product thickness, product width, product length, product temperature, rolling speed, etc.); operational condition data in the cold rolling process (product thickness, product width, product length, rolling speed, etc.); operational condition data in the surface treatment process (product thickness, product width, product length, surface treatment furnace temperature, surface treatment speed, etc.); and surface defect inspection result data (quality data) of the steel plate inspected at the end of the surface treatment process. At this time, data was collected at intervals equivalent to every 1 m in the longitudinal direction of the product in the surface treatment process. In other words, the above-mentioned "unit length" was set to 1 m.

In the data creation step, the operating condition data for the steelmaking process, hot rolling process, cold rolling process, and surface treatment process, and the surface defect inspection result data at the end of the surface treatment process were linked and stored in the memory unit 20. At this time, the longitudinal area of the product data was divided into 100 m segments and stored. In other words, the "predetermined length" was set to 100 m.

In the model creation step, a quality prediction model was created that applies a two-dimensional CNN that outputs the classification probability (defect occurrence probability score) of the presence or absence of surface defects in response to two-dimensional input in which the longitudinal position of the steel plate after passing through the surface treatment process is arranged in the row direction and the process operating condition data is arranged in the column direction. In the model creation step, data equivalent to approximately 400 products was used, which amounts to 5,367 data samples when divided into 100 m segments. In this quality prediction model, a classification probability threshold was set so that if the classification probability of surface defect occurrence is less than 0.5, it is classified and predicted as "no surface defect," and if the classification probability is 0.5 or more, it is classified and predicted as "surface defect present."

In the quality prediction step, the quality prediction model was used to verify the accuracy of quality prediction for 627 input data samples, each of which was equivalent to approximately 500 random products and divided into 100m segments. As a result, as shown in Table 1 below, the accuracy rate was 74.2%.

To compare and evaluate the prediction accuracy of 2D CNN, a quality prediction model was created in which CNN was applied to one-dimensional input of a sequence of operating condition items across multiple processes every 1 m, using the same data equivalent to approximately 400 products as used for 2D CNN, and the accuracy rate of predicting the occurrence of surface defects was also examined. With one-dimensional input, the feature map is also one-dimensional, and it is not possible to extract longitudinal patterns of the product data. Using the same data equivalent to approximately 50 products as used for 2D CNN, the accuracy rate of predicting the occurrence of surface defects using 1D CNN was 70.0%.

Therefore, by taking into account information in the length direction using a two-dimensional CNN, it is possible to capture more of the characteristics of the data, thereby improving prediction performance from the same data set.

In this way, by inputting the operating condition data of the target from which the cause of surface defects, a type of quality defect, is to be extracted into the quality prediction model created by the present invention, it is now possible to predict the occurrence of surface defects before the manufacturing of the product is completed.

According to the quality prediction model creation method, quality prediction method, operating condition presentation method, quality prediction model creation device, quality prediction device, and operating condition presentation device of the embodiments described above, it is possible to create a quality prediction model that can predict product quality with high accuracy by learning two-dimensional data that links one or more operating conditions in each process to the quality of the final product for each specific position in a specific direction of the final product as input data.

In other words, in the quality prediction model creation method, quality prediction method, operating condition presentation method, quality prediction model creation device, quality prediction device, and operating condition presentation device according to the embodiments, two-dimensional data is created by associating and linking corresponding operating condition data and quality data with a specific position in a specific direction of the final product, and this two-dimensional data is used to input information on the operating conditions of each process and the specific position in a specific direction of the product, including surrounding information, all at once, to create a highly accurate quality prediction model that predicts product quality. This makes it possible to predict product quality with high accuracy from information on any operating conditions and the specific position in a specific direction of the product. Therefore, it becomes possible to determine in advance whether or not the final product completed after the final process will be of defective quality with higher accuracy than before.

The quality prediction model creation method, quality prediction method, operating condition presentation method, quality prediction model creation device, quality prediction device, and operating condition presentation device according to the present invention have been specifically described above using the form and examples for implementing the invention, but the spirit of the present invention is not limited to these descriptions and must be interpreted broadly based on the descriptions in the claims. Furthermore, it goes without saying that various changes, modifications, etc. based on these descriptions are also included in the spirit of the present invention.

Reference Signs List 1 Information processing device 10 Input unit 20 Storage unit 21 Operation DB
30 Calculation section 31 Data creation section 32 Model creation section 33 Quality prediction section 34 Influence degree calculation section 35 Operation condition presentation section 40 Display section

Claims (12)

A method for creating a quality prediction model for a product manufactured through a plurality of processes, comprising the steps of:
a data creation step of creating two-dimensional data for each specific position in a specific direction of a final product after a final process by linking one or more operation conditions in each process with the quality of the final product;
a model creation step of creating a quality prediction model by machine learning from a plurality of the two-dimensional data, the quality prediction model predicting quality at each specific position in the specific direction , the quality prediction model being input data of the two-dimensional data as one unit and the quality of the final product being output data;
Including,
The quality of the final product is acquired at each predetermined length,
the data creation step creates the two-dimensional data by linking one or more operation conditions in each process with the quality of the final product for each predetermined unit length in the length direction of the final product, for the quality for each predetermined length;
How to create a quality prediction model. 2. The method for creating a quality prediction model according to claim 1 , wherein the data creation step determines an order of operation conditions in each process by searching and rearranging the order using an optimization algorithm, or by rearranging the order based on an association between each operation condition data related to quality, and creates the two-dimensional data by linking the determined order with the quality of the final product. 3. The method for creating a quality prediction model according to claim 1 , wherein the model creating step creates the model using a convolutional neural network. The product is a steel product manufactured through a plurality of processes,
The method for creating a quality prediction model according to claim 1 , wherein the plurality of processes include one or more of ironmaking, steelmaking, hot rolling, cold rolling, annealing, and surface treatment. 5. The method for creating a quality prediction model according to claim 1 , wherein the operating conditions are preset operating conditions set as standard conditions in each process, or actual operating conditions measured in each process. 6. The method for creating a quality prediction model according to claim 5 , wherein the actual operating conditions are an average value, a maximum value or a minimum value of measured data. 7. A quality prediction method comprising: a quality prediction step of predicting product quality by inputting two -dimensional data linking one or more operating conditions in each process with the quality of the final product for each specific position in a specific direction of the final product after the final process into the quality prediction model created by the quality prediction model creation method according to any one of claims 1 to 6. a quality prediction step of predicting a quality of a product when the product is manufactured under the actual operating conditions and the set operating conditions, which are set as standard conditions for the subsequent processes, by inputting actual operating conditions up to a process intermediate in the manufacture of the product and set operating conditions for each specific position in a specific direction of the final product after the final process, into the quality prediction model created by the method for creating a quality prediction model according to any one of claims 1 to 6;
an operating condition search step of searching for set operating conditions such that the defect occurrence probability score indicating the quality is equal to or less than a predetermined value;
an operation condition presentation step of presenting the searched set operation conditions;
A method for presenting operating conditions, including: 9. The operational condition presentation method according to claim 8, wherein the operational condition search step searches for a set operational condition among the set operational conditions, the set operational condition having a defect occurrence probability score value exceeding a predetermined value, by inputting the set operational condition together with the actual operational condition to the quality prediction model while changing the set operational condition, so as to search for the set operational condition such that the defect occurrence probability score is equal to or less than the predetermined value. An apparatus for creating a quality prediction model for a product manufactured through a plurality of processes, comprising:
a data creation means for creating two-dimensional data for each specific position in a specific direction of a final product after a final process by linking one or more operating conditions in each process with the quality of the final product;
a model creation means for creating a prediction model by machine learning from a plurality of the two-dimensional data, the prediction model predicting quality at each specific position in the specific direction , the two-dimensional data being one unit of input data and the quality of the final product being output data;
Equipped with
The quality of the final product is acquired at each predetermined length,
the data creation means creates the two-dimensional data by linking one or more operation conditions in each process with the quality of the final product for each predetermined unit length in the length direction of the final product, for the quality for each predetermined length.
A device for creating quality prediction models. 11. A quality prediction device comprising a quality prediction means for predicting product quality by inputting two-dimensional data linking one or more operating conditions in each process with the quality of the final product for each specific position in a specific direction of the final product after the final process, for the quality prediction model created by the quality prediction model creation device described in claim 10. a quality prediction means for predicting the quality of a product when manufactured under actual operational conditions and set operational conditions, which are set as standard conditions for the subsequent processes, by inputting actual operational conditions up to a midway point in the manufacturing process of the product and set operational conditions, for each specific position in a specific direction of a final product after a final process, into the quality prediction model created by the quality prediction model creating device according to claim 10;
an operating condition search means for searching for a set operating condition such that the defect occurrence probability score indicating the quality is equal to or less than a predetermined value;
an operation condition presentation means for presenting the searched set operation conditions;
An operation condition presentation device comprising:

Images