Intelligent anomaly detection device and its control method

Abstract

To provide an intelligent abnormality detection device 100 and its control method.SOLUTION: In an intelligent abnormality detection system 1 in which a user terminal 200 interlocks with an intelligent abnormality detection device 100, an external server 300 and a target manufacturing facility A through a network, the intelligent abnormality detection device 100 includes a data collection part for collecting manufacturing information of the target manufacturing facility, an abnormality detection part for generating a manufacturing quality abnormal degree on the basis of the manufacturing information, and an interpretation generation part for generating interpretation information on the basis of the manufacturing quality abnormal degree. The interpretation generation part generates the interpretation information including visualization information for a cause corresponding to abnormality and explanatory information corresponding to the visualization information when the abnormality exists in manufacturing quality.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present application relates to an intelligent anomaly detection device and a control method thereof. [Background technology]

[0002] In the manufacturing industry, research is ongoing into AI-based anomaly detection models that can distinguish between good and defective products using machine vision to improve production quality. However, due to the excessive cost and time required for technological development, it is difficult for small and medium-sized manufacturers to adopt this model.

[0003] In addition, even if an AI-based quality verification system is established at a manufacturing site, workers on the site have low trust in AI technology, so they do not fully utilize the applied system and instead solve problems based on their own experience or conduct quality verification themselves.

[0004] Specifically, when looking at examples of applying AI-based anomaly detection technology to manufacturing quality in Korea, time-series sensor data or image and video data are collected and analyzed, and if it is determined that the manufacturing quality is poor, it only has a general system of classifying it by process. However, there are many shortcomings in the technology that can detect the cause of poor manufacturing quality or explain the reason for the defect, making it difficult for workers to use it reliably. [Prior art documents] [Patent documents]

[0005] Patent Document 1: Korean Patent Publication No. 10-2367597 Summary of the Invention [Problem to be solved by the invention]

[0006] The present application is directed to solving the problems of the prior art described above, and aims to provide an intelligent anomaly detection device and control method thereof that can solve the problems of high development costs, long development times, and low versatility of artificial intelligence-based manufacturing processes and product quality verification technologies.

[0007] The present application is directed to solving the problems of the prior art described above, and aims to provide an intelligent anomaly detection device and a control method thereof that can solve the problem of insufficient technology to detect the cause of defects in manufacturing quality or explain the reason for the defect.

[0008] However, the technical objectives that the embodiments of the present application aim to achieve are not limited to those described above, and other technical objectives may exist. [Means for solving the problem]

[0009] As a technical means for achieving the above technical objectives, an intelligent anomaly detection device according to one embodiment of the present application may include a collection unit that collects manufacturing information of a target manufacturing equipment, an anomaly detection unit that generates a manufacturing quality anomaly level based on the manufacturing information, and an interpretation generation unit that generates interpretation information based on the manufacturing quality anomaly level.

[0010] According to one embodiment of the present application, when an abnormality exists in manufacturing quality, the interpretation generation unit may generate the interpretation information including visualization information for a cause corresponding to the abnormality and explanatory information corresponding to the visualization information.

[0011] According to an embodiment of the present application, the intelligent anomaly detection device may further include a pre-processing unit that performs pre-processing on the manufacturing information, including at least one of a refinement technique, a normalization technique, and an enhancement technique.

[0012] According to one embodiment of the present application, if the manufacturing information is time-series information, the pre-processing unit may perform pre-processing on the time-series information, including at least one correction selected from the group consisting of distribution conversion correction, missing value correction, and outlier correction. If the manufacturing information is image information, the pre-processing unit may perform pre-processing on the image information, including at least one correction selected from the group consisting of noise correction, size correction, and hue correction.

[0013] According to an embodiment of the present application, if the manufacturing information is image information, the pre-processing unit may generate the image information processed based on the image information to enhance the image information.

[0014] According to one embodiment of the present application, the intelligent anomaly detection device may further include a characteristic variable extraction unit that analyzes the manufacturing information based on pre-set manufacturing quality verification elements and extracts characteristic variables based on the analysis results of the manufacturing information.

[0015] According to one embodiment of the present application, the characteristic variable extraction unit inputs the manufacturing information and the analysis results to a variable extraction artificial neural network to output characteristic variables, and the variable extraction artificial neural network can learn the manufacturing information, the manufacturing information analysis results, and the characteristic variables as a learning dataset.

[0016] According to one embodiment of the present application, the intelligent anomaly detection device may include a model generation unit that generates a plurality of artificial neural networks that have been trained using the manufacturing information, the characteristic variables, and manufactured product information as a learning dataset, and sets the artificial neural network having the highest accuracy in outputting a manufacturing quality anomaly level among the plurality of artificial neural networks as an anomaly detection model.

[0017] According to an embodiment of the present application, the anomaly detection unit may input the characteristic variables to the anomaly detection model and output the degree of anomaly in the manufacturing quality.

[0018] According to one embodiment of the present application, if the degree of manufacturing quality anomaly is equal to or greater than a preset critical anomaly level, the anomaly detection unit can apply at least one of a plurality of analysis techniques to the degree of manufacturing quality anomaly to calculate the process causing the problem.

[0019] According to an embodiment of the present application, the interpretation generating unit may generate the interpretation information including the visualization information and the explanation information based on the problem cause process.

[0020] According to an embodiment of the present application, the intelligent anomaly detection device may further include a control unit that controls the target manufacturing equipment based on the degree of the manufacturing quality anomaly.

[0021] According to an embodiment of the present application, the control unit may control the target manufacturing equipment based on the problem-causing process.

[0022] A control method of an intelligent anomaly detection device according to one embodiment of the present application may include a step of collecting manufacturing information of a target manufacturing equipment, a step of generating a manufacturing quality anomaly level based on the manufacturing information, and a step of generating interpretation information based on the manufacturing quality anomaly level.

[0023] The above-mentioned problem solving means are merely exemplary and should not be construed as limiting the present application. In addition to the above-mentioned exemplary embodiments, there may be additional embodiments in the drawings and detailed description of the invention. Effect of the Invention

[0024] According to the above-mentioned problem solving means of the present application, by providing an intelligent anomaly detection device and a control method thereof, it is possible to solve the problems of high development costs, long development times, and low versatility of artificial intelligence-based manufacturing processes and product quality verification technologies.

[0025] According to the above-mentioned solution to the problem of the present application, by providing an intelligent anomaly detection device and a control method thereof, it is possible to solve the problem of insufficient technology to detect the cause of defects in manufacturing quality or explain the reason for the defect.

[0026] However, the effects obtained by the present invention are not limited to the above-mentioned effects, and other effects may also exist. [Brief description of the drawings]

[0027] [Figure 1] 1 is a schematic configuration diagram of a first form of an intelligent anomaly detection system according to an embodiment of the present application; [Diagram 2] 1 is a schematic configuration diagram of a second embodiment of an intelligent anomaly detection system according to an embodiment of the present application; [Diagram 3] 1 is a schematic block diagram of an intelligent anomaly detection device according to an embodiment of the present application; [Figure 4] 1 is a schematic block diagram of an intelligent anomaly detection device according to another embodiment of the present application; [Diagram 5] 2 is a diagram illustrating time series information and image information generated in a manufacturing process according to an embodiment of the present disclosure; [Figure 6] 1 is a diagram illustrating an image corresponding to an anomaly output by an anomaly detection model according to an embodiment of the present application; [Figure 7] 1 is a diagram illustrating an example of a masking process according to an embodiment of the present disclosure; [Figure 8] 1 is a diagram illustrating a graph related to manufacturing information according to an embodiment of the present application; [Figure 9] 1 is a diagram illustrating a graph of an enhancement technique according to an embodiment of the present application; [Figure 10] 1 is a diagram illustrating a structural diagram of an enhancement technique according to an embodiment of the present application. [Figure 11] 1 is a schematic diagram illustrating an example of enhanced manufacturing information according to an embodiment of the present application. [Figure 12]1 is a diagram illustrating an example of visualization material relating to manufacturing quality anomalies according to an embodiment of the present disclosure; [Figure 13] 4 is an operational flowchart of a control method of an intelligent anomaly detection device according to an embodiment of the present disclosure. [Figure 14] 11 is an operational flowchart of a control method of an intelligent anomaly detection device according to another embodiment of the present disclosure. [Figure 15] 11 is an operational flowchart of a control method of an intelligent anomaly detection device according to another embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0028] Hereinafter, the embodiments of the present application will be described in detail with reference to the accompanying drawings so that a person skilled in the art to which the present application pertains can easily carry out the present application. However, the present application may be embodied in various different forms and is not limited to the embodiments described herein. In order to clearly explain the present application in the drawings, parts that are not related to the description are omitted, and similar parts are designated by similar reference numerals throughout the specification.

[0029] Throughout this specification, when a part is said to be "connected" to another part, this includes not only "directly connected" but also "electrically connected" or "indirectly connected" with another element therebetween.

[0030] Throughout this specification, when an element is referred to as being "on," "upper," "top," "upper end," "under," "lower," or "bottom" of another element, this includes not only when the element is in contact with the other element, but also when there is another element between the two elements.

[0031] Throughout this specification, when a part "comprises" a certain element, this means that it can further include other elements, rather than excluding other elements, unless specifically stated to the contrary.

[0032] The present application relates to an intelligent anomaly detection apparatus and a control method thereof, and more particularly, to an intelligent anomaly detection apparatus and a control method thereof based on explainable artificial intelligence (XAI) that utilizes manufacturing information.

[0033] The present application may have, but is not limited to, an objective of the present application to provide a machine learning-based anomaly detection model that selectively applies data refinement technology, normalization technology, and enhancement technology to sensing information generated by a sensor module installed in a target manufacturing equipment, extracts characteristic variables, and determines whether or not there is an abnormality in manufacturing quality based on the extracted characteristic variables.

[0034] Another object of the present application may be to provide an anomaly detection model based on explainable artificial intelligence that can interpret the cause of a malfunction or problem when it occurs by utilizing explainable artificial intelligence, but is not limited to this.

[0035] FIG. 1 is a schematic diagram of an intelligent anomaly detection system according to an embodiment of the present application.

[0036] Referring to FIG. 1, an intelligent anomaly detection system 1 (hereinafter also referred to as "this system 1") may include an intelligent anomaly detection device 100 (hereinafter also referred to as "this device 100"), a user terminal 200, an external server 300, and a target manufacturing equipment A.

[0037] According to the present system 1, the present device 100 can collect manufacturing information of the target manufacturing equipment A, generate a manufacturing quality abnormality degree based on the manufacturing information, and control the target manufacturing equipment A based on the manufacturing quality abnormality degree.

[0038] According to an embodiment of the present application, the apparatus 100 may provide a manufacturing information collection menu, a manufacturing quality generation menu, a quality abnormality interpretation menu, and a manufacturing equipment control menu to the user terminal 200. For example, the user terminal 200 may download and install an application program provided by the apparatus 100, and the manufacturing information collection menu, the manufacturing quality generation menu, the quality abnormality interpretation menu, and the manufacturing equipment control menu may be provided through the installed application.

[0039] The apparatus 100 can include any type of server, terminal, or device that transmits and receives data, content, and various communication signals to and from the user terminal 200, the external server 300, and the target manufacturing equipment A via a network, and has the function of storing and processing data.

[0040] The user terminal 200 is a device that interfaces with the present apparatus 100, the external server 300, and the target manufacturing equipment A through a network, and may be, for example, a smartphone, a smart pad, a tablet PC, a wearable device, etc., all kinds of wireless communication devices such as PCS (Personal Communication System), GSM (Global System for Mobile communication), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), and Wibro (Wireless Broadband Internet) terminals, and fixed terminals such as desktop computers and smart TVs.

[0041] The external server 300 is a server that is linked to the present apparatus 100, the user terminal 200, and the target manufacturing equipment A, and can receive and store or transmit information handled by the present system. The external server 300 can store manufacturing information including all information that can be generated in the target manufacturing equipment A, such as production volume, production results, production date, operation time, defective quantity, planned quantity, target quantity, operation rate, etc., and image information visualizing photos of products after a process is completed or vibration data through machine vision.

[0042] Examples of networks for sharing information between the present device 100, the user terminal 200, and the target manufacturing equipment A include, but are not limited to, a 3GPP (registered trademark) (3rd Generation Partnership Project) network, a LTE (Long Term Evolution) network, a 5G network, a WIMAX (registered trademark) (World Interoperability for Microwave Access) network, wired and wireless Internet, a LAN (Local Area Network), a Wireless LAN (Wireless Local Area Network), a WAN (Wide Area Network), a PAN (Personal Area Network), a Bluetooth (registered trademark) network, a Wifi network, an NFC (Near Field Communication) network, a satellite broadcasting network, an analog broadcasting network, a DMB (Digital Multimedia Broadcasting) network, and the like.

[0043] The target manufacturing facility A means a facility for producing the target product, and can be controlled by the present apparatus 100 .

[0044] FIG. 2 is a schematic diagram of an intelligent anomaly detection system 1 according to another embodiment of the present invention.

[0045] Referring to FIG. 2, the present system 1 includes a plurality of sensors attached to the target manufacturing equipment A, and is capable of sharing data based on a 5G edge computing infrastructure designed to monitor and even control the presence or absence of algorithm abnormalities.

[0046] The apparatus 100 can analyze and pre-process the manufacturing process of the target manufacturing equipment A using the collected manufacturing information, extract characteristic variables, and detect or predict the presence or absence of an abnormality in the manufacturing quality through the extracted characteristic variables. In addition, the apparatus 100 can output visualized content generated based on the cause of the problem when an abnormality occurs in the manufacturing quality to the user terminal 200, and perform equipment control suited to the environment of the target manufacturing equipment A to solve the cause of the problem.

[0047] FIG. 3 is a schematic block diagram of an intelligent anomaly detection device 100 according to an embodiment of the present application.

[0048] 3, the apparatus 100 may include a collection unit 110, an anomaly detection unit 120, and an interpretation generation unit 130. In addition, although not shown in FIG 3, the apparatus 100 may include a preprocessing unit, a variable extraction unit, a model generation unit, and a control unit.

[0049] According to an embodiment of the present application, the collection unit 110 can collect manufacturing information of the target manufacturing facility A.

[0050] For example, the collecting unit 110 may collect data (manufacturing information) generated in a manufacturing process (target manufacturing equipment A) for quality verification. The manufacturing information may include all information that can be generated in the target manufacturing equipment A, such as production volume, production results, production date, operation time, defective quantity, planned quantity, target quantity, operation rate, etc., and image information visualizing a photograph or vibration data of a product after the process is completed through machine vision, but is not limited thereto.

[0051] The collecting unit 110 can collect time series information including vibration, status, and operation rate of the target manufacturing equipment A as manufacturing information through a sensor module attached to the target manufacturing equipment A. The collecting unit 110 can collect image information such as front, back, and side of a completed product (manufactured product) after a process is completed through machine vision as manufacturing information. The manufacturing information can include time series information that is generated in the process of the target manufacturing equipment A on a time series basis and can be quantified, and image information capturing the state of a product or the state of a completed product during the process of the target manufacturing equipment A.

[0052] Furthermore, the manufacturing information may include manufacturing process information of the target manufacturing equipment A. The manufacturing information may include power, status, operation rate, etc. of the target manufacturing equipment A. The manufacturing information may include sensing information received from a sensor attached to the target manufacturing equipment A, and the sensing information received from the sensor attached to the target manufacturing equipment A may include time-series information that is generated in the manufacturing process of the target manufacturing equipment A and can be digitized, and the sensor attached to the target manufacturing equipment A may include various types of sensors such as optical sensors, ultrasonic sensors, and infrared sensors. The manufacturing information may include multi-angle image information of the product produced by the target manufacturing equipment A.

[0053] As another example, the collection unit 110 may include a pre-processing unit that pre-processes the manufacturing information. The collection unit 110 may also include a variable analysis unit that analyzes the manufacturing information, and a variable output unit that inputs the manufacturing information and the analysis result to a variable extraction artificial neural network and outputs characteristic variables, and the variable extraction artificial neural network may learn the manufacturing information, the manufacturing information analysis result, and the characteristic variables as a learning data set.

[0054] According to an embodiment of the present application, the anomaly detector 120 may generate a manufacturing quality anomaly level based on the manufacturing information.

[0055] For example, the anomaly detector 120 may input characteristic variables to an anomaly detection model to enable output of the degree of anomaly in manufacturing quality.

[0056] In addition, as an example, the present device 100 can generate a manufacturing quality abnormality level including at least one of information for detecting whether or not there is an abnormality in a manufactured product produced by the target manufacturing equipment A based on the manufacturing information, and information for predicting whether or not there is an abnormality in a manufactured product to be produced by the target manufacturing equipment A.

[0057] According to an embodiment of the present application, the interpretation generating unit 130 may generate interpretation information based on the degree of manufacturing quality abnormality.

[0058] As an example, the interpretation generation unit 130 may input image information corresponding to the degree of manufacturing quality abnormality output by the anomaly detection model described below into a previously generated explainable artificial intelligence model to generate interpretation information including visualization information regarding the interpretation of the degree of manufacturing quality abnormality.

[0059] FIG. 4 is a schematic block diagram of an intelligent anomaly detection device 100 according to another embodiment of the present application.

[0060] 4, the apparatus 100 may include a data collection unit 101, a data pre-processing unit 102, a data model learning unit 103, an explainable artificial intelligence model visualization unit 104, and an automatic process control system unit 105. The apparatus 100 may include a database (not shown) including big data. The database includes all information that may be generated by equipment, including, but not limited to, production volume, production results, production date, operation time, defective quantity, planned quantity, target quantity, operation rate, part code, momentary stop time, etc.

[0061] In addition, the automatic process control system unit 105 of the present apparatus 100 can construct infrastructure including heterogeneous equipment and process monitoring technology (not shown), complex event analysis processing technology (not shown), abnormality diagnosis monitoring alarm technology (not shown), and 5G edge computing infrastructure-based data sharing technology (not shown), etc., to perform automatic process control for the target manufacturing equipment A.

[0062] The heterogeneous equipment and process monitoring technology (not shown) can use a database that stores information in computer memory rather than disk storage to collect manufacturing information of the target manufacturing equipment A and create an environment that can process more than a preset number of events per hour (e.g., more than 500,000 events per second).

[0063] Complex event analysis processing technology (not shown) can quickly process manufacturing information of parallel processes of the target manufacturing equipment A by using complex event analysis processing technology to process large volumes of manufacturing information generated in real time by the target manufacturing equipment A.

[0064] The abnormality diagnosis monitoring alarm technology (not shown) activates an alarm and warning light function when an abnormality occurs in the production quality, and provides real-time notification functions (including text and e-mail) to the user terminals 200 of workers and managers. In addition, the device 100 can apply mobile web core technologies such as HTML5 to visually represent production information. That is, the device 100 can establish communication with the client server and provide web services (additional functions) without using other external (Active-X, Plug-in) functions.

[0065] A data sharing technology (not shown) based on 5G edge computing infrastructure refers to performing computing at or around the physical location of a user or data source, and sensing information generated during the manufacturing process of the target manufacturing equipment A is used for pre-processing and learning through the edge system, and the sensing information can be analyzed on a big data platform and then transmitted to the user terminal 200 to facilitate situational awareness.

[0066] According to an embodiment of the present application, when an abnormality exists in the manufacturing quality, the interpretation generating unit 130 may generate interpretation information including visualization information for a cause corresponding to the abnormality and explanation information corresponding to the visualization information.

[0067] As an example, when image information corresponding to the degree of manufacturing quality abnormality, which is output by the anomaly detection model described below and indicates that an abnormality exists in the manufacturing quality, is input to the explainable artificial intelligence model described above, the explainable artificial intelligence model is divided into a generation part that generates visualization information and an explanation part that generates explanation information that explains the generated visualization information. The explainable artificial intelligence model generates visualization information using at least one of the image object recognition algorithms YOLO algorithm, SSD algorithm, Faster R-CNN algorithm, Mask R-CNN algorithm, and RetinaNet algorithm, and at the same time generates explanation information in which the content of the visualization information is generated in words using an artificial intelligence algorithm having an encoder and a decoder built in.

[0068] In addition, the interpretation generating unit 130 may recognize an object from the image information or visualization information in which the explanation part of the explainable artificial intelligence model corresponds to the degree of abnormality in manufacturing quality, generate at least one word based on the recognized object, and generate explanation information by combining the recognized object and the generated word to interpret the content of the image information or the visualization information as a sentence. When an abnormality occurs in manufacturing, the present application provides interpretation information including the visualization information and explanation information to personnel related to the target manufacturing facility A, thereby providing an effect that even non-experts can easily understand the cause of the manufacturing problem.

[0069] According to an embodiment of the present disclosure, the pre-processing unit may perform pre-processing on the manufacturing information, including at least one of a refinement technique, a normalization technique, and an enhancement technique.

[0070] As an example, the preprocessing unit may perform preprocessing of the manufacturing information in a time series manner, whereby the preprocessing unit performs preprocessing of the manufacturing information in a time series manner, making it possible to check the state of the manufacturing equipment.

[0071] As another example, the preprocessing technique includes at least one of a refinement technique, a normalization technique, and an enhancement technique, and the present apparatus 100 may preprocess the manufacturing information by applying the preprocessing technique to the manufacturing information. If at least one of the refinement requirement degree, normalization requirement degree, and enhancement requirement degree in the manufacturing information is equal to or greater than a preset degree, the present apparatus 100 may selectively apply a preprocessing technique to the manufacturing information based on the corresponding requirement degree.

[0072] As another example, the apparatus 100 may pre-process the manufacturing information until the accuracy of the degree of abnormality in manufacturing quality output when the manufacturing information is input to the anomaly detection model is equal to or higher than a preset level.

[0073] FIG. 5 is a diagram illustrating time series information and image information generated in a manufacturing process according to an embodiment of the present application.

[0074] 5, the manufacturing information may include at least one of time series information as shown in Fig. 5(a) and image information as shown in Fig. 5(b), where the time series information may refer to information that has been processed in a time series.

[0075] According to an embodiment of the present application, if the manufacturing information is time-series information, the pre-processing unit may perform pre-processing on the time-series information, including at least one of distribution transformation correction, missing value correction, and abnormal value correction.

[0076] As an example, the preprocessing unit can perform preprocessing including at least one of distribution transformation correction, missing value correction, and outlier correction on manufacturing information including time series information, thereby generating a learning data set suitable for learning by an artificial intelligence model (anomaly detection model).

[0077] In addition, if the manufacturing information is formula information including time series information, the pre-processing unit may augment the manufacturing information using at least one of a plurality of augmentation techniques. Specifically, the formula information may include time series information. If the manufacturing information is processed in a time series or is time series information, it may be regarded as formula information. Therefore, if the manufacturing information is processed in a time series or is time series information, the pre-processing unit may augment the manufacturing information using at least one of a plurality of augmentation techniques described below.

[0078] Specifically, the plurality of augmentation techniques may include a first augmentation technique that adds noise to the manufacturing information to enhance it. If the manufacturing information is processed in a time series or is time series information, the present apparatus 100 may augment the manufacturing information using the first augmentation technique. Through the first augmentation technique, the present apparatus 100 may augment the manufacturing information by adding noise. By applying the first augmentation technique to the manufacturing information, robustness against noise is increased, and thus performance can be improved. The first augmentation technique may be implemented by randomly generating numbers using a Gaussian distribution and matching them with the manufacturing information. The first augmentation technique may include a jittering technique.

[0079] In addition, the plurality of augmentation techniques may include a second augmentation technique that augments a variable of the manufacturing information by applying a preset amount of magnitude change. If the manufacturing information is processed in a time series or is time series information and a label can be maintained even if the value of the manufacturing information changes by a certain magnitude, the apparatus 100 may augment the manufacturing information using the second augmentation technique. Through the second augmentation technique, the apparatus 100 may augment the manufacturing information by collectively applying a preset amount of magnitude change to each variable of the time series processed manufacturing information. The second augmentation technique may be implemented by performing multiplication using an arbitrary value for each variable of the manufacturing information. The second augmentation technique may include a scaling technique, data scaling, etc.

[0080] In addition, the plurality of augmentation techniques may include a third augmentation technique that converts the time point of the manufacturing information to augment it. If the manufacturing information is processed in a time series or is time series information and a preset change in the time point of the action does not affect the label to a preset degree, the present apparatus 100 may augment the manufacturing information using the third augmentation technique. Through the third augmentation technique, the present apparatus 100 may generate manufacturing information similar to the manufacturing information by converting the time point of the time series value of the manufacturing information to augment the manufacturing information. The third augmentation technique may be implemented by distorting the time interval between samples of the manufacturing information to a preset degree and changing the time position of the samples. The third augmentation technique may include a warping technique, time warping, etc.

[0081] According to an embodiment of the present disclosure, if the manufacturing information is image information, the pre-processing unit may perform pre-processing on the image information, including at least one of noise correction, size correction, and color correction.

[0082] As an example, the pre-processing unit may perform pre-processing including at least one of noise correction, size correction, and color correction on manufacturing information including image information, thereby generating a learning data set suitable for learning an artificial intelligence model (anomaly detection model).

[0083] According to an embodiment of the present disclosure, if the manufacturing information is image information, the pre-processing unit may generate processed image information based on the image information to enhance the image information.

[0084] As an example, if the manufacturing information is atypical information, the pre-processing unit may generate manufacturing information processed to satisfy a preset criticality based on the manufacturing information to enhance the manufacturing information. Specifically, the atypical information may include information that is not processed in time series. The atypical information may include image information. Thus, if the manufacturing information is not processed in time series or is image information, the pre-processing unit may generate manufacturing information processed to satisfy a preset criticality based on the manufacturing information to enhance the manufacturing information. The pre-processing unit may enhance the manufacturing information by applying image information of the manufactured product included in the manufacturing information to a GAN algorithm. Image information of various cases where anomalies may occur may be held by performing an enhancement technique such as inversion, cutting, or rotation on the image information included in the manufacturing information, or by applying a GAN algorithm to generate fake image information.

[0085] In addition, if the manufacturing information is non-standard information, the pre-processing unit can utilize the CycleGan algorithm to augment the manufacturing information.

[0086] According to an embodiment of the present application, the characteristic variable extraction unit may analyze manufacturing information based on pre-defined manufacturing quality verification elements.

[0087] For example, the characteristic variable extraction unit may analyze the entire manufacturing process based on the manufacturing information and perform an analysis on the finished product.

[0088] In addition, the characteristic variable extraction unit can grasp the manufacturing quality verification elements necessary for quality verification based on the manufacturing information, analyze the manufacturing information based on the manufacturing quality verification elements, and construct a learning dataset that can be applied to the anomaly detection model taking into account the manufacturing information and the analysis results.

[0089] In addition, the apparatus 100 can analyze the manufacturing information. Specifically, the apparatus 100 can analyze whether the manufacturing information is related to factors that induce the presence or absence of abnormality in manufacturing quality, which are preset in the manufacturing information. The manufacturing quality verification factors can include factors that induce the presence or absence of abnormality in manufacturing quality. Factors that induce the presence or absence of abnormality in manufacturing quality can include, but are not limited to, the assembly of defective parts during the manufacturing process, aging of equipment, rapid operation of manufacturing equipment, breakage due to pressure, irregular power supply, temperature drop, etc.

[0090] Furthermore, the present device 100 can analyze the manufacturing process of the target manufacturing equipment A based on the manufacturing information, and can analyze the products produced by the target manufacturing equipment A. Furthermore, the present device 100 can grasp whether or not the state of the target manufacturing equipment A is correlated with an abnormality in the produced products by performing a time-series analysis on the time-series processed manufacturing information.

[0091] According to an embodiment of the present application, the characteristic variable extraction unit may extract characteristic variables based on an analysis result of the manufacturing information.

[0092] For example, the characteristic variable extraction unit may extract a variable corresponding to information that is outside a preset critical range corresponding to a preset manufacturing quality verification element from among information in the manufacturing information as a characteristic variable. If all information included in the manufacturing information is within a preset critical range corresponding to the manufacturing quality verification element, the characteristic variable extraction unit may extract the characteristic variable as a preset value.

[0093] According to an embodiment of the present application, the characteristic variable extraction unit may input manufacturing information and analysis results to a variable extraction artificial neural network to output characteristic variables.

[0094] As an example, the present apparatus 100 may extract characteristic variables based on manufacturing information and time series analysis results, and input the manufacturing information and the time series analysis results to a variable extraction artificial neural network to output characteristic variables (statistical characteristic (feature) variables). The present apparatus 100 may extract meaningful variables for the characteristic variables. In this case, the variable extraction artificial neural network may be a principal component analysis algorithm, but is not limited thereto, and may apply various algorithms that have already been developed or will be developed in the future.

[0095] According to an embodiment of the present application, the variable extraction artificial neural network can learn manufacturing information, manufacturing information analysis results, and characteristic variables as a learning data set.

[0096] For example, the variable extraction artificial neural network can be generated through artificial intelligence-based learning such as machine learning and deep learning, but is not limited to these, and various neural network systems that have already been developed or will be developed in the future can be applied.

[0097] According to an embodiment of the present application, the model generating unit can generate a plurality of artificial neural networks trained using manufacturing information, characteristic variables, and manufactured product information as training data sets.

[0098] As an example, the model generating unit may generate a plurality of artificial neural networks that have learned abnormalities in manufacturing quality by having a machine learning algorithm learn the constructed learning data set. In addition, when time series information is input, the model generating unit may generate a plurality of artificial neural networks based on at least one of a Random Forest algorithm, an RNN algorithm, a CNN algorithm, an Autoencoder algorithm, a GAN algorithm, and a Transformer algorithm. Conversely, when image information or video information is input, the model generating unit may generate a plurality of artificial neural networks based on at least one of a CNN algorithm, a GAN algorithm, a PANDA algorithm, a Patchcore algorithm, and a DEVNET algorithm. The model generating unit may generate a plurality of artificial neural networks that determine the presence or absence of abnormalities in manufacturing quality based on a deep learning algorithm that receives image information (which may refer to video information) as an input.

[0099] As another example, the device 100 can generate multiple artificial neural networks based on at least one of a Random Forest algorithm, an RNN algorithm, a CNN algorithm, an Autoencoder algorithm, a GAN algorithm, and a Transformer algorithm, and can train manufacturing information, characteristic variables, and manufactured product (manufactured product or finished product) information as a training dataset.

[0100] The Random Forest algorithm is an algorithm in which many decision trees form a forest and average each prediction result into one result variable. The RNN algorithm is a type of artificial neural network that contains an internal cyclic structure and is used for time-dependent or sequential data learning. The cyclic structure allows information expression by accumulating previous information in the current information. The CNN algorithm is an artificial neural network that learns directly from data without the need to manually extract features, and is useful when searching for patterns for face, object, and scene recognition in videos or images. The Autoencoder algorithm compresses the input data, shrinks it, and then expands it to make the result data the same as the input data. The GAN algorithm is an artificial neural network in which the generator and the classifier compete with each other and learn by gradually improving each other's performance. The Transformer algorithm is a neural network that complements the existing RNN algorithm, processes sequences in parallel at once, and conveys which parts are important to reflect information.

[0101] In other words, the CNN algorithm uses neural networks to learn patterns and features of image data to identify abnormal images and detect anomalies. The GAN algorithm uses a competitive neural network consisting of a generator and a discriminator to learn normal image distributions and detect anomalies in them. The PANDA algorithm and Patchcore algorithm are algorithms that combine reinforcement learning and neural networks to detect and classify anomalies in image data. However, the difference between the two models is the way the data is processed and calculated internally. The DEVNET algorithm is a deep learning algorithm that combines neural networks and time-series data to detect image anomalies.

[0102] According to an embodiment of the present application, the model generating unit may set an artificial neural network having the highest accuracy of outputting a degree of abnormality in manufacturing quality among a plurality of artificial neural networks as an anomaly detection model.

[0103] As an example, the apparatus 100 may set an artificial neural network having the highest accuracy of the output level of the manufacturing quality anomaly among a plurality of artificial neural networks generated by applying a plurality of machine learning algorithms as an anomaly detection model. The anomaly detection model may have a purpose of detecting and predicting the manufacturing quality anomaly and accurately classifying the quality process. That is, the apparatus 100 may generate an artificial neural network by applying each of a Random Forest algorithm, an RNN algorithm, a CNN algorithm, an Autoencoder algorithm, a GAN algorithm, and a Transformer algorithm, and set an artificial neural network having the highest accuracy of the output level of the manufacturing quality anomaly as an anomaly detection model for detecting and predicting the manufacturing quality anomaly.

[0104] In addition, the model generation unit can select the most accurate model among multiple artificial neural networks generated by applying multiple deep learning algorithms as a model for accurately classifying the quality process. In other words, the model generation unit can generate multiple artificial intelligence models that predict manufacturing defects by applying each of the Random Forest algorithm, RNN algorithm, CNN algorithm, Autoencoder algorithm, GAN algorithm, Transformer algorithm, CNN algorithm, GAN algorithm, PANDA algorithm, Patchcore algorithm, and DEVNET algorithm, and compare the accuracy to select the artificial intelligence model suitable for on-site use as the ideal detection model.

[0105] According to an embodiment of the present application, the anomaly detector 120 may input characteristic variables to an anomaly detection model and output an anomaly level of manufacturing quality.

[0106] For example, the anomaly detection unit may input characteristic variables to an anomaly detection model and output a manufacturing quality anomaly level regarding whether or not there is an anomaly in the manufacturing quality, or image information corresponding to the manufacturing quality anomaly level.

[0107] In addition, when an abnormality is detected in a manufactured product or a future occurrence of an abnormality is predicted based on the degree of abnormality in the manufacturing quality, the device 100 can transmit the characteristic variable corresponding to the abnormality and the manufacturing process associated with the characteristic variable to the user terminal 200.

[0108] According to one embodiment of the present application, if the degree of manufacturing quality anomaly is equal to or greater than a preset critical anomaly level, the anomaly detection unit 120 can apply at least one of a plurality of analysis techniques to the degree of manufacturing quality anomaly to calculate the process causing the problem.

[0109] As an example, if the number or proportion of objects corresponding to a pre-set abnormality contained in image information corresponding to the degree of manufacturing quality abnormality is equal to or greater than a critical abnormality degree (meaning a critical quantity or critical proportion, respectively), the anomaly detection unit can apply at least one of a plurality of analysis techniques to the degree of manufacturing quality abnormality to calculate the cause process of the problem within the target manufacturing equipment A.

[0110] In addition, if the degree of manufacturing quality abnormality is equal to or higher than a preset critical abnormality degree, the device 100 can apply at least one of the SHAP technique, the Anchors technique, the LIME technique, the Counterfactual instances technique, and the ICE technique to the degree of manufacturing quality abnormality to generate an explainable artificial intelligence model, and calculate the cause process of the problem using the explainable artificial intelligence model.

[0111] The SHAP technique is an artificial intelligence technique that analyzes and visualizes the relationship between input variables and model result values ​​by calculating a SHAP value for each input variable. The Anchors technique is a technique that captures predictions and generates local explanations when changes in other characteristic values ​​do not affect the prediction. The LIME technique is an algorithm that provides a new technique to explain the results of all predictive models in an interpretable and reliable manner. The Counterfactual Explanations technique is a technique that explains causal situations in the manner that "if situation A had not occurred, situation B would not have occurred." The ICE technique is a technique that visualizes how output values ​​change due to changes in input values, allowing you to interpret which input values ​​have a significant impact on the target.

[0112] FIG. 6 is a diagram illustrating an image corresponding to an anomaly output by an anomaly detection model according to an embodiment of the present application.

[0113] Referring to FIG. 6, the present device uses an anomaly detection model, which is a deep learning model, to determine that an anomaly exists in the manufacturing information if there is a difference between the manufacturing information and a predicted heat map and a predicted mask that is greater than a preset level, and can output an image (segmentation result) corresponding to the existence of the anomaly.

[0114] FIG. 7 is a diagram illustrating an example of a masking process according to an embodiment of the present disclosure.

[0115] Referring to Figure 7, when image information corresponding to the degree of manufacturing quality abnormality that is output by the anomaly detection model indicating the presence of an abnormality in the manufacturing information is input to an explainable artificial intelligence, the interpretation generation unit 130 recognizes objects that are determined to be problematic in the image information corresponding to the degree of manufacturing quality abnormality, and generates visualization information by masking the recognized objects as shown in Figure 7a.

[0116] According to an embodiment of the present application, the interpretation generating unit 130 may generate interpretation information including visualization information and explanation information based on the problem cause process.

[0117] As an example, the interpretation generating unit 130 can generate visualization information based on the problem cause process. Specifically, the present device 100 can generate visualization information based on the problem cause process calculated using an explainable artificial intelligence model. When the present device 100 determines that an abnormality exists in a manufactured product after the manufacturing process is completed, the present device 100 can provide a visualization of which part of the manufacturing process of the target manufacturing equipment caused the abnormality, making it easy for non-experts to understand and trust.

[0118] As another example, when the anomaly detection model of the above-mentioned machine learning algorithm determines that an anomaly has occurred in manufacturing, the interpretation generation unit 130 can confirm and interpret the cause through an explainable artificial intelligence model.

[0119] In addition, the interpretation generation unit 130 includes an explainable artificial intelligence model that is roughly divided into a generation part and an explanation part, in which the generation part generates object words that are characterized in image information corresponding to the degree of abnormality in manufacturing quality input using a deep learning algorithm based on CNN and RNN, and in which the explanation part generates explanatory information that explains the input image information in text using the generated words. In addition, in the generation part step, the user can simultaneously recognize an object (causing anomaly) desired by the user in the image information through an object recognition model, and perform bounding box and masking processing to generate visualization information.

[0120] In addition, when the anomaly detection model determines that there is an abnormality in the image information or video information included in the manufacturing information, the interpretation generation unit 130 provides a visualization of which part the cause is, making it easy for non-experts to understand and trust the results of the artificial intelligence.

[0121] According to an embodiment of the present application, the control unit can control the target manufacturing equipment A based on the degree of abnormality in the manufacturing quality.

[0122] As an example, based on the degree of abnormality in manufacturing quality, the apparatus 100 can control the manufacturing process included in the target manufacturing equipment A so that the manufactured products produced by the target manufacturing equipment A are free of abnormalities and the manufactured products to be produced do not contain abnormalities.

[0123] According to an embodiment of the present application, the control unit can control the target manufacturing equipment A based on the problem-causing process.

[0124] As an example, when the device 100 determines that there is an abnormality in a manufactured product after the manufacturing process has been completed, the control unit can transmit the process causing the problem to the user terminal 200 of a pre-configured manager, and control the target manufacturing equipment A based on the user input entered into the user terminal 200.

[0125] As another example, the control unit may automatically perform process control corresponding to the problem-causing process output based on a preset manual.

[0126] FIG. 8 is a diagram illustrating a graph related to manufacturing information according to an embodiment of the present application.

[0127] Referring to FIG. 8, (a) may represent a graph corresponding to a normal power pattern included in the manufacturing information, (c) may represent a graph corresponding to an abnormal power pattern included in the manufacturing information, and (b) and (d) may represent graphs corresponding to time-series processed manufacturing information or time-series information included in the manufacturing information.

[0128] FIG. 9 is a diagram that illustrates a graph of an enhancement technique according to an embodiment of the present application.

[0129] 9, it can be seen that the original graph is a graph of manufacturing information that has been subjected to time series processing without being augmented. Here, it can be seen that when the jittering technique is applied to the manufacturing information corresponding to the original graph (a) to be augmented, the manufacturing information is augmented as shown in the jittering graph (b). In addition, it can be seen that when the scaling technique is applied to the manufacturing information corresponding to the original graph (a) to be augmented, the manufacturing information is augmented as shown in the scaling graph (c). In addition, it can be seen that when the time warping technique is applied to the manufacturing information corresponding to the original graph (a) to be augmented, the manufacturing information is augmented as shown in the time warping graph (d).

[0130] FIG. 10 is a diagram illustrating a structural diagram of an enhancement technique according to one embodiment of the present application.

[0131] Referring to FIG. 10, it can be seen that this is a structural diagram regarding the case where the CycleGan algorithm is applied to manufacturing information, which is non-standard information, to increase it. Whereas the existing GAN algorithm is composed of one generator and discriminator, CycleGAN can be composed of two generators and two discriminators. Here, (a) refers to the first generator and can generate image information of the first domain as image information of the second domain. (b) refers to the first discriminator and can judge the authenticity of the image (fake) generated by the first generator. (c) refers to the second generator and can convert the image information of the second domain back to image information of the first domain. Also, (d) refers to the second discriminator and can judge the authenticity of the image information generated by the second generator.

[0132] FIG. 11 is a schematic diagram illustrating an example of enhanced manufacturing information according to an embodiment of the present application.

[0133] 11, it can be seen that the image information, which is non-standard information included in the manufacturing information, is enhanced through a pre-processing process. The image information included in the manufacturing information may be image information of a verification stage after the manufacturing process of the target manufacturing equipment A is completed.

[0134] FIG. 12 is a schematic diagram illustrating an example of visualization material related to manufacturing quality anomalies according to an embodiment of the present disclosure.

[0135] Referring to FIG. 12, it can be seen that the cause of the problem at the point in time when it is determined to be abnormal based on the characteristic variable, temperature A, can be visualized as shown in (a).

[0136] The following briefly describes the operation flow of the present application based on the above detailed description.

[0137] FIG. 13 is a flowchart illustrating an operation of a control method of the intelligent anomaly detection device 100 according to an embodiment of the present application.

[0138] Referring to FIG. 13, the control method of the intelligent anomaly detection device 100 may include steps S110 to S130.

[0139] In step S110, the apparatus 100 can collect manufacturing information of the target manufacturing facility A.

[0140] Next, in step S120, the apparatus 100 can generate a manufacturing quality abnormality level based on the manufacturing information.

[0141] Next, in step S130, the apparatus 100 can generate interpretation information based on the degree of abnormality in the manufacturing quality.

[0142] FIG. 14 is a flowchart illustrating an operation of a control method of the intelligent anomaly detection device 100 according to another embodiment of the present application.

[0143] Referring to FIG. 14, the apparatus 100 can perform steps S101 to S109.

[0144] FIG. 15 is a flowchart illustrating an operation of a control method of the intelligent anomaly detection device 100 according to another embodiment of the present application.

[0145] Referring to FIG. 15, the apparatus 100 can perform steps S1001 to S1005.

[0146] 13 to 15 may be performed by the above-described intelligent anomaly detection apparatus 100. Therefore, even if content is omitted below, the content described about the intelligent anomaly detection apparatus 100 may be equally applied to the description of the control method of the intelligent anomaly detection apparatus 100.

[0147] In the above description, steps S110 to S130, S101 to S109, and S1001 to S1005 may be further divided into additional steps or combined into fewer steps depending on the embodiment of the present application. Also, some steps may be omitted or the order between steps may be changed as necessary.

[0148] The control method of the intelligent anomaly detection device 100 according to an embodiment of the present application may be embodied in the form of program instructions that can be executed by various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specifically designed and constructed for the present invention, or may be those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specifically configured to store and execute program instructions, such as ROMs, RAMs, flash memories, and the like. Examples of program instructions include not only machine language codes such as those created by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

[0149] In addition, the above-described method for controlling the intelligent anomaly detection apparatus 100 may be embodied in the form of a computer program or application stored in a recording medium and executed by a computer.

[0150] The above description of the present application is for illustrative purposes only, and a person having ordinary skill in the art to which the present application pertains will understand that the present application can be easily modified into other specific forms without changing the technical idea or essential features of the present application. Therefore, it should be understood that the above-described embodiments are illustrative in all respects and are not limiting. For example, each component described as a single type may be implemented in a distributed form, and similarly, each component described as a distributed type may be implemented in a combined form.

[0151] The scope of the present application is defined by the claims set forth below rather than the above detailed description, and all modifications and variations derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present application. [Explanation of symbols]

[0152] 1: Intelligent Anomaly Detection System 100: Intelligent Anomaly Detection Device 110: Collection Department 120: Anomaly Detection Department 130: Interpretation generation unit 200: User terminal 300: External server A: Target manufacturing equipment

Claims

1. In an intelligent anomaly detection device, a collection unit for collecting manufacturing information, which is data for verifying the quality of products generated in a manufacturing process, acquired through a sensor module installed in a target manufacturing facility, the manufacturing information including at least one of a production volume, a production record, a production date, a working time, a defective quantity, a planned quantity, a target quantity, and an operating rate, and image information visualizing a photograph or vibration data of a product after the process is completed through a machine vision; an anomaly detection unit that uses manufacturing information as input, and outputs manufacturing quality anomaly information and image information corresponding to the manufacturing quality anomaly information, the manufacturing quality anomaly information including at least one of information for detecting whether the product produced by the target manufacturing facility has an anomaly and information for predicting whether the product to be produced by the target manufacturing facility has an anomaly, by using an anomaly detection model machine-learned to output the manufacturing quality anomaly information and image information corresponding to the manufacturing quality anomaly information corresponding to the manufacturing information collected by the collection unit; an interpretation generation unit that generates interpretation information corresponding to the manufacturing quality anomaly information output by the anomaly detection unit and the image information corresponding to the manufacturing quality anomaly information output by the anomaly detection unit, using an artificial intelligence model that has been trained to receive as input manufacturing quality anomaly information, which is learning data, and image information corresponding to the manufacturing quality anomaly information, which is learning data, and output interpretation information including visualization information for a cause corresponding to the anomaly and explanatory information corresponding to the visualization information; Including, Intelligent anomaly detection device.

2. a pre-processing unit that performs pre-processing on the manufacturing information, the pre-processing including at least one of a refinement technique, a normalization technique, and an enhancement technique; Further comprising:

2. The intelligent anomaly detection device according to claim 1.

3. The pre-treatment unit includes: If the manufacturing information is time-series information, a pre-processing is performed on the time-series information, the pre-processing including at least one correction selected from the group consisting of distribution transformation correction, missing value correction, and outlier correction; If the manufacturing information is image information, a pre-processing step is performed on the image information, which includes at least one of noise correction, size correction, and color correction, to generate image information that is detected or predicted to have an abnormality in the product.

3. The intelligent anomaly detection device according to claim 2.

4. A characteristic variable extraction unit which performs a time series analysis of the manufacturing information based on manufacturing quality verification elements which indicate factors which induce the presence or absence of an abnormality in the manufacturing quality, determines whether the state of the target manufacturing equipment is correlated with an abnormality in the manufactured product produced, and extracts characteristic variables which correspond to information contained in the manufacturing information and which exist outside the critical range corresponding to the manufacturing quality verification elements based on the determination results for the manufacturing information, and outputs the characteristic variables to the anomaly detection model. Further comprising: The anomaly detection model is trained using the manufacturing information and the characteristic variables as a training data set. An intelligent anomaly detection device according to any one of claims 1 to 3.

5. The characteristic variable extraction unit inputting the manufacturing information and the identification result into a variable extraction artificial neural network to output the characteristic variables; The variable extraction artificial neural network comprises: The manufacturing information, the identification result, and the characteristic variable are learned as a learning data set.

5. An intelligent anomaly detection device according to claim 4.

6. a model generation unit that generates a plurality of artificial neural networks that are trained using the manufacturing information, the characteristic variables, and manufactured product information as a learning data set, and sets an artificial neural network that has the highest accuracy of outputting the manufacturing quality anomaly information among the plurality of artificial neural networks as the anomaly detection model; It includes, 6. An intelligent anomaly detection device according to claim 5.

7. The abnormality detection unit is inputting the characteristic variables into the anomaly detection model to output the manufacturing quality anomaly information; 7. An intelligent anomaly detection device according to claim 6.

8. The abnormality detection unit is When the manufacturing quality abnormality information indicates information that is outside the critical range, a problem cause process is calculated by applying at least one of a plurality of analysis techniques to the manufacturing quality abnormality information.

8. An intelligent anomaly detection device according to claim 7.

9. The interpretation generating unit is generating the interpretation information including the visualization information and the explanation information based on the problem cause process; 9. The intelligent anomaly detection device according to claim 8.

10. A control unit that controls the target manufacturing equipment based on the manufacturing quality abnormality information; Further comprising:

10. The intelligent anomaly detection device according to claim 9.

11. The control unit is The target manufacturing equipment is controlled based on the problem cause process.

11. The intelligent anomaly detection device according to claim 10.

12. A method for controlling an intelligent anomaly detection device, comprising: a step of collecting manufacturing information, which is data for verifying the quality of products generated in a manufacturing process, acquired through a sensor module installed in a target manufacturing facility, the manufacturing information including at least one of a production volume, a production record, a production date, a working time, a defective quantity, a planned quantity, a target quantity, and an operating rate, and image information visualizing a photograph or vibration data of a product after the process is completed through a machine vision; outputting, using an anomaly detection model machine-learned using manufacturing information as input, manufacturing quality anomaly information including at least one of information for detecting whether the product produced by the target manufacturing facility has an anomaly and information for predicting whether the product to be produced by the target manufacturing facility has an anomaly, and image information corresponding to the manufacturing quality anomaly information as output, the manufacturing quality anomaly information corresponding to the manufacturing information collected in the collecting step, and the image information corresponding to the manufacturing quality anomaly information; generating interpretation information corresponding to the manufacturing quality anomaly information output in the outputting step and the image information corresponding to the manufacturing quality anomaly information output in the outputting step, using an artificial intelligence model trained to receive manufacturing quality anomaly information as learning data and image information corresponding to the manufacturing quality anomaly information as learning data, and output interpretation information including visualization information for a cause corresponding to the anomaly and explanatory information corresponding to the visualization information; Including, A method for controlling an intelligent anomaly detection device.

13. A computer-readable recording medium having recorded thereon a program for causing a computer to execute the method of claim 12.

Images