Quality defect diagnostic device and quality defect diagnostic method

Abstract

To provide a quality defect diagnostic device and a quality defect diagnostic method that are capable of predicting quality defects of a product with high accuracy and estimating the causes of the quality defects.SOLUTION: A quality defect diagnostic device that diagnoses quality defects occurring in products includes: a data dividing unit that divides product data to be diagnosed composed of operation data and quality data for each of prescribed blocks in a two-dimensional manner; a data aggregation unit that extracts the operation data for each predetermined quality defect classification on the basis of a tendency of occurrence of quality defects in the blocks after the division, and aggregates the extracted operation data for each block and for each area including two or more blocks; a quality defect determination unit that determines whether or not a quality defect has occurred using a plurality of prediction models that has been created for each quality defect classification and trained to learn a relation between the aggregated operation data and the quality data; and a cause estimation unit that estimates the cause of the quality defect on the basis of the determination result for the quality defect.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a quality defect diagnosis apparatus and a quality defect diagnosis method.

Background Art

[0002] In Patent Document 1, as a method for predicting and diagnosing the quality of a product, operation condition data at the time of product manufacturing and quality determination data in the final process are associated in consideration of the quality determination position on the product, and a prediction model obtained by learning the data is used to predict the quality.

Prior Art Document

Patent Document

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, when predicting the quality of a product, if quality data is uniformly used regardless of its characteristics, information regarding multiple types of quality defects generated by different mechanisms will be missing.That is, even if a prediction model is created based on such quality data, a problem may occur in that sufficient prediction performance cannot be achieved.

[0005] The present invention has been made in view of the above, and an object thereof is to provide a quality defect diagnosis apparatus and a quality defect diagnosis method capable of accurately predicting a quality defect of a product and estimating the cause of the quality defect.

Means for Solving the Problems

[0006] To solve the above-mentioned problems and achieve the objective, the present invention provides a quality defect diagnostic device for diagnosing quality defects occurring in a product, comprising: a data division unit that divides product data to be diagnosed, consisting of operational data and quality data, into two-dimensional blocks; a data aggregation unit that extracts the operational data for each predetermined quality defect classification based on the tendency for quality defects to occur in the divided blocks, and aggregates the extracted operational data for each block and for regions containing two or more blocks; a quality defect determination unit that determines whether or not a quality defect has occurred using a plurality of predictive models created for each quality defect classification and which have learned the relationship between the aggregated operational data and the quality data; and a factor estimation unit that estimates the factors of the quality defect based on the quality defect determination result.

[0007] Furthermore, in the quality defect diagnostic device according to the present invention, when the factor estimation unit determines that a quality defect has occurred in a specific product to be diagnosed, it estimates the operational data necessary to suppress the occurrence of the quality defect.

[0008] Furthermore, in the quality defect diagnostic device according to the present invention, the factor estimation unit estimates the operational data that caused the quality defect when a quality defect occurs in a specific product.

[0009] Furthermore, in the quality defect diagnostic device according to the present invention, the factor estimation unit estimates the operational data that caused the quality defects when quality defects occur in a large number of products.

[0010] Furthermore, the quality defect diagnostic device according to the present invention is created by the predictive model which divides past product data consisting of operational data and quality data into two dimensions for each predetermined block, classifies the quality defects included in the quality data based on the tendency for quality defects to occur in the divided blocks, aggregates the operational data extracted for each classification of quality defects for each block and for areas containing two or more blocks, and learns the relationship between the aggregated operational data and the quality data for each classification of quality defects using machine learning.

[0011] Furthermore, in the quality defect diagnostic device according to the present invention, the quality defect classification unit classifies the quality defect into two or more of the following categories: tip, tail, central, width edge, single random, and group random.

[0012] Furthermore, in the quality defect diagnostic device according to the present invention, the data aggregation unit aggregates the operational data by calculating one or more of the average value, maximum value, minimum value, standard deviation, variance, and the difference between the maximum and minimum values ​​of the operational data on a block or product basis.

[0013] Furthermore, the quality defect diagnostic device according to the present invention correlates the quality data with templates created in advance for each tendency of quality defects, and classifies the quality defects based on the magnitude of the correlation.

[0014] To solve the above-mentioned problems and achieve the objective, the present invention provides a quality defect diagnosis method for diagnosing quality defects occurring in a product, comprising: a data division step in which a data division means provided by a computer divides product data to be diagnosed, consisting of operational data and quality data, into predetermined blocks in a two-dimensional manner; a data aggregation step in which a data aggregation means provided by the computer extracts the operational data for each predetermined quality defect classification based on the tendency for quality defects to occur in the divided blocks, and aggregates the extracted operational data for each block and for areas containing two or more blocks; a quality defect determination step in which a quality defect determination means provided by the computer determines whether or not a quality defect has occurred using a plurality of predictive models created for each quality defect classification and which have learned the relationship between the aggregated operational data and the quality data; and a factor estimation step in which a factor estimation means provided by the computer estimates the factors of the quality defect based on the quality defect determination result. [Effects of the Invention]

[0015] According to the quality defect diagnostic device and quality defect diagnostic method of the present invention, by creating and using a predictive model of the relationship between operational data and quality defects for each classification according to the tendency of quality defects to occur, it is possible to predict product quality defects with high accuracy and estimate the causes of such quality defects. [Brief explanation of the drawing]

[0016] [Figure 1] Figure 1 shows a schematic configuration of an information processing device that realizes a quality defect diagnostic device according to an embodiment of the present invention. [Figure 2] Figure 2 is a diagram illustrating the details of the data division step performed by the data division unit of the quality defect diagnosis device according to an embodiment of the present invention. [Figure 3] Figure 3 is a diagram illustrating the details of the quality defect classification step performed by the quality defect classification unit of the quality defect diagnostic device according to an embodiment of the present invention. [Figure 4]Figure 4 is a diagram for explaining the aggregation of operation data at the tip in the data aggregation step performed by the data aggregation unit of the quality defect diagnosis apparatus according to an embodiment of the present invention. [Figure 5] Figure 5 is a diagram for explaining the aggregation of operation data at the tail end in the data aggregation step performed by the data aggregation unit of the quality defect diagnosis apparatus according to an embodiment of the present invention. [Figure 6] Figure 6 is a diagram for explaining the aggregation of operation data in the central part in the data aggregation step performed by the data aggregation unit of the quality defect diagnosis apparatus according to an embodiment of the present invention. [Figure 7] Figure 7 is a diagram for explaining the aggregation of operation data in the width edge part in the data aggregation step performed by the data aggregation unit of the quality defect diagnosis apparatus according to an embodiment of the present invention. [Figure 8] Figure 8 is a diagram for explaining the aggregation of operation data of single-shot random and burst random in the data aggregation step performed by the data aggregation unit of the quality defect diagnosis apparatus according to an embodiment of the present invention. [Figure 9] Figure 9 is a flowchart showing the procedure of a method for creating a prediction model used in the quality defect diagnosis method according to an embodiment of the present invention. [Figure 10] Figure 10 is a flowchart showing the procedure of the quality defect diagnosis method according to an embodiment of the present invention.

Embodiments for Carrying Out the Invention

[0017] The quality defect diagnosis apparatus and the quality defect diagnosis method according to an embodiment of the present invention will be described with reference to the drawings. The quality defect diagnosis apparatus and the quality defect diagnosis method are for diagnosing quality defects occurring in products. Examples of the "products" in the present embodiment include steel products (such as coils) manufactured through a plurality of processes. Further, examples of the "quality defects" in the present embodiment include surface defects, internal defects, characteristic values of materials (such as determination results of mechanical characteristics), and the like.

[0018] 〔Information Processing Apparatus〕 FIG. 1 shows the configuration of an information processing apparatus 1 that realizes a quality defect diagnosis apparatus according to an embodiment. As shown in FIG. 1, the information processing apparatus 1 includes an input unit 10, a storage unit 20, an arithmetic unit 30, and an output unit 40.

[0019] The quality defect diagnosis apparatus according to the embodiment can be realized by the components of the information processing apparatus 1 excluding the quality defect classification unit 32 and the model creation unit 34 of the arithmetic unit 30. Further, the apparatus for creating the prediction model used in the quality defect diagnosis apparatus according to the embodiment can be realized by the components of the information processing apparatus 1 excluding the quality defect determination unit 35 and the factor estimation unit 36 of the arithmetic unit 30. Hereinafter, each component of the information processing apparatus 1 will be described.

[0020] The input unit 10 is an input means for the arithmetic unit 30, and is realized by an input device such as a keyboard, a mouse pointer, a numeric keypad, etc. The input unit 10 inputs information necessary for various processes in the arithmetic unit 30.

[0021] The storage unit 20 is composed of recording media such as an EPROM (Erasable Programmable ROM), a hard disk drive (HDD), and a removable medium. Examples of the removable medium include disk recording media such as a USB (Universal Serial Bus) memory, a CD (Compact Disc), a DVD (Digital Versatile Disc), and a BD (Blu-ray (registered trademark) Disc). The storage unit 20 can store an operating system (OS), various programs, various tables, various databases, etc. A product DB (database) 21 is stored in the storage unit 20.

[0022] Product DB21 stores data on products manufactured in the past. This product data includes operational data and quality data. "Operational data" is data obtained by measuring or estimating equipment operating conditions and product condition during product manufacturing. "Quality data" is data indicating the quality of the product, based on the results of measuring or estimating the quality of the product. The above product data includes "past performance" product data used when creating predictive models (see Figure 9) and "product data to be diagnosed" used when performing quality defect diagnosis (see Figure 10).

[0023] In addition to the product database 21, the storage unit 20 also stores, for example, product data associated with each block by the data partitioning unit 31, information regarding the classification of quality defects by the quality defect classification unit 32, and operational data aggregated by the data aggregation unit 33 on a block or product basis. Furthermore, the storage unit 20 may also store the quality defect judgment results by the quality defect determination unit 35 and the quality defect cause estimation results by the cause estimation unit 36.

[0024] The arithmetic unit 30 is implemented by a processor, such as a CPU (Central Processing Unit), and memory (main memory) such as RAM (Random Access Memory) and ROM (Read Only Memory).

[0025] The arithmetic unit 30 loads a program into the working area of ​​the main memory and executes it, and by controlling each component through the execution of the program, it realizes a function that matches a predetermined purpose. Through the execution of the program described above, the arithmetic unit 30 functions as a data partitioning unit 31, a quality defect classification unit 32, a data aggregation unit 33, a model creation unit 34, a quality defect determination unit 35, and a factor estimation unit 36. In Figure 1, for example, an example is shown in which the functions of each unit are realized by a single computer (arithmetic unit), but the means of realizing the functions of each unit are not particularly limited, and for example, each function of each unit may be realized by multiple computers.

[0026] The data division unit 31 divides the product data into predetermined blocks in a two-dimensional manner. For example, as shown in Figure 2, the data division unit 31 divides the total length of the product data in the longitudinal direction into p blocks and divides the total length of the product data in the width direction into q blocks. In the same figure, the leftmost block in the longitudinal direction of the product is designated as the 1st block, and the rightmost block as the qth block. Also in the same figure, the uppermost block in the width direction of the product is designated as the 1st block, and the lowermost block as the pth block. The data division unit 31 stores the product data, consisting of operational data and quality data, in the storage unit 20, associating it with each block into which the product has been divided as shown in Figure 2.

[0027] The data partitioning unit 31 partitions past product data when creating a predictive model (see Figure 9). The data partitioning unit 31 also partitions product data to be diagnosed when performing quality defect diagnosis (see Figure 10). Furthermore, it is desirable that the data partitioning method (number of partitions, area of ​​each block, etc.) be consistent when creating a predictive model and when performing quality defect diagnosis.

[0028] The number of divisions (q,p values) for the product data is not particularly limited, but it is desirable to divide it into at least three divisions (q,p values ​​> 3). Furthermore, the area of ​​each block after division may be the same for each product data, or it may differ. For example, in the classification of quality defects described later (see Figure 3), the blocks classified as width edge parts or central parts may have different areas (regions) for each product data.

[0029] The quality defect classification unit 32 classifies the quality defects included in the quality data into two or more categories based on the tendency of quality defects to occur in the divided blocks. Here, "tendency of quality defects to occur" refers to, for example, the location and distribution of quality defects in the divided blocks (see Figure 1).

[0030] Specifically, the quality defect classification unit 32 classifies quality defects in the quality data into those occurring at specific locations along the longitudinal direction of the product (e.g., the front end, tail end, middle section, width edge section), and those occurring sporadically or in clusters across the entire length and width of the product. The quality defect classification unit 32 classifies quality defects in the quality data into two or more of the following categories, as shown in Figure 2, for example: front end, tail end, middle section, width edge section, sporadically random, and in clusters random.

[0031] Specifically, the quality defect classification unit 32 classifies a quality defect as "leading edge" if it occurs in the first block from the leading edge of the product in a block division along the longitudinal direction of the product. Furthermore, the quality defect classification unit 32 classifies a quality defect as "tail end" if it occurs in the p-th block from the leading edge of the product in a block division along the longitudinal direction of the product.

[0032] Furthermore, the quality defect classification unit 32 classifies a quality defect as "central area" if it occurs in the blocks from the 2nd to p-1th from the front in the longitudinal direction of the product, and in the blocks from the 2nd to q-1th from the front in the width direction of the product. Also, if a quality defect occurs in the 1st or qth block in the width direction of the product, the quality defect classification unit 32 classifies a quality defect as "width edge area".

[0033] Furthermore, the quality defect classification unit 32 classifies a quality defect as "single random" if it contains only one quality defect within a block division in the longitudinal and width directions of the product. Also, the quality defect classification unit 32 classifies a quality defect as "swarm random" if it contains two or more quality defects within a block division in the longitudinal and width directions of the product.

[0034] Here, p is an integer such that p≧3 and q is an integer such that q≧3, and the values ​​of p and q are predetermined arbitrarily. In this case, with this method, the quality defects classified for each block may be two or more types, or a priority order (e.g., tip, tail end > width edge) may be assigned to the classification criteria to limit it to a single type.

[0035] For example, in the quality defect classification shown in Figure 3, the tip, tail, and central parts do not overlap, and the central part and the width edge part do not overlap either. On the other hand, the tip and width edge part, and the tail and width edge part may overlap. Also, single random and group random do not overlap with each other, but single random or group random can overlap with other classifications (tip, tail, central part, width edge part). The quality defect classification unit 32 stores information regarding the quality defect classification as shown in Figure 3 in the storage unit 20.

[0036] The quality defect classification unit 32 classifies past performance quality data when creating a prediction model (see Figure 9). Then, when performing quality defect diagnosis (see Figure 10), it uses this quality defect classification from when the prediction model was created to aggregate the operational data to be diagnosed.

[0037] Here, the quality defect classification unit 32 may correlate the quality data with templates created in advance for each type of quality defect occurrence trend, and classify the quality defects based on the magnitude of this correlation. In this case, templates for each type of quality defect classification (tip, tail, center, width edge, single random, group random) are prepared in advance, and the quality defect classification associated with the template with the highest correlation is adopted.

[0038] One possible template is the distribution of quality defects by category, obtained by labeling and aggregating quality defect data from numerous past products. In this case, for the product being diagnosed, the type of quality defect should be determined to be the same as the template type with the highest occurrence rate in the block containing the quality defect. By using this method, quality defects can be classified quickly for large amounts of product data.

[0039] The data aggregation unit 33 aggregates operational data on a block or product basis. First, the data aggregation unit 33 extracts operational data for each classification of quality defects (see Figure 3). The quality defect classification used here is the same as that classified by the quality defect classification unit 32 when creating the prediction model. Next, the data aggregation unit 33 aggregates the extracted operational data for each block and for each region containing two or more blocks. For example, the data aggregation unit 33 aggregates the operational data as follows.

[0040] As shown in Figure 4, for example, the data aggregation unit 33 extracts only the operational data from the first block out of p blocks in the longitudinal direction of the product for quality defects classified as "tip section," and aggregates the extracted operational data on a block-by-block basis. In this way, since the tip section is a part where quality defects are particularly likely to occur, the operational data is aggregated on a block-by-block basis.

[0041] Furthermore, as shown in Figure 5, for example, the data aggregation unit 33 extracts only the operational data from the p-th block from the beginning out of p blocks in the longitudinal direction of the product for quality defects classified as "tail end," and aggregates the extracted operational data on a block-by-block basis. In this way, since the tail end is a part where quality defects are particularly likely to occur, similar to the tip, the operational data is aggregated on a block-by-block basis.

[0042] Furthermore, the data aggregation unit 33 extracts only the operational data for blocks from the second to p-1th block from the beginning in the longitudinal direction of the product, and from the second to q-1th block from the beginning in the width direction of the product, for quality defects classified as "central part," as shown in Figure 6. The data aggregation unit 33 then aggregates the extracted operational data on a product-by-product basis. That is, in the same figure, the blocks enclosed by thick lines are considered as a single data set.

[0043] Furthermore, the data aggregation unit 33 extracts, for example, the operational data for all blocks in the first longitudinal direction in the product width direction and the operational data for all blocks in the q-th longitudinal direction in the product width direction, for quality defects classified as "width edge portion," as shown in Figure 7. The data aggregation unit 33 then aggregates the extracted operational data on a product-by-product basis. That is, in the same figure, the blocks enclosed by thick lines are considered as a single data set.

[0044] Furthermore, the data aggregation unit 33, for example as shown in Figure 8, extracts operational data for blocks corresponding to quality defects classified as "single random occurrences" and operational data for blocks where no quality defects occurred. The extracted operational data is then aggregated on a block-by-block basis.

[0045] Furthermore, the data aggregation unit 33, for example as shown in Figure 8, extracts operational data from blocks corresponding to the "swarm random" quality defects and operational data from blocks where no quality defects occurred. The extracted operational data is then aggregated on a block-by-block basis.

[0046] The data aggregation unit 33 can use basic statistics that represent the values ​​at each location within a block or product as a method of aggregation. Specifically, these include values ​​such as the mean, maximum, minimum, standard deviation, variance, and the range represented by the difference between the maximum and minimum values. In this way, the data aggregation unit 33 aggregates operational data by calculating one or more of the mean, maximum, minimum, standard deviation, variance, and the difference between the maximum and minimum values ​​of the operational data on a block-by-block or product-by-product basis. When aggregating operational data, at least one type of basic statistics is used, but multiple types of basic statistics may be used. The data aggregation unit 33 stores the operational data aggregated on a block-by-block or product-by-product basis in the storage unit 20.

[0047] The data aggregation unit 33 aggregates past operational data when creating a predictive model (see Figure 9). Furthermore, the data aggregation unit 33 aggregates operational data of the target of the quality defect diagnosis when performing a quality defect diagnosis (see Figure 10).

[0048] The model creation unit 34 creates multiple predictive models according to the classification of quality defects. The model creation unit 34 creates multiple predictive models according to the classification of quality defects by learning the relationship between aggregated operational data and quality data using machine learning for each classification of quality defects.

[0049] In other words, the model creation unit 34 constructs a predictive model that shows the relationship between operational data aggregated by block or product unit according to the classification of quality defects, and the quality defects that occurred in the corresponding block or product. The model creation unit 34 creates six types of predictive models, for example, to predict quality defects in the tip, tail, middle, width edge, single random, and swarm random parts. Various machine learning algorithms such as deep neural networks, decision trees, and support vector machines, as well as other statistical algorithms, can be used to create these predictive models.

[0050] The quality defect determination unit 35 determines whether or not a quality defect has occurred using multiple prediction models created by the model creation unit 34. The quality defect determination unit 35 determines whether or not a quality defect has occurred for each category by inputting the operational data to be diagnosed into the prediction model for each category. Here, the operational data input into the prediction model is data that has been aggregated in block units or product units after data division in the data division unit 31 and data aggregation in the data aggregation unit 33 (see Figure 10).

[0051] The quality defect determination unit 35 performs quality defect determination, for example, before or during the manufacturing of a product. When performed before manufacturing, the unit predicts quality defects in the manufactured product by inputting the pre-processing manufacturing condition settings as operational data for the product to be predicted into the prediction model and performing calculations. When performed during manufacturing, the unit predicts quality defects in the manufactured product by inputting the actual values ​​of the processed manufacturing conditions and the pre-processing manufacturing condition settings as operational data for the product to be predicted into the prediction model and performing calculations.

[0052] When the quality defect determination result from the quality defect determination unit 35 is output to the output unit 40, for example, a product plan view divided into product units or block units according to the classification of quality defects may be displayed on a screen where information about the product to be diagnosed can be viewed. Then, a circle mark may be placed on the corresponding part of the product plan view if the predicted result is no quality defects, and an X mark may be placed if the predicted result is quality defects. Alternatively, different colors may be used depending on whether or not quality defects are present. Or, only if the predicted result is quality defects, the output unit 40 may output a statement such as "The product to be diagnosed (or a specific block of the product to be diagnosed) is predicted to have quality defects."

[0053] The quality defect determination unit 35 inputs the operational data to be diagnosed for all prediction models for the tip, tail, center, width edge, single random, and cluster random defects to make a determination. However, it is also acceptable to input the operational data to be diagnosed for only some of the prediction models. For example, if the operational data to be diagnosed is for the "corner," that is, the data for the first block from the front in the longitudinal direction of the product and the first block from the front in the width direction of the product, the quality defects that occur will be limited to the tip, width edge, single random, or cluster random defects. In this case, it is sufficient to input the operational data to be diagnosed only for the prediction models that predict quality defects in the tip, width edge, single random, and cluster random defects.

[0054] The factor estimation unit 36 ​​estimates the factors causing the quality defect based on the quality defect determination result from the quality defect determination unit 35. "Factors causing the quality defect" refers to, for example, operational data that are important factors in the quality defect of the product.

[0055] The factor estimation unit 36 ​​can estimate the operational data necessary to prevent the occurrence of a quality defect in a specific product being diagnosed, for example, if it is determined that a quality defect will occur in that product. In other words, if it is predicted that a quality defect will occur in a product before or during manufacturing, the unit estimates the operational data that needs to be changed or adjusted to prevent the occurrence of this quality defect. This allows for consideration of countermeasures to prevent the occurrence of quality defects.

[0056] Furthermore, the factor estimation unit 36 ​​can estimate the operational data that caused the quality defect in a particular product when a quality defect occurs in that product. In other words, if a product has a quality defect after manufacturing, it estimates the operational data that caused that quality defect. This makes it possible to identify the cause of the quality defect.

[0057] Furthermore, the factor estimation unit 36 ​​can estimate the operational data that caused the quality defects when a large number of products have quality defects. In other words, if a large number of products have quality defects after manufacturing, it can estimate the operational data that caused those quality defects. This makes it possible to identify the cause of the quality defects.

[0058] When the factor estimation unit 36 ​​determines that a quality defect will occur in a specific product under diagnosis, and when it needs to estimate the operational data necessary to prevent the occurrence of the quality defect, it calculates an index corresponding to the contribution of each factor in the above determination to the data of the product. Then, it estimates the factors that show large values ​​in this index as important factors. In this case, it uses a general-purpose method for calculating important predictive factors that does not depend on the predictive model creation algorithm, such as the method known as "SHAP (SHapley Additive exPlanations)".

[0059] Furthermore, when a quality defect occurs in a specific product and the operational data that caused the defect is to be estimated, the factor estimation unit 36 ​​calculates an index corresponding to the contribution of each factor in the above determination to the data of the product in question. Then, it estimates the factors that show a large value for this index as important factors. In this case, a general-purpose method for calculating important predictive factors that does not depend on the prediction model creation algorithm, such as SHAP described above, is used.

[0060] Furthermore, when a large number of products experience quality defects and the operational data that caused the defects are estimated, the factor estimation unit 36 ​​calculates an index corresponding to the degree of influence of each factor in the overall prediction model. Then, it estimates the factors that show large values ​​for this index as important factors.

[0061] Furthermore, the factor estimation unit 36 ​​may use a method for calculating prediction importance factors specific to the algorithm used to create the prediction model, such as the importance based on Gini impurity in decision tree algorithms. Alternatively, it may use a general method for calculating prediction importance factors that is independent of the prediction model creation algorithm, such as "Permutation Importance".

[0062] The output unit 40 is implemented by a display device such as an LCD display or a CRT display. Based on the display signals input from the calculation unit 30, the output unit 40 displays, for example, the result of quality defect determination by the quality defect determination unit 35, the result of quality defect cause estimation by the cause estimation unit 36, etc., in the form of characters, graphics, etc.

[0063] [How to create a predictive model] A method for creating a predictive model used in the quality defect diagnosis method according to the embodiment will be explained with reference to Figure 9. The method for creating a predictive model involves a block division step (step S1), a quality defect classification step (step S2), a data aggregation step (step S3), and a model creation step (step S4).

[0064] In the block division step, the data division unit 31 divides the past performance product data into predetermined blocks in a two-dimensional manner (step S1). Subsequently, in the quality defect classification step, the quality defect classification unit 32 classifies the quality defects included in the quality data according to the tendency of quality defects to occur in the divided blocks (step S2).

[0065] Next, in the data aggregation step, the data aggregation unit 33 aggregates the operational data extracted for each quality defect classification into blocks and into regions containing two or more blocks (step S3). Subsequently, in the model creation step, the model creation unit 34 creates a predictive model for each quality defect classification by learning the relationship between the aggregated operational data and quality data using machine learning (step S4).

[0066] [Methods for diagnosing quality defects] A quality defect diagnosis method according to the embodiment will be explained with reference to Figure 10. The method for creating a predictive model involves a block division step (step S11), a data aggregation step (step S12), a quality defect determination step (step S13), and a factor estimation step (step S14).

[0067] In the following explanation, we assume that a predictive model created in advance at a different time is used in steps S1 to S4 of Figure 9. However, the predictive model may also be created at the time of quality defect diagnosis. That is, steps S1 to S4 of Figure 9 and steps S11 to S14 of Figure 10 may be performed at different times, or steps S11 to S14 of Figure 9 may be performed immediately after steps S1 to S4 of Figure 9.

[0068] In the block division step, the data division unit 31 divides the product data to be diagnosed into predetermined blocks in a two-dimensional manner (step S11). Subsequently, in the data aggregation step, the data aggregation unit 33 aggregates the operational data extracted for each predetermined quality defect classification into blocks and into areas containing two or more blocks (step S12). In step S12, for example, operational data is extracted using information on the quality defect classification created in step S2 of Figure 9.

[0069] Next, in the quality defect determination step, the quality defect determination unit 35 determines whether or not a quality defect has occurred using multiple prediction models (step S13). Subsequently, in the factor estimation step, the factor estimation unit 36 ​​estimates the factors of the quality defect based on the quality defect determination result (step S14).

[0070] In the quality defect diagnostic device and quality defect diagnostic method according to the embodiments described above, the location of the quality defect is classified, and the quality defect is diagnosed using a predictive model that incorporates the classification results. In other words, in the quality defect diagnostic device and quality defect diagnostic method according to the embodiments, a predictive model is created and used for each classification according to the tendency of quality defect occurrence, showing the relationship between operational data and quality defects. This makes it possible to predict product quality defects with high accuracy and to estimate the causes of such quality defects.

[0071] The quality defect diagnostic device and quality defect diagnostic method according to the present invention have been specifically described above with reference to embodiments and examples for carrying out the invention. However, the spirit of the present invention is not limited to these descriptions and must be interpreted broadly based on the claims. Furthermore, it goes without saying that various modifications and alterations based on these descriptions are also included in the spirit of the present invention. [Explanation of symbols]

[0072] 1. Information Processing Device 10 Input section 20 Memory section 21 Product DB 30 Arithmetic section 31 Data partitioning section 32 Quality defect classification department 33 Data Aggregation Department 34 Model Creation Department 35 Quality defect determination section 36 Factor Estimation Unit 40 Output section

Claims

1. A quality defect diagnostic device for diagnosing quality defects that occur in a product, A data division unit divides the product data to be diagnosed, consisting of operational data and quality data, into two dimensions for each predetermined block, A data aggregation unit extracts the operational data for each predetermined classification of quality defects based on the tendency of quality defects to occur in the divided blocks, and aggregates the extracted operational data for each block and for areas containing two or more blocks. A quality defect determination unit determines whether or not a quality defect has occurred using multiple predictive models created for each classification of quality defects and which have learned the relationship between aggregated operational data and the quality data. Based on the quality defect determination result, a factor estimation unit estimates operational data that may be the cause of the quality defect, Equipped with, The factor estimation unit calculates an index corresponding to the degree of contribution to the quality defect judgment result for the operational data input to the prediction model, and estimates the operational data that are the cause of the quality defect based on the magnitude of the calculated index. Quality defect diagnostic device.

2. The quality defect diagnostic device according to claim 1, wherein the factor estimation unit estimates operational data necessary to suppress the occurrence of a quality defect when it is determined that a quality defect occurs in a specific product to be diagnosed.

3. The quality defect diagnostic device according to claim 1, wherein the factor estimation unit estimates the operational data that caused the quality defect when a quality defect occurs in a specific product.

4. The quality defect diagnostic device according to claim 1, wherein the factor estimation unit estimates the operational data that caused the quality defects when quality defects occur in a large number of products.

5. The aforementioned prediction model, Past product data consisting of operational data and quality data is divided into two dimensions according to predetermined blocks. Based on the tendency for quality defects to occur in the divided blocks, the quality defects included in the quality data are classified. The operational data extracted for each classification of quality defects is aggregated for each block and for each area containing two or more blocks. For each of the aforementioned quality defect classifications, the relationship between the aggregated operational data and the aforementioned quality data was learned using machine learning and created. The quality defect diagnostic device according to claim 1.

6. The quality defect diagnostic device according to claim 1, wherein the classification of the quality defect is two or more of the following: tip, tail end, central part, width edge, single random, and swarm random.

7. The quality defect diagnostic device according to claim 1, wherein the data aggregation unit aggregates the operational data by calculating one or more of the mean, maximum, minimum, standard deviation, variance, and difference between the maximum and minimum values ​​of the operational data on a block or product basis.

8. The quality defect diagnostic device according to claim 3, which correlates the quality data with templates created in advance for each tendency of quality defects to occur, and classifies the quality defects based on the magnitude of the correlation.

9. A quality defect diagnosis method for diagnosing quality defects that occur in a product, A data partitioning means provided by a computer divides the product data to be diagnosed, consisting of operational data and quality data, into predetermined blocks in a two-dimensional manner, as a data partitioning step. The data aggregation means provided by the computer extracts the operational data for each predetermined classification of quality defects based on the tendency of quality defects to occur in the divided blocks, and aggregates the extracted operational data for each block and for each area containing two or more blocks, in a data aggregation step. The quality defect determination means provided by the computer includes a quality defect determination step in which it determines whether or not a quality defect has occurred using multiple predictive models created for each classification of the quality defect and which have learned the relationship between aggregated operational data and the quality data, The factor estimation means provided by the computer includes a factor estimation step of estimating operational data that are the cause of the quality defect based on the quality defect determination result, Includes, The factor estimation step involves calculating an index corresponding to the contribution of the operational data input to the prediction model to the result of the quality defect determination, and estimating the operational data that is the cause of the quality defect based on the magnitude of the calculated index. A quality defect diagnosis method that includes this.

Images