Defect diagnosis device and defect diagnosis method

Abstract

The quality defect diagnosis device includes: a data division unit that divides product data of a diagnosis target into a two-dimensional shape for each predetermined block; a data collection unit that extracts work data and equipment maintenance data for each predetermined quality defect classification based on a quality defect generation tendency in the divided blocks, and collects the extracted work data and equipment maintenance data for each region including two or more blocks in a specific direction of the product; and a quality defect determination unit that determines whether or not a quality defect is generated using a plurality of prediction models that are generated for each quality defect classification and that learn a relationship between the collected work data and equipment maintenance data and quality data.

Description

Technical Field

[0001] This invention relates to a device and method for diagnosing poor quality. Background Technology

[0002] In Patent Document 1, as a method for predicting and diagnosing product quality, the following method is disclosed: considering the quality determination location on the product, the working condition data during product manufacturing is associated with the quality determination data in the final process, and a prediction model learned from the data is used to predict the quality.

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent Document 1: Japanese Patent No. 6953990 Summary of the Invention

[0006] The problem that the invention aims to solve

[0007] However, when predicting product quality, if quality data is used uniformly without relying on its characteristics, information related to the various quality defects arising from different mechanisms is missing. Therefore, even if a predictive model is generated based on such quality data, it may still result in insufficient predictive performance.

[0008] The present invention was made in view of the above circumstances, and its object is to provide a defect diagnosis device and a defect diagnosis method that can accurately predict product defects and presume the main causes of such defects.

[0009] Methods for solving problems

[0010] To address the aforementioned issues and achieve the objective, the quality defect diagnosis device of the present invention comprises: a data segmentation unit that segments product data of the diagnostic object, consisting of work data, equipment maintenance data, and quality data, into a two-dimensional shape according to each predetermined block; a data aggregation unit that extracts the work data and the equipment maintenance data according to each predetermined quality defect classification based on the tendency of quality defects to occur in the segmented blocks, and aggregates the extracted work data and the equipment maintenance data into regions of two or more blocks in a specific direction containing the product; and a quality defect determination unit that uses multiple prediction models generated according to each quality defect classification and having learned the relationship between the aggregated work data, the equipment maintenance data, and the quality data to determine whether a quality defect has occurred.

[0011] Furthermore, the quality defect diagnosis device of the present invention, based on the above invention, includes a main cause estimation unit, which estimates the main cause of the quality defect based on the quality defect determination result.

[0012] Furthermore, the defective quality diagnostic device of the present invention, based on the above invention, includes a result display unit, which causes the output unit to display at least one of the determination result in the defective quality determination unit and the main cause estimation result in the main cause estimation unit.

[0013] Furthermore, based on the above invention, the quality defect diagnosis device of the present invention generates the prediction model in the following manner: Product data representing past performance, consisting of the operation data, equipment maintenance data, and quality data, is divided into two dimensions according to each defined block; based on the tendency of quality defects to occur in the divided blocks, the quality defects contained in the quality data are classified; the operation data and equipment maintenance data extracted according to each quality defect classification are aggregated according to regions containing two or more blocks in a specific direction of the product; and according to each quality defect classification, machine learning is used to learn the relationship between the aggregated operation data, equipment maintenance data, and quality data.

[0014] Furthermore, based on the above-mentioned invention, the defective quality diagnostic device of the present invention provides that the operating data is data that varies along the entire length of the product in a specific direction.

[0015] Furthermore, based on the above invention, the defective quality diagnosis device of the present invention includes equipment maintenance data that varies throughout the entire period during which the product is processed and handled in the equipment.

[0016] Furthermore, based on the above-mentioned invention, the quality defect diagnosis device of the present invention provides that the quality data is data that varies along the entire length of the product in a specific direction.

[0017] Furthermore, based on the above-mentioned invention, the equipment maintenance data in the defective product diagnostic device of the present invention represents the equipment status and maintenance status that affect a specific location of the product through direct or indirect action on the product.

[0018] Furthermore, based on the above-mentioned invention, the quality defect diagnosis device of the present invention, wherein the data collection unit extracts the operation data and the equipment maintenance data according to each predetermined quality defect classification, based on the quality defect generation tendency of having multiple quality defect generation locations along the entire length direction of the product or the quality defect generation tendency of having multiple quality defect generation locations along the entire width direction of the product.

[0019] Furthermore, based on the above invention, the defect diagnosis device of the present invention uses multiple prediction models generated according to the classification of each defect and which have learned the relationship between the aggregated equipment maintenance data and the quality data to determine whether a defect has occurred. The main cause estimation unit estimates the main cause of the defect based on the defect determination result.

[0020] Furthermore, based on the above-described invention, the defective quality diagnostic device of the present invention allows the output unit to display data groups that have undergone alignment processing by the data collection unit.

[0021] To address the aforementioned issues and achieve the objectives, the present invention provides a method for diagnosing quality defects, comprising: a data segmentation step, wherein a computer-equipped data segmentation unit segments product data of the diagnostic object, consisting of work data, equipment maintenance data, and quality data, into a two-dimensional structure according to each defined block; a data aggregation step, wherein the computer-equipped data aggregation unit extracts the work data and the equipment maintenance data according to each predetermined category of quality defects based on the tendency of quality defects to occur in the segmented blocks, and aggregates the extracted work data and the equipment maintenance data according to regions of two or more blocks in a specific direction containing the product; and a quality defect determination step, wherein the computer-equipped quality defect determination unit uses multiple prediction models generated according to each category of quality defects and having learned the relationship between the aggregated work data, the equipment maintenance data, and the quality data to determine whether a quality defect has occurred.

[0022] Furthermore, the method for diagnosing quality defects of the present invention, based on the above invention, includes a main cause estimation step, wherein the main cause estimation unit of the computer estimates the main cause of the quality defect based on the determination result of the quality defect.

[0023] Invention Effects

[0024] According to the defect diagnosis device and method of the present invention, for each category corresponding to the tendency of defect generation, the relationship between work data and equipment maintenance data and defect is generated and used as a prediction model, thereby enabling high-precision prediction of product defect and estimation of the main cause of the defect. Attached Figure Description

[0025] Figure 1 This is a diagram showing the schematic structure of the information processing device of the defective quality diagnosis device that implements the embodiments of the present invention.

[0026] Figure 2This is a diagram illustrating in detail the data segmentation steps performed by the data segmentation unit of the defective diagnostic apparatus according to an embodiment of the present invention.

[0027] Figure 3 This is a diagram illustrating in detail the quality defect classification steps performed by the quality defect classification unit of the quality defect diagnosis device according to an embodiment of the present invention.

[0028] Figure 4 This is a diagram illustrating a first example of classifying continuous quality defects along the length of a product in a quality defect classification step performed by the quality defect classification unit of the quality defect diagnosis device according to an embodiment of the present invention.

[0029] Figure 5 This is a second example illustrating the classification of continuous quality defects along the length of a product in the quality defect classification step performed by the quality defect classification unit of the quality defect diagnosis device according to an embodiment of the present invention.

[0030] Figure 6 This is a diagram illustrating a first example of classifying continuous quality defects in the product width direction during a quality defect classification step performed by the quality defect classification unit of the quality defect diagnosis device according to an embodiment of the present invention.

[0031] Figure 7 This is a second example illustrating the classification of continuous quality defects in the product width direction during the quality defect classification step performed by the quality defect classification unit of the quality defect diagnosis device according to an embodiment of the present invention.

[0032] Figure 8 This is a diagram illustrating in detail the first example of the result display steps performed by the result display unit of the defective quality diagnostic device according to an embodiment of the present invention.

[0033] Figure 9 This is a diagram illustrating a second example of the result display steps performed by the result display unit of the defective quality diagnostic device according to an embodiment of the present invention.

[0034] Figure 10 This is a flowchart illustrating the sequence of methods for generating the prediction model used in the defective quality diagnosis method according to an embodiment of the present invention.

[0035] Figure 11 This is a flowchart illustrating the sequence of the defect diagnosis method according to embodiments of the present invention. Detailed Implementation

[0036] The defect diagnosis apparatus and method of the present invention will be described with reference to the accompanying drawings. The defect diagnosis apparatus and method are used to diagnose quality defects occurring in a product.

[0037] In this embodiment, "product" refers to steel products manufactured through multiple processes, such as semi-finished products like slabs and steel plates manufactured by rolling the slabs. In this embodiment, "quality defects" include, for example, surface defects, internal defects, and material property values ​​(such as the results of mechanical property determination).

[0038] (Information processing device)

[0039] Figure 1 The structure of the information processing device 1, which implements the defective quality diagnosis device of the embodiment, is shown. For example... Figure 1 As shown, the information processing device 1 includes an input unit 10, a storage unit 20, an arithmetic unit 30, and an output unit 40.

[0040] The defect diagnosis device of the embodiment can be implemented by components of the information processing device 1 other than the defect classification unit 32 and the model generation unit 34 of the calculation unit 30. Furthermore, the predictive model generation device used in the defect diagnosis device of the embodiment can be implemented by components of the information processing device 1 other than the defect determination unit 35, the main cause estimation unit 36, and the result display unit 37 of the calculation unit 30. Hereinafter, each component of the information processing device 1 will be described.

[0041] The input unit 10 is an input unit for the arithmetic unit 30, and is implemented, for example, through an input device such as a keyboard, mouse pointer, or numeric keys. The input unit 10 inputs information required for various processing operations in the arithmetic unit 30.

[0042] Storage unit 20 consists of recording media such as EPROM (Erasable Programmable ROM), hard disk drive (HDD), and removable media. Examples of removable media include USB (Universal Serial Bus) memory, CD (Compact Disc), DVD (Digital Versatile Disc), and BD (Blu-ray Disc). Storage unit 20 can store operating systems (OS), various programs, various tables, various databases, etc. Product database 21 is stored in storage unit 20.

[0043] Product DB21 stores data on products manufactured in the past. This product data includes, for example, operational data, equipment maintenance data, and quality data.

[0044] Work data refers to data obtained by measuring the manufacturing state of a product during manufacturing or by estimating it through some means. Preferably, this work data varies over the entire length of the product in a specific direction. "Data varying over the entire length of the product in a specific direction" means, for example, data that varies continuously relative to the length or width direction of the product.

[0045] As operational data, examples include manufacturing condition data for the steelmaking process, hot rolling process, cold rolling process, and annealing process, among others, in various manufacturing steps of steel products. Specifically, operational data may include data on steel composition, steel plate temperature during hot rolling, hot rolling speed, cold rolling speed, steel plate temperature during annealing, and annealing time. Furthermore, operational data can be measured at predetermined intervals based on the distance the steel plate travels along its length relative to the transport direction of each step.

[0046] Equipment maintenance data is data measured or estimated by some means regarding the state and maintenance status of equipment during product manufacturing. Preferably, this equipment maintenance data reflects changes throughout the entire period during which the product is processed or handled within the equipment. Furthermore, it is preferable that the equipment maintenance data directly or indirectly represents the contact status of equipment that comes into contact with the product during the manufacturing process, or that represents the state and maintenance status of equipment that affects a specific location on the product due to direct or indirect actions on the product.

[0047] As equipment maintenance data, for example, data related to the contact state of transport rollers that move products by contacting and rotating them in various manufacturing processes of steel products can be listed. Additionally, as equipment maintenance data, data related to the contact state of walking beams that move products by contacting and lifting them in the heating furnace during the heating process before rolling can be cited. Specifically, as equipment maintenance data, data on the shape of the transport rollers along their entire width and data on the contact position of the walking beam relative to the product can be envisioned.

[0048] Furthermore, the shape data of the transport rollers along their entire width in the equipment maintenance data is, for example, data measured by taking the wear depth of the rollers at predetermined intervals along the width direction relative to the transport rollers as the roller profile. Additionally, the contact position data of the walking beam relative to the product in the equipment maintenance data is, for example, data calculated by measuring the front and rear end positions of the product along its length in the heating furnace and based on the total length of the product in the heating furnace and the position of the walking beam within the furnace. Thus, by using the equipment maintenance data, continuous quality defects along specific directions (length direction and width direction) of the product can be predicted with high accuracy when generating predictive models and diagnosing quality defects.

[0049] Quality data is data indicating the quality of a product, based on measurements or inferences derived through some means. Preferably, this quality data varies along the entire length of the product in a specific direction. Furthermore, examples of quality data include, for instance, data on surface defects (surface blemishes) on the product. More specifically, quality data may include information such as the location and shape of surface blemishes on the product.

[0050] It should be noted that the product data mentioned above includes data generated when creating the predictive model (refer to...). Figure 10 The product data used in the "past performance" and the data used when diagnosing quality defects (refer to...) Figure 11 The product data used by the "diagnostic object".

[0051] In addition to the product DB21, the storage unit 20 also stores, for example, product data corresponding to each block via the data segmentation unit 31, and information related to the classification of quality defects by the quality defect classification unit 32. Furthermore, in addition to the product DB21, the storage unit 20 also stores, for example, operational data and equipment maintenance data aggregated in blocks by the data aggregation unit 33. Additionally, the storage unit 20 may also store, as needed, the quality defect determination results from the quality defect determination unit 35, and the main cause estimation results from the main cause estimation unit 36.

[0052] The arithmetic unit 30 is implemented, for example, by a processor consisting of a CPU (Central Processing Unit) and a memory (main storage unit) consisting of RAM (Random Access Memory) and ROM (Read Only Memory).

[0053] The arithmetic unit 30 loads the program into the working area of ​​the main storage unit and executes it. Through program execution, it controls various component units, thereby achieving the intended function. The arithmetic unit 30, through the execution of the aforementioned program, functions as a data segmentation unit 31, a defect classification unit 32, a data collection unit 33, a model generation unit 34, a defect determination unit 35, a primary cause estimation unit 36, and a result display unit 37. It should be noted that in... Figure 1 The example shown is an example of implementing the functions of each part through a single computer (computation unit), but the means of implementing the functions of each part are not particularly limited. For example, the functions of each part can also be implemented separately through multiple computers.

[0054] The data segmentation unit 31 divides the product data into two dimensions according to each defined block. First, the data segmentation unit 31 collects, for example, operational data, equipment maintenance data, and quality data from the product DB21. Then, for example... Figure 2 As shown, the data segmentation unit 31 divides the product data along its length into p blocks and along its width into q blocks. In this diagram, the leftmost block along the product length is designated as the first block, and the rightmost block is designated as the qth block. Similarly, in this diagram, the topmost block along the product width is designated as the first block, and the bottommost block is designated as the pth block. The data segmentation unit 31 segments the product data, which consists of work data and quality data, along with... Figure 2 Each block obtained from dividing the product is correspondingly stored in the storage unit 20.

[0055] When generating the prediction model, the data segmentation unit 31 (refer to...) Figure 10 The data segmentation unit 31 performs segmentation of product data based on past performance. Additionally, when performing quality defect diagnosis (refer to...), the data segmentation unit 31... Figure 11 The product data for the diagnostic target is segmented. Furthermore, the preferred data segmentation method (number of segments, area of ​​each segment, etc.) is consistent when generating the predictive model and when diagnosing quality defects.

[0056] The data segmentation unit 31 divides the entire length of each product data into p blocks and the entire width of the product data into q blocks. The number of segments (the values ​​of p and q) is not particularly limited, but preferably at least 3 segments (the values ​​of q and p > 3). That is, p is an integer "p ≥ 3", q is an integer "q ≥ 3", and the values ​​of p and q can be arbitrarily determined beforehand. Furthermore, the area of ​​each segmented block can be consistent for each product data, or it can be different. For example, in the classification of quality defects described later (refer to...). Figure 3 In the context of a product, a block classified as having continuous defects in the product length direction and continuous in the product width direction can also be an area (region) that varies depending on the data for each product.

[0057] The quality defect classification unit 32 classifies the quality defects contained in the quality data into two or more categories based on the tendency of quality defects to occur in the segmented blocks. Here, "tendency of quality defects to occur" refers to, for example, the segmented blocks (see reference...). Figure 2 The location and distribution of quality defects in the product.

[0058] Specifically, the defect classification unit 32 classifies quality defects in the quality data into two categories: continuous defects occurring along the length of the product (continuous along the product length direction) and continuous defects occurring throughout the width of the product (continuous along the product width direction). For example, the defect classification unit 32... Figure 3 As shown, the quality defects in the quality data are classified into either continuous defects in the product length direction or continuous defects in the product width direction. In addition, the quality defect classification unit 32 classifies quality defects that are continuous in the product length direction and quality defects that are continuous in the product width direction based on the following criteria (1) to (4), for example.

[0059] (1) When there are multiple defective blocks along the product length direction at various positions along the product width direction, the number of defective blocks is set to s. When s / p≥x, the defective blocks are classified as "continuous along the product length direction". In this case, x is any real number of 0<x≤1, and is set individually according to the type of defect.

[0060] (2) When there are multiple defective blocks along the product length direction at various positions along the product width direction, and when there are more than m consecutive defective blocks, the defect is classified as "continuous along the product length direction". In this case, m is any integer greater than 2, and is set individually according to the type of defect.

[0061] (3) When there are multiple defective blocks along the product width direction at various positions along the product length direction, the number of defective blocks is set to t. When t / q≥y, the defect is classified as "continuous in the product width direction". In this case, y is any real number of 0<y≤1, and is set individually according to the type of defect.

[0062] (4) When there are multiple defective blocks in the product width direction at various positions along the product length direction, and when there are more than n consecutive adjacent defective blocks, the defect is classified as "continuous in the product width direction". In this case, n is any integer greater than 2, and is set individually according to the type of defect.

[0063] Figure 4This indicates a case where the defect classification unit 32 classifies the defect as continuous in the product length direction based on the above-mentioned criterion (1). For example, in the case of "x = 0.7" in the above-mentioned criterion (1), "s = 5", "s / p = 0.71 ≥ x", therefore the defect in this figure is classified as "continuous in the product length direction". It should be noted that this figure shows an example of a defect-generating block that is continuous in the product length direction, but even if the defect-generating block is not continuous in the product length direction, it can still be consistent with the above-mentioned criterion (1). For example, even if the defect-generating blocks are generated in a dispersed manner in the product length direction, if it is consistent with the above-mentioned criterion (1), it is also classified as "continuous in the product length direction".

[0064] Figure 5 This indicates a case where the quality defect classification unit 32 classifies the defect as continuous in the product length direction based on the above-mentioned criterion (2). For example, in the case of "m=3" in the above-mentioned criterion (2), the number of consecutive adjacent blocks in the product length direction is "3", so the quality defect of this figure is classified as "continuous in the product length direction".

[0065] Figure 6 This indicates a case where the defect classification unit 32 classifies the defect as continuous in the product length direction based on the above-mentioned criterion (3). For example, in the case of "y = 0.7" in the above-mentioned criterion (3), "t = 6", "t / q = 0.86 ≥ y", therefore the defect in this figure is classified as "continuous in the product width direction". It should be noted that this figure shows an example where the defect generation blocks are continuous in the product width direction, but even if the defect generation blocks are not continuous in the product width direction, it can still be consistent with the above-mentioned criterion (1). For example, even if the defect generation blocks are generated in a dispersed manner in the product width direction, if it is consistent with the above-mentioned criterion (3), it is also classified as "continuous in the product width direction".

[0066] Figure 7 This indicates a case where the quality defect classification unit 32 classifies the defect as continuous in the product length direction based on the above-mentioned criterion (4). For example, in the case where “n=4” in the above-mentioned criterion (4), the number of consecutive adjacent blocks in the product width direction is “5”, so the quality defect of this diagram is classified as “continuous in the product width direction”.

[0067] The defect classification unit 32 stores information related to the classification of defective products processed as described above in the storage unit 20. Furthermore, when generating a prediction model, the defect classification unit 32 (see...) Figure 10 This involves categorizing past performance quality data. Furthermore, when diagnosing quality defects (refer to...),... Figure 11The model uses the classification of defects generated during the prediction process to collect operational data and equipment maintenance data of the diagnostic objects.

[0068] The data collection unit 33 collects operational data and equipment maintenance data in blocks. First, based on the tendency of quality defects to occur within the divided blocks, the data collection unit 33 categorizes each quality defect according to its classification (see reference). Figure 3 Extract operational data and equipment maintenance data. The defect classification used here is the defect classification (continuous in product length direction, continuous in product width direction) of the defect classification department 32, which was used when generating the prediction model.

[0069] In this case, the data collection unit 33 bases its analysis on the tendency of quality defects to occur (refer to) the presence of multiple defect-causing locations (defect-causing blocks) along the entire length of the product. Figure 4 and Figure 5 The data collection unit 33 extracts operational data and equipment maintenance data according to each defect category. Additionally, the data collection unit 33 analyzes the defect generation trends based on multiple defect locations along the entire product width (refer to...). Figure 6 and Figure 7 ), extract operation data and equipment maintenance data according to each category of defective quality.

[0070] Next, the data collection unit 33 collects the extracted work data and equipment maintenance data for each region containing two or more blocks in a specific direction (continuous in the product length direction and continuous in the product width direction). The data collection unit 33 collects work data and equipment maintenance data, for example, as follows.

[0071] For example, Figure 4 and Figure 5 As shown, for quality defects classified as "continuous along the product length direction", the data collection unit 33 extracts only the operation data and equipment maintenance data of p blocks along the product length direction at each position along the product width direction. Then, the data collection unit 33 collects the extracted operation data and equipment maintenance data in block units.

[0072] In addition, for example, Figure 6 and Figure 7 As shown, for quality defects classified as "continuous in the product width direction," the data collection unit 33 extracts only q blocks of work data and equipment maintenance data along the product width direction at each location along the product length direction. Then, the data collection unit 33 aggregates the extracted work data and equipment maintenance data in block units.

[0073] As a method of data aggregation in the data aggregation unit 33, for example, basic statistics representing the values ​​at each position within the block can be used. Specifically, these basic statistics are values ​​such as mean, maximum, minimum, standard deviation, variance, and the range represented by the difference between the maximum and minimum values.

[0074] In this way, the data collection unit 33 collects the work data and equipment maintenance data by calculating one or more of the following on a block-by-block basis: average, maximum, minimum, standard deviation, variance, and the difference between the maximum and minimum values. It should be noted that at least one basic statistical measure is used when collecting the work data and equipment maintenance data, but multiple basic statistical measures may also be used. The data collection unit 33 saves the collected work data and equipment maintenance data to the storage unit 20 at the block-by-block level.

[0075] When generating the prediction model, the data collection unit 33 (refer to...) Figure 10 This involves collecting past performance data and equipment maintenance data. Additionally, the data collection unit 33 performs quality defect diagnosis (refer to...). Figure 11 This involves collecting operational data and equipment maintenance data of the diagnostic objects.

[0076] The model generation unit 34 generates multiple predictive models corresponding to the categories of quality defects. For each category of quality defects, the model generation unit 34 uses machine learning to learn the relationship between the aggregated work data, equipment maintenance data, and quality data, thereby generating multiple predictive models corresponding to the category of quality defects.

[0077] That is, the model generation unit 34 constructs a predictive model according to the classification of quality defects. This predictive model represents the relationship between the operational data and equipment maintenance data collected in blocks and the quality defects generated in that block or product. The model generation unit 34 generates two predictive models, for example, predicting quality defects that are continuous in the product length direction and continuous in the product width direction. In the generation of these predictive models, various machine learning algorithms, such as deep neural networks, decision trees, and support vector machines, as well as other statistical algorithms, can be used.

[0078] The defect determination unit 35 uses multiple prediction models generated by the model generation unit 34 to determine whether a defect has occurred. The defect determination unit 35 determines whether a defect has occurred in each category by inputting the operational data and equipment maintenance data of the diagnostic object into the prediction model for each category. Here, the operational data and equipment maintenance data input into the prediction model are data collected in blocks after data segmentation in the data segmentation unit 31 and data collection in the data collection unit 33 (see reference). Figure 11 ).

[0079] The quality defect determination unit 35 determines quality defects, for example, before or during product manufacturing. When determined before manufacturing, the unit inputs pre-processing manufacturing conditions and equipment condition settings into a prediction model to calculate and predict quality defects in the manufactured product, using operational and equipment maintenance data for the product to be predicted. Conversely, when determined during manufacturing, the unit inputs the actual values ​​of completed manufacturing and equipment conditions into a prediction model to calculate and predict quality defects in the manufactured product, using operational and equipment maintenance data for the product to be predicted.

[0080] The main cause estimation unit 36 ​​estimates the main cause of the quality defect based on the quality defect determination result of the quality defect determination unit 35. The "main cause of the quality defect" refers to, for example, operational data and equipment maintenance data that are important factors in the quality defect of the product.

[0081] The primary cause estimation unit 36, for example, when a specific product identified as a diagnostic target experiences a quality defect, can estimate the operational data and equipment maintenance data required to suppress the occurrence of such a defect. That is, if it is predicted that a quality defect will occur in a product before or during manufacturing, the operational data and equipment maintenance data that need to be changed or adjusted to prevent the occurrence of such a defect can be estimated. Therefore, it is possible to study countermeasures to suppress the occurrence of quality defects.

[0082] Furthermore, the primary cause estimation unit 36 ​​can estimate the operational data and equipment maintenance data that are the primary cause of a quality defect in a specific product. That is, it estimates the operational data and equipment maintenance data that are the primary cause of a quality defect when it exists in a manufactured product. This allows the primary cause of the quality defect to be determined.

[0083] Furthermore, the primary cause estimation unit 36 ​​can estimate the operational data and equipment maintenance data that are the main causes of quality defects when multiple products experience quality defects. That is, when multiple manufactured products have quality defects, operational data and equipment maintenance data that are the main causes of these defects can be estimated. Therefore, the primary cause of the quality defects can be determined.

[0084] When the main cause estimation unit 36 ​​determines that a specific product targeted for diagnosis has a quality defect, and estimates the operational and equipment maintenance data necessary to suppress the occurrence of the product quality defect, it calculates an index for the product data that is equivalent to the contribution of each factor in the above determination. Then, the factor with a large value for the index is estimated as an important factor. At this time, a general method for calculating the predictive importance factor, which does not rely on the generation algorithm of the predictive model, is used, for example, a method called "SHAP (SHapley Additive exPlanations)".

[0085] Furthermore, when the primary cause estimation unit 36 ​​encounters a situation where a quality defect has occurred in a specific product, and operational data and equipment maintenance data are estimated to be the primary cause of the defect, it calculates an index for that product that corresponds to the contribution of each factor in the aforementioned determination. Then, factors with large values ​​for this index are estimated as important factors. At this point, a general method for calculating the predictive importance factor, such as SHAP described above, which does not rely on a predictive model, is used.

[0086] Furthermore, when the main cause estimation unit 36 ​​calculates an index that represents the relative influence of each factor in the overall prediction model, based on operational data and equipment maintenance data where quality defects have occurred in multiple products and are presumed to be the primary cause of these defects, the factors with larger values ​​in this index are then presumed to be important factors.

[0087] Furthermore, the principal cause estimation unit 36 ​​may, for example, use a predictive importance factor calculation method specific to the algorithm that generates the predictive model, such as the importance of Gini impurity based on the decision tree algorithm. Alternatively, the principal cause estimation unit 36 ​​may use a general predictive importance factor calculation method that is independent of the predictive model generation algorithm, such as "Permutation Importance".

[0088] The result display unit 37 causes the output unit 40 to display (output) at least one of the judgment result from the quality defect judgment unit 35 and the main cause estimation result from the main cause estimation unit 36. Furthermore, it is preferable that the result display unit 37 uniformly displays the data groups that have undergone alignment processing by the data collection unit 33 on the output unit 40.

[0089] When the output unit 40 outputs the quality defect determination result of the quality defect determination unit 35, for example, on a screen where information related to the product being diagnosed can be viewed, the product length direction is set to horizontal and the product width direction is set to vertical, displaying a product plan view divided into blocks according to the classification of quality defects. Then, the quality prediction result for each position of the product is displayed on this product plan view.

[0090] In this case, for example, Figure 8 As shown, defective parts (blocks) can also be marked with an "×" on the product plan view. Alternatively, parts without defects can be marked with an "○" and parts with defects can be marked with an "×". Alternatively, different colors can be used to display the results based on whether or not defects are predicted. Alternatively, only when defects are predicted, the output unit 40 can output a statement such as "It is predicted that the product to be diagnosed (or a specific block of the product to be diagnosed) will have defects".

[0091] In addition, for example, Figure 9 As shown, one or more operational and equipment maintenance data points varying along the product's length can also be displayed graphically at the bottom (or top) of the product plan view. Additionally, one or more operational and equipment maintenance data points varying along the product's width can also be displayed graphically at the left (or right) of the product plan view.

[0092] When the output unit 40 outputs the main cause estimation results of the quality defects from the main cause estimation unit 36, for example, on a screen where information related to the product being diagnosed can be viewed, the product length direction is set to horizontal and the product width direction to vertical, displaying a product plan view divided into block units according to the classification of quality defects. Then, the quality data and quality prediction results for each location of the product are displayed on this product plan view. Alternatively, two product plan views can be displayed, with the quality data for each location of the product displayed on one plan view and the quality prediction results for each location of the product displayed on the other plan view.

[0093] In addition, for example, Figure 9 As shown, the main cause estimation results of quality defects from the main cause estimation section 36 can also be arranged and displayed next to the product plan view. In this case, regarding the main cause estimation results, all items of each input data and the calculated importance index values ​​can be displayed together, or they can be displayed sequentially from the items with the largest calculated importance index values ​​to any given number. The method of displaying the main cause estimation results is not limited, including other methods.

[0094] Furthermore, when displaying the presumed main causes of quality defects in the main cause estimation section 36, one or more operational data and equipment maintenance data varying along the product's length can be displayed graphically at the top or bottom of the product plan view. The displayed data can also be selected sequentially from the data with the highest calculated importance index values.

[0095] Furthermore, when displaying the presumed main causes of quality defects in the main cause estimation section 36, one or more operational data and equipment maintenance data that vary in the product width direction can be displayed graphically on the left or right side of the product plan view. The displayed data can also be selected sequentially from the data with the highest calculated importance index values.

[0096] The output unit 40 is implemented by a display device such as an LCD display or a CRT display. Based on the display signal input from the result display unit 37, the output unit 40 displays, for example, the quality defect determination result of the quality defect determination unit 35 and the main cause estimation result of the quality defect of the main cause estimation unit 36 ​​in the form of text or graphics.

[0097] (Methods for generating prediction models)

[0098] Reference Figure 10 The method for generating the prediction model used in the defective quality diagnosis method of the implementation method will be described. The method for generating the prediction model includes a block segmentation step (step S1), a defective quality classification step (step S2), a data collection step (step S3), and a model generation step (step S4).

[0099] In the block segmentation step, the data segmentation unit 31 divides the past performance product data into two dimensions according to each prescribed block (step S1). Next, in the defect classification step, the defect classification unit 32 classifies the defects contained in the quality data according to the tendency of defects to occur in the segmented blocks (step S2).

[0100] Next, in the data aggregation step, the data aggregation unit 33 aggregates the operational data and equipment maintenance data extracted for each defect category for each region containing two or more blocks in a specific direction (product length direction, product width direction) (step S3). Next, in the model generation step, the model generation unit 34 uses machine learning to learn the relationship between the aggregated operational data and equipment maintenance data and the quality data for each defect category, thereby generating a predictive model (step S4).

[0101] (Methods for diagnosing poor quality)

[0102] Reference Figure 11The method for diagnosing quality defects in the implementation method is described. The method for generating the prediction model includes a block segmentation step (step S11), a data collection step (step S12), a quality defect determination step (step S13), a main cause estimation step (step S14), and a result display step (step S15).

[0103] It should be noted that the following explanation assumes that... Figure 10 Steps S1-S4 utilize predictive models pre-generated at different time points, but predictive models can also be generated during the diagnosis of poor quality. That is, they can be implemented at different time points. Figure 10 Steps S1~S4 and Figure 11 Steps S11~S15, or you can continue... Figure 10 Steps S1 to S4 are performed Figure 11 Steps S11 to S15.

[0104] In the block segmentation step, the data segmentation unit 31 segments the product data of the diagnostic target into a two-dimensional shape according to each prescribed block (step S11). Next, in the data aggregation step, the data aggregation unit 33 aggregates the operational data and equipment maintenance data extracted according to each predetermined quality defect classification for each region containing two or more blocks in a specific direction (product length direction, product width direction) of the product (step S12). It should be noted that in step S12, for example, using... Figure 10 The information related to the classification of defective products generated in step S2 is used to extract work data and equipment maintenance data.

[0105] Next, in the defect determination step, the defect determination unit 35 uses multiple prediction models to determine whether a defect has occurred (step S13). Next, in the main cause estimation step, the main cause estimation unit 36 ​​estimates the main cause of the defect based on the defect determination result (step S14). Next, in the result display step, the result display unit 37 causes the output unit 40 to display the processing results of the defect determination step and the main cause estimation step (step S15).

[0106] In the defect diagnosis apparatus and method described above, the location of the defect in the product is categorized, and a predictive model incorporating the categorization results is used to diagnose the defect. Specifically, in the defect diagnosis apparatus and method described in these embodiments, for each category corresponding to the tendency of defect occurrence, the relationship between operational data, equipment maintenance data, and the defect is generated and used as a predictive model. This allows for high-precision prediction of product defects and the estimation of their root causes.

[0107] The foregoing has provided a detailed description of the defective quality diagnosis device and method of the present invention through embodiments for carrying out the invention. However, the scope of the present invention is not limited to these descriptions and must be broadly interpreted based on the claims. Furthermore, various modifications and alterations based on these descriptions are also included within the scope of the present invention, which is self-evident.

[0108] For example, in the above implementation, multiple predictive models that have learned the relationship between work data, equipment maintenance data and quality data are used to diagnose quality defects, but multiple predictive models that have learned the relationship between equipment maintenance data and quality data can also be used for diagnosis.

[0109] In this case, in the model generation unit 34, for each category of quality defect, machine learning is used to learn the relationship between the aggregated equipment maintenance data and quality data, thereby generating multiple prediction models corresponding to the category of quality defect. Furthermore, in the quality defect determination unit 35, using the multiple prediction models generated for each category of quality defect and having learned the relationship between the aggregated equipment maintenance data and quality data, it is determined whether a quality defect has occurred. In addition, in the main cause estimation unit 36, the main cause of the quality defect is estimated based on the quality defect determination result. Thus, even when only equipment maintenance data and quality data are used as input data for the prediction models, product quality defects can be predicted with high accuracy, and the main cause of the quality defect can be estimated.

[0110] Explanation of reference numerals in the attached figures

[0111] 1 Information processing device

[0112] 10 Input Section

[0113] 20 Storage Department

[0114] 21 Product DB

[0115] 30 Computing Department

[0116] 31 Data Segmentation Department

[0117] 32 Defective Product Classification Department

[0118] 33 Data Collection Department

[0119] 34 Model Generation Department

[0120] 35 Quality Defect Judgment Department

[0121] 36. Presumed Main Causes

[0122] 37 Results Display Department

[0123] 40 Output section.

Claims

1. A defective product diagnostic device, comprising: The data segmentation department divides the product data of the diagnostic object, which consists of operational data, equipment maintenance data, and quality data, into two-dimensional blocks according to each specified block. The data aggregation unit, based on the tendency of quality defects to occur in the segmented blocks, extracts the operation data and the equipment maintenance data according to each predetermined quality defect category, and aggregates the extracted operation data and equipment maintenance data according to regions containing two or more blocks in a specific direction of the product; and The quality defect determination unit uses multiple prediction models generated according to the classification of each quality defect and learned the relationship between the aggregated operation data, equipment maintenance data and quality data to determine whether a quality defect has occurred.

2. The defective quality diagnostic device according to claim 1, wherein, The defective quality diagnostic device includes a primary cause estimation unit, which estimates the primary cause of the defect based on the defective quality determination result.

3. The defective quality diagnostic device according to claim 2, wherein, The defective quality diagnostic device includes a result display unit, which causes the output unit to display at least one of the determination result in the defective quality determination unit and the main cause estimation result in the main cause estimation unit.

4. The defective quality diagnostic device according to claim 1, wherein, The prediction model is generated in the following manner: The product data representing past performance, which consists of the operational data, the equipment maintenance data, and the quality data, is divided into two dimensions according to each defined block. Based on the tendency of the quality defects to occur in the segmented blocks, the quality defects contained in the quality data are classified. The operational data and equipment maintenance data extracted according to each of the aforementioned quality defects are aggregated into regions containing two or more blocks in a specific direction for each product. Based on each of the aforementioned quality defects, machine learning is used to learn the relationship between the aggregated job data, the equipment maintenance data, and the quality data.

5. The defective quality diagnostic device according to claim 1, wherein, The work data is data that varies over the entire length of the product in a specific direction.

6. The defective quality diagnostic device according to claim 1, wherein, The equipment maintenance data refers to the data that changes throughout the entire period during which the product is processed and handled in the equipment.

7. The defective quality diagnostic device according to claim 1, wherein, The quality data is data that varies over the entire length of the product in a specific direction.

8. The defective quality diagnostic device according to claim 2, wherein, The equipment maintenance data represents the equipment status and maintenance status that affect a specific location of the product through direct or indirect actions on the product.

9. The defective quality diagnostic device according to claim 1, wherein, The data collection unit extracts the operation data and the equipment maintenance data according to each predetermined defect category, based on the defect tendency of having multiple defect locations along the entire length direction of the product or the defect tendency of having multiple defect locations along the entire width direction of the product.

10. The defective quality diagnostic device according to claim 8, wherein, The defect determination unit uses multiple prediction models generated according to the classification of each defect and learned the relationship between the aggregated equipment maintenance data and the quality data to determine whether a defect has occurred. The main cause estimation section estimates the main cause of the quality defect based on the determination result of the quality defect.

11. The defective quality diagnostic device according to claim 3, wherein, The result display unit causes the output unit to display the data group that has been aligned by the data collection unit.

12. A method for diagnosing quality defects, comprising: In the data segmentation step, the computer's data segmentation unit divides the product data of the diagnostic object, which consists of operational data, equipment maintenance data, and quality data, into two-dimensional blocks according to each specified block. The data collection step involves the computer using data collection capabilities to extract the operation data and equipment maintenance data based on the tendency of quality defects to occur in the segmented blocks, according to each predetermined category of quality defects, and then collecting the extracted operation data and equipment maintenance data according to regions containing two or more blocks in a specific direction of the product. and In the defect determination step, the defect determination unit of the computer uses multiple prediction models generated according to each defect category and learned the relationship between the aggregated operation data, equipment maintenance data and quality data to determine whether a defect has occurred.

13. The method for diagnosing quality defects according to claim 12, wherein, The system includes a primary cause estimation step, in which the primary cause estimation unit of the computer estimates the primary cause of the quality defect based on the result of the quality defect determination.

Images