Fault identification system

Abstract

This invention provides a fault identification system that can more easily identify the cause of a failure in a manufacturing line compared to conventional systems, and can display the cause of the failure in natural language. [Solution] The configuration includes a language data acquisition unit that acquires unstructured data relating to the state or settings of a manufacturing line written or entered by an operator in natural language; a sensor acquisition unit that acquires sensor data from each manufacturing device; a first feature extraction unit that extracts first feature data from the unstructured data; a model analysis unit that uses a failure analysis model including a large-scale language model to identify the cause of the failure and the corresponding part of the unstructured data relating to the cause of the failure, and generates a first objective variable including failure display data for displaying the cause of the failure and the corresponding part in natural language, based on the first feature data, sensor data during normal operation and when a failure occurs, and first explanatory variables including measured or predicted values ​​of the quality of the product or intermediate product during normal operation and when a failure occurs; and a display unit that displays the failure display data.

Description

Technical Field

[0001] The present invention relates to a failure identification system for identifying the cause of a failure in a production line. Background Art

[0002] Conventionally, products have been manufactured by a production line constituted by a plurality of manufacturing apparatuses (for example, Patent Document 1). In such a production line, when a failure occurs in one manufacturing apparatus among the plurality of manufacturing apparatuses, the failure of the manufacturing apparatus may be caused not by a defect of the manufacturing apparatus itself but by an external factor from another manufacturing apparatus. In such a case, even if the failed manufacturing apparatus is repaired, there is a possibility that the manufacturing apparatus will fail again or other manufacturing apparatuses will fail in a chain reaction. Conventionally, a skilled worker has verified the factors causing the failure of the manufacturing apparatus one by one based on empirical rules to identify the failure. Prior Art Documents Patent Documents

[0003] Patent Document 1 Japanese Unexamined Patent Application Publication No. 2024-92754 Summary of the Invention Problems to be Solved by the Invention

[0004] However, conventional failure identification methods require the empirical rules of skilled workers, and cannot allow any worker to easily identify a failure in a manufacturing apparatus regardless of their skill level.

[0005] Accordingly, an object of the present invention is to provide a failure identification system that can more easily identify the cause of a failure in a production line than conventional systems and can display the cause of the failure in natural language. Means for Solving the Problems

[0006] One approach to solving the above-mentioned problems is a fault identification system for identifying the cause of a failure when a product manufacturing line including one or more manufacturing devices fails, comprising: a language data acquisition unit that acquires unstructured data relating to the state or settings of the manufacturing line written or entered by an operator in natural language; a sensor acquisition unit that acquires sensor data provided on each manufacturing device of the manufacturing line; a first feature extraction unit that extracts first feature data from the unstructured data acquired by the language data acquisition unit; the first feature data extracted by the first feature extraction unit, sensor data acquired by the sensor acquisition unit when the manufacturing line is operating normally, and the sensor acquisition unit The fault identification system comprises: a model analysis unit that estimates the cause of the failure and identifies the relevant portion of the unstructured data related to the cause of the failure, using a failure analysis model including a large-scale language model, from first explanatory variables including sensor data acquired when the manufacturing line fails, measured or predicted values ​​of the quality of the product or intermediate product of the product when the manufacturing line is operating normally, and measured or predicted values ​​of the quality of the product or intermediate product of the product when the manufacturing line fails, and generates a first objective variable including fault display data for displaying the cause of the failure and the relevant portion in natural language; and a display unit that displays the fault display data generated by the model analysis unit.

[0007] According to this configuration, it is possible to identify the cause of a malfunction by reflecting unstructured data regarding the status or settings of the manufacturing line, which is written or entered by the operator in natural language. Furthermore, the display unit shows the cause of the malfunction along with the relevant section of the unstructured data related to the cause of the malfunction, allowing the operator to take action based on the evidence for the cause of the malfunction. As a result, even inexperienced operators can easily handle malfunctions compared to conventional methods.

[0008] A preferred configuration is that the model analysis unit, using the failure analysis model, further identifies countermeasures for the cause of the failure from the first explanatory variables and generates a first objective variable that includes the failure display data for displaying the cause of the failure, the relevant location, and the countermeasures in natural language.

[0009] A preferred configuration further comprises a second feature extraction unit that analyzes a photographic image of the product or an intermediate product and extracts second feature data of the product or intermediate product from the photographic image, wherein the first explanatory variable includes second feature data extracted by the second feature extraction unit when the manufacturing line is operating normally and second feature data extracted by the second feature extraction unit when the manufacturing line is malfunctioning.

[0010] A preferred configuration further comprises a quality prediction unit that generates a second objective variable, which includes a predicted quality of the product or an intermediate product of the product, using a machine learning model from a second explanatory variable, which includes first feature data of the manufacturing line extracted by the first feature extraction unit and sensor data of the manufacturing line acquired by the sensor acquisition unit, wherein the first explanatory variable includes a predicted quality of the product or an intermediate product of the product when the manufacturing line is operating normally and a predicted quality of the product or an intermediate product of the product when the manufacturing line fails, as predicted by the quality prediction unit.

[0011] A preferred configuration is that the unstructured data is based on the worker's daily work report and / or worker's anomaly report.

[0012] A preferred configuration is one in which the unstructured data includes multiple reporting items, each including the time when the state of the production line changed or the settings of the production line were changed, and a report from the worker who confirmed the change in the state of the production line at that time or the worker who changed the settings of the production line at that time, and the first feature extraction unit extracts a feature vector from the unstructured data for each of the reporting items.

[0013] The aspects described above can be dependent on each other, refer to some of their components, or substitute for some of their components, as long as they fall within the technical scope of the present invention. [Effects of the Invention]

[0014] According to the present invention, the cause of a failure in a manufacturing line can be identified more easily than in the conventional method, and the cause of the failure can be displayed in natural language. [Brief explanation of the drawing]

[0015] [Figure 1] This is a schematic block diagram showing the fault identification system of the first embodiment of the present invention. [Figure 2] (a) is a block diagram showing an example of a failure analysis model by the model analysis unit in Figure 1, and (b) is a block diagram showing an example of a machine learning model by the quality prediction unit in Figure 1. [Figure 3] (a) is a diagram showing an example of a reporting medium, and (b) is a diagram showing an example of a display image displayed by the display unit. [Figure 4] This is a schematic block diagram showing a fault identification system according to another embodiment of the present invention. [Modes for carrying out the invention]

[0016] Embodiments of the present invention will be described in detail below.

[0017] The first embodiment of the present invention, the fault identification system 1, when a fault occurs in a manufacturing line 200 having one or more manufacturing devices, uses a fault identification program to identify the cause and location of the fault in the manufacturing line 200 and proposes a solution to the worker. As shown in Figure 1, the fault identification system 1 comprises a diagnostic device 2, an unstructured input unit 3, an imaging unit 4, and a display unit 5.

[0018] <Diagnostic device 2> The diagnostic device 2 is a computer having a hardware configuration including a central processing unit configured of a control device that controls each device and an arithmetic device that performs computation on data, a storage device that stores data, an input device that inputs data from the outside, and an output device that outputs data to the outside. As shown in FIG. 1, the diagnostic device 2 includes, as main constituent components, a language data acquisition unit 10, a sensor acquisition unit 11, an image acquisition unit 12, a first feature amount extraction unit 13, a second feature amount extraction unit 14, a model analysis unit 15, a quality prediction unit 16, and a data storage unit 17.

[0019] (Language Data Acquisition Unit 10) The language data acquisition unit 10 is a data acquisition unit that acquires unstructured data input by the unstructured input unit 3, and is capable of outputting the acquired unstructured data to the first feature amount extraction unit 13 and the data storage unit 17. The unstructured data is obtained by converting, by the unstructured input unit 3, a reporting medium 201 such as a daily work report or an abnormality report in which an operator describes data related to the state of the production line 200 or the settings of the production line 200 in natural language as shown in FIG. 3(a). As shown in FIG. 3(a), the reporting medium 201 preferably includes a plurality of report items A to D including the time when the state of the production line 200 changed, the report from the operator who confirmed the change in the state of the production line 200 at that time, and / or the time when the operator changed the settings of the production line 200 and the report from the operator who changed the settings of the production line 200 at that time. The reporting medium 201 is not particularly limited as long as it is one in which an operator describes data related to the state of the production line 200 or the settings of the production line 200 in natural language, and may be a handwritten notebook of the operator which is a paper medium such as a foreman's notebook or a manufacturing notebook.

[0020] (Sensor Acquisition Unit 11) The sensor acquisition unit 11 is a data acquisition unit that converts detection values of sensors of each manufacturing apparatus from the production line 200 into structured data, and acquires the converted data as sensor data. In other words, the sensor acquisition unit 11 converts the detected values ​​(e.g., temperature, pressure, flow rate, tension, etc.) measured by the sensors of each manufacturing device into structured data and can output them as sensor data to the model analysis unit 15, the quality prediction unit 16, and the data storage unit 17 as needed.

[0021] (Image acquisition unit 12) The image acquisition unit 12 is a data collection unit that acquires images of products or intermediate products (hereinafter collectively referred to as "products, etc.") captured by the shooting unit 4, and is capable of outputting the acquired images of products, etc. to the second feature extraction unit 14 and the data storage unit 17.

[0022] (First feature extraction unit 13) The first feature extraction unit 13 is the part that uses a first feature extraction model to extract first features, which are latent variables related to quality, from the unstructured data acquired by the language data acquisition unit 10, and generates first feature data that includes a first feature vector. The first feature extraction unit 13 extracts feature vectors for each reporting item from unstructured data that reflects the reporting medium 201. In other words, if there are multiple reporting items in the unstructured data, the first feature extraction unit 13 extracts a multi-dimensional feature vector, which is a numerical representation of each reporting item, as the first feature vector. For example, the first feature extraction unit 13 divides the unstructured data corresponding to the reporting medium 201 into blocks according to reporting items A to D, as shown in Figure 3(a), and extracts a first feature vector (a, b, c, d) corresponding to each block. For example, large-scale language models such as GTP, Gemeni, and Llama can be used as the first feature extraction model in the first feature extraction unit 13.

[0023] (Second feature extraction unit 14) The second feature extraction unit 14 is a part that uses a second feature extraction model, which is a computer vision model, to extract second features from captured images of products, etc., acquired by the image acquisition unit 12, and generates second feature data that includes a second feature vector. The second feature quantities extracted by the second feature extraction unit 14 are not particularly limited as long as they can reflect the unique characteristics of the product, etc. For example, they may include shape-related features (shape indices) such as aspect ratio, area, roundness, perimeter, equivalent diameter, length of the longer side of the aspect ratio, length of the shorter side of the aspect ratio, and normality, or color-related features (color indices) such as hue, saturation, and brightness. The second feature extraction model used in the second feature extraction unit 14 is not particularly limited to computer vision models; for example, segmentation models such as Segment Anything Model, Seg-Net, PSP-Net, U-Net, and MaskR-CNN can be used.

[0024] (Model Analysis Section 15) The model analysis unit 15 calculates the first objective variable from the first explanatory variable using a failure analysis model. As shown in Figure 2(a), the model analysis unit 15 of this embodiment uses a failure analysis model to identify the cause of failure in the manufacturing line 200 and the corresponding location in the unstructured data related to the cause of failure (the corresponding location in the reporting medium 201) from a first explanatory variable which includes first feature data, second feature data when the manufacturing line 200 is operating normally and when a failure occurs, sensor data when the manufacturing line 200 is operating normally and when a failure occurs, and quality data of products, etc. when the manufacturing line 200 is operating normally and when a failure occurs. The model analysis unit 15 of this embodiment generates a first objective variable which includes failure display data for displaying the cause of failure, the corresponding location, and the countermeasures in natural language. In other words, the model analysis unit 15 is an explanation generation unit that creates data to explain the cause of the failure, the relevant location, and countermeasures in natural language, and it is also a countermeasure proposal unit that proposes countermeasures for the failure. For example, large-scale language models such as GTP, Gemeni, and Llama can be used as failure analysis models in the model analysis unit 15. The quality data includes measured values ​​and / or predicted values ​​predicted by the quality prediction unit 16 that evaluate the quality of the product, etc. (e.g., tactile feel, gloss, etc.). The quality data in this embodiment includes predicted values ​​of the product quality predicted by the quality prediction unit 16, but does not include measured values ​​that evaluate the quality of the product.

[0025] (Quality Forecasting Department 16) The quality prediction unit 16 calculates the second objective variable from the second explanatory variable using a machine learning model. As shown in Figure 2(b), the quality prediction unit 16 of this embodiment generates a second objective variable, which includes quality data containing predicted values ​​of product quality, using a machine learning model, from a second explanatory variable that includes first feature data of the manufacturing line 200 extracted by the first feature extraction unit 13, second feature data of captured images of products, etc. extracted by the second feature extraction unit 14, and sensor data of the manufacturing line 200 acquired by the sensor acquisition unit 11. Machine learning models that can be used include supervised learning models such as regression, decision trees, and neural networks, as well as unsupervised learning models such as k-means and PCA.

[0026] (Data storage unit 17) The data storage unit 17 is a part that stores various types of data, and stores and accumulates the following: unstructured data acquired by the past language data acquisition unit 10, sensor data acquired by the sensor acquisition unit 11, captured images acquired by the image acquisition unit 12, first feature data extracted by the first feature extraction unit 13, second feature data extracted by the second feature extraction unit 14, fault indication data calculated by the model analysis unit 15, and quality data predicted by the quality prediction unit 16.

[0027] <Unstructured Input Section 3> The unstructured input unit 3 is the part that uses a reading device such as a scanner to convert a report medium 201, such as a document in which an operator has written data regarding the status or settings of the manufacturing line 200 in natural language, into unstructured data and input it into the language data acquisition unit 10. Reporting medium 201 is not particularly limited, but examples include handwritten notes by workers such as daily reports, supervisor notes, and manufacturing notes.

[0028] <Photography Department 4> The imaging unit 4 is the part that photographs the product or an intermediate product (product, etc.) and inputs the captured image to the image acquisition unit 12. The images used are not particularly limited as long as they relate to the product and express its unique characteristics (shape and color), but for example, cross-sectional images of the product can be used.

[0029] <Display section 5> The display unit 5 is a notification unit that displays the fault indication data generated by the model analysis unit 15 as an image, and informs the operator of the cause of the fault, the relevant location, and the countermeasures.

[0030] Next, we will describe a fault identification method performed by a fault identification program using the fault identification system 1 of this embodiment.

[0031] The fault identification method performed by the fault identification program of the fault identification system 1 of this embodiment mainly includes a first feature extraction step, a second feature extraction step, a sensor data generation step, a quality data generation step, a fault analysis step, and a display step.

[0032] (First feature extraction step) The first feature extraction step involves the language data acquisition unit 10 acquiring a reporting medium 201 containing data on the status or settings of the manufacturing line 200, which has been written or entered by an operator in natural language, converting it into unstructured data, and outputting it to the first feature extraction unit 13 and the data storage unit 17. The first feature extraction unit 13 then uses a first feature extraction model to extract first features from the input unstructured data and generates first feature data including a first feature vector.

[0033] (Second feature extraction process) The second feature extraction step includes a normal feature generation step in which the second feature extraction unit 14 extracts second features from images of products, etc., taken by the imaging unit 4 in the past during normal operation and stored in the data storage unit 17, and generates second feature data for normal operation that includes a second feature vector, and a failure feature generation step in which the second feature extraction unit 14 extracts second features from images of products, etc., taken by the image acquisition unit 12 in the past or most recently when a failure occurred, and generates second feature data for failure operation that includes a second feature vector.

[0034] (Sensor data generation process) The sensor data generation process is a process in which the sensor acquisition unit 11 converts the detected values ​​detected by each sensor on the manufacturing line 200 into structured sensor data.

[0035] (Quality data generation process) The quality data generation process includes a normal quality prediction process in which the quality prediction unit 16 calculates quality data, including a predicted value of the quality of the product, etc., under normal operation, as a second objective variable using machine learning from a second explanatory variable that includes a first feature data, a second feature data under normal operation, and sensor data under normal operation, and a second objective variable; and a failure quality prediction process in which the quality prediction unit 16 calculates quality data, including a predicted value of the quality of the product, etc., at the time of failure, as a second objective variable using machine learning from a second explanatory variable that includes a first feature data, a second feature data at the time of failure, and sensor data at the time of failure.

[0036] (Failure analysis process) The failure analysis process involves the model analysis unit 15 using a failure analysis model to identify the cause of failure in the manufacturing line 200 and the relevant locations in the unstructured data related to the cause of failure, creating countermeasures for the cause of failure, and calculating failure display data as the first objective variable for displaying the cause of failure, the relevant locations, and countermeasures in natural language. This data includes first feature data extracted in the first feature extraction process, second feature data extracted in the second feature extraction process for normal operation and failure occurrence, sensor data generated in the sensor data generation process for normal operation and failure occurrence, and quality data for products, etc., predicted in the quality data generation process.

[0037] (Display process) The display process involves displaying the failure display data generated in the failure analysis process using the display unit 5. The display image 100 displayed by the display unit 5 includes, as shown in Figure 3(b), a status display unit 101 showing the status of the manufacturing line 200, a cause display unit 102 showing the cause of the failure, a countermeasure display unit 103 showing the countermeasure for the cause of the failure, and a relevant display unit 104 showing the relevant section in the reporting medium 201.

[0038] The failure identification system 1 of this embodiment identifies the cause of a failure when a product manufacturing line 200, which includes one or more manufacturing devices, fails. It comprises a language data acquisition unit 10 that acquires unstructured data from a reporting medium 201 concerning the status or settings of the manufacturing line 200, written in natural language by an operator; a sensor acquisition unit 11 that acquires detection values ​​from sensors provided on each manufacturing device of the manufacturing line 200 as sensor data; a first feature extraction unit 13 that extracts first feature data from the unstructured data acquired by the language data acquisition unit 10; and the first feature data extracted by the first feature extraction unit 13 and the manufacturing line 200 acquired by the sensor acquisition unit 11. The system includes a model analysis unit 15 that uses a failure analysis model including a large-scale language model to estimate the cause of failure and identify the relevant parts of unstructured data related to the cause of failure, and generates a first objective variable that includes failure display data for displaying the cause of failure and the relevant parts in natural language, based on first explanatory variables including sensor data when the system is always in operation, sensor data acquired by the sensor acquisition unit 11 when the manufacturing line 200 fails, predicted quality values ​​of products or intermediate products when the manufacturing line 200 is operating normally, and predicted quality values ​​of products or intermediate products when the manufacturing line 200 fails. The system also includes a display unit 5 that displays the failure display data generated by the model analysis unit 15. This configuration allows for the identification of the cause of a malfunction by reflecting unstructured data regarding the status or settings of the manufacturing line 200, which is written in natural language by the operator. Furthermore, the display unit 5 displays the cause of the malfunction along with the relevant section of the unstructured data related to the cause of the malfunction, allowing the operator to take action based on the rationale behind the malfunction. As a result, even inexperienced operators can easily handle malfunctions compared to conventional methods, reducing maintenance time.

[0039] In the fault identification system 1 of this embodiment, it is preferable that the model analysis unit 15 further identifies countermeasures for the cause of the failure using a failure analysis model from the first explanatory variables, and generates a first objective variable that includes fault display data for displaying the cause of the failure, the relevant location, and the countermeasures in natural language. This makes it easier for workers to deal with malfunctions.

[0040] In the fault identification system 1 of this embodiment, it is preferable that the system further includes a second feature extraction unit 14 that analyzes captured images of a product or an intermediate product and extracts second feature data of the product or intermediate product from the captured images, and that the first explanatory variable includes second feature data extracted by the second feature extraction unit 14 in the past when the manufacturing line 200 was operating normally and second feature data extracted by the second feature extraction unit 14 when the manufacturing line 200 was in failure. This allows for a more accurate identification of the cause of the malfunction.

[0041] In the fault identification system 1 of this embodiment, the system further includes a quality prediction unit 16 that generates a second objective variable, which includes a predicted value of the quality of the product or an intermediate product of the product, using a machine learning model from second explanatory variables, which include first feature data of the manufacturing line 200 extracted by a first feature extraction unit 13 and sensor data of the manufacturing line 200 acquired by a sensor acquisition unit 11. Preferably, the first explanatory variables include the predicted value of the quality of the product or an intermediate product of the product when the manufacturing line 200 is operating normally, as predicted by the quality prediction unit 16, and the predicted value of the quality of the product or an intermediate product of the product when the manufacturing line 200 fails. This allows for the identification of the cause of failures, etc., without actually evaluating the product or intermediate products.

[0042] In the fault identification system 1 of this embodiment, the unstructured data is preferably data based on the worker's work report and / or the worker's anomaly report. This allows for highly accurate identification of the cause of the malfunction, based on the operator's expertise, and enables the presentation of compelling solutions.

[0043] In the fault identification system 1 of this embodiment, the unstructured data includes multiple reporting items A to D, which include the time when the state of the manufacturing line 200 changed or the settings of the manufacturing line 200 were changed, and reports from the worker who confirmed the change in the state of the manufacturing line 200 at that time or the worker who changed the settings of the manufacturing line 200 at that time. The first feature extraction unit 13 preferably extracts a feature vector from the unstructured data for each reporting item. This makes it easier to identify the relevant section in unstructured data that is causing the failure.

[0044] In the embodiment described above, the model analysis unit 15 used the predicted quality values ​​of the product or intermediate product predicted by the quality prediction unit 16 as quality data, but the present invention is not limited thereto. The model analysis unit 15 may use the measured quality values ​​of the product or intermediate product as quality data instead of the predicted quality values ​​of the product or intermediate product predicted by the quality prediction unit 16. In this case, as shown in Figure 4, the quality evaluation unit 300 for evaluating the quality of the product or intermediate product will be included instead of the quality prediction unit 16.

[0045] In the embodiment described above, the unstructured input unit 3 converted the report medium 201, which was written by the worker, into unstructured data. However, the present invention is not limited to this. The unstructured input unit 3 may also convert the report medium 201, which was entered by the worker using an input device such as a mouse or keyboard, into unstructured data.

[0046] In the embodiment described above, the diagnostic device 2 was composed of a single computer, but the present invention is not limited thereto. It may be composed of multiple computers. In this case, the computers constituting the diagnostic device 2 may be connected wirelessly or by wire, or they may be connected via a network such as the Internet or an intranet.

[0047] In the embodiments described above, the components can be freely substituted or added between each embodiment, as long as they fall within the technical scope of the present invention. [Explanation of symbols]

[0048] 1. Fault Identification System 5 Display section 10. Language data acquisition unit 11 Sensor acquisition unit 13. First Feature Extraction Unit 14. Second Feature Extraction Unit 15 Model Analysis Department 16 Quality Forecasting Department 200 production lines

Claims

1. A fault identification system for identifying the cause of a failure when a product manufacturing line including one or more manufacturing devices fails, A language data acquisition unit that acquires unstructured data relating to the status or settings of the manufacturing line, which is written or entered by an operator in natural language, A sensor acquisition unit that acquires sensor data provided in each manufacturing device of the aforementioned manufacturing line, A first feature extraction unit extracts first feature data from the unstructured data acquired by the language data acquisition unit, A model analysis unit generates a first objective variable that uses a failure analysis model including a large-scale language model to estimate the cause of the failure and identify the relevant portion of the unstructured data related to the cause of the failure, and displays the cause of the failure and the relevant portion in natural language, from first feature data extracted by the first feature extraction unit, sensor data acquired by the sensor acquisition unit when the manufacturing line is operating normally, sensor data acquired by the sensor acquisition unit when the manufacturing line is malfunctioning, measured or predicted quality values ​​of the product or intermediate product of the product when the manufacturing line is operating normally, and first explanatory variables including measured or predicted quality values ​​of the product or intermediate product of the product when the manufacturing line is malfunctioning, A fault identification system comprising: a display unit that displays the fault indication data generated by the model analysis unit.

2. The failure identification system according to claim 1, wherein the model analysis unit further identifies countermeasures for the cause of the failure using the failure analysis model from the first explanatory variables, and generates a first target variable that includes the cause of the failure, the relevant location, and the failure display data for displaying the countermeasures in natural language.

3. The system further includes a second feature extraction unit that analyzes a photographic image of the product or an intermediate product and extracts second feature data of the product or intermediate product from the photographic image. The fault identification system according to claim 1 or 2, wherein the first explanatory variable includes second feature data extracted by the second feature extraction unit when the manufacturing line is operating normally and second feature data extracted by the second feature extraction unit when the manufacturing line is in failure.

4. The system further includes a quality prediction unit that generates a second objective variable, which includes a predicted quality value of the product or an intermediate product of the product, using a machine learning model, from a second explanatory variable that includes first feature data of the manufacturing line extracted by the first feature extraction unit and sensor data of the manufacturing line acquired by the sensor acquisition unit. The fault identification system according to claim 1 or 2, wherein the first explanatory variable includes a predicted value of the quality of the product or an intermediate product of the product when the manufacturing line is operating normally, and a predicted value of the quality of the product or an intermediate product of the product when the manufacturing line fails, as predicted by the quality prediction unit.

5. The fault identification system according to claim 1 or 2, wherein the unstructured data is data based on the worker's work report and / or the worker's anomaly report.

6. The aforementioned unstructured data includes multiple reporting items, which include the time when the state of the production line changed or the settings of the production line were changed, and reports from the worker who confirmed the change in the state of the production line at that time or the worker who changed the settings of the production line at that time. The fault identification system according to claim 1 or 2, wherein the first feature extraction unit extracts a feature vector from the unstructured data for each of the reporting items.

Images