Diagnostic system using artificial intelligence model and method of operating same

Abstract

The diagnostic system may include: a first device configured to train a machine learning-based diagnostic model using a training image; and a second device configured to determine whether the diagnostic target is defective using the diagnostic model provided from the first device and the inferred image of the diagnostic target. Here, the second device may determine whether the inference image is a target for retraining based on whether one or more predefined conditions are satisfied, and the first device may store an original of the inference image that has been determined by the second device as the target for retraining in a predefined storage space.

Description

Technical Field

[0001] This application claims priority and benefit to Korean Patent Application No. 10-2024-0090819, filed on July 10, 2024, with the Korean Intellectual Property Office, the entire contents of which are incorporated herein by reference.

[0002] This invention relates to a diagnostic method, a diagnostic system, and an operating method thereof for detecting abnormalities, and more specifically, to a diagnostic system and an operating method thereof for using an artificial intelligence-based diagnostic model to diagnose a diagnostic target. Background Technology

[0003] Rechargeable and reusable secondary batteries can be used as power sources for small devices such as mobile phones, tablet PCs, and vacuum cleaners, as well as for medium and large-sized devices such as personal mobility devices, automobiles, and energy storage systems (ESS) for smart grids.

[0004] Depending on system requirements, secondary batteries are used in the form of components—such as battery modules in which multiple battery cells are connected in series and parallel, or battery packs in which battery modules are connected in series and parallel.

[0005] Manufacturing battery cells or battery modules requires various processes. Automated equipment can be installed on the production line to perform each process, and the seamless connection of these automated devices enables the mass production of battery cells or battery modules.

[0006] To ensure product quality meets required standards, diagnostics are performed at critical processing points on the production line. This is typically achieved using a video-based monitoring system (Factory Monitoring Visual System, FMVS) that utilizes cameras installed at these critical processing points.

[0007] Recently, FMVS (Frequency-Based Visualization System) utilizing artificial intelligence (AI) models has been proposed to improve the diagnostic accuracy of monitoring systems. These diagnostic systems input images of the target for diagnosis into an AI-based diagnostic model and determine whether the target has defects based on the results output by the diagnostic model.

[0008] To further improve the diagnostic accuracy of these diagnostic systems, the diagnostic model must be updated. This can be achieved by retraining the diagnostic model using the original images used in the diagnostic process as training data.

[0009] However, since the raw images collected during the diagnostic process are high-resolution, high-volume data, building a storage library to store these large numbers of raw images is very costly.

[0010] Among the prior art documents related to this invention, KR 10-2018-0009563 A is somewhat relevant. Summary of the Invention

[0011] [Technical Issues]

[0012] To eliminate one or more problems of related technologies, embodiments of this disclosure provide a diagnostic system using an artificial intelligence (AI) model.

[0013] To eliminate one or more problems in the related technologies, embodiments of this disclosure also provide a method of operating a diagnostic system.

[0014] [Technical Solution]

[0015] To achieve the objectives of this disclosure, the diagnostic system may include: a first device configured to train a machine learning-based diagnostic model using training images; and a second device configured to determine whether a diagnostic target is defective using the diagnostic model provided by the first device and an inferred image of the diagnostic target. Here, the second device may determine whether an inferred image is a target for retraining based on whether one or more predefined conditions are met, and the first device may store the original inferred image of a target already determined by the second device for retraining in a predefined storage space.

[0016] The first device can compress inference images that are determined not to be targets for retraining according to a predefined image processing procedure, and store the compressed inference images separately from the original inference images.

[0017] The first device includes a storage device for storing the inferred image. Here, the storage device may include: a first storage space for storing a first image, which is the original of the inferred image; and a second storage space for storing a second image, which is a converted image of the original with lower capacity or quality.

[0018] The first device can use multiple first images stored in the first storage space as training data to retrain the diagnostic model.

[0019] The first device can reconstruct at least some of the second images stored in the second storage space according to a predefined image processing procedure, and use the reconstructed images to retrain the diagnostic model.

[0020] The first device can convert a first image that has been stored for a preset period of time into a second image from among a plurality of first images stored in the first storage space, and store the second image in the second storage space.

[0021] The second device can identify the inferred image as the target for retraining if two conditions are met: a first condition in which the inferred image is determined to be defective by the diagnostic model; and a second condition in which the inferred image is determined to be defective by a defect detection model defined separately from the diagnostic model.

[0022] Even if the first condition is not met, if the third condition is met, the second device can identify the inferred image as the target for retraining, in which the abnormal region is detected by the anomaly detection model, which is defined separately from the diagnostic model and the defect detection model.

[0023] According to another embodiment of this disclosure, a method of operating a diagnostic system is provided. The diagnostic system includes: a first device configured to train a machine learning-based diagnostic model using training images; and a second device configured to determine whether a diagnostic target is defective using the diagnostic model provided by the first device and an inferred image of a diagnostic target. The method may include the second device determining whether the inferred image is a target for retraining based on whether one or more predefined conditions are met; and the first device storing the original of the inferred image that has been determined by the second device to be a target for retraining in a predefined storage space.

[0024] The operation method of the diagnostic system may further include: by a first device, compressing an inferred image determined not to be a target for retraining according to a predefined image processing procedure; and by the first device, storing the compressed image of the inferred image separately from the original inferred image.

[0025] The first device may include a storage device for storing an inferred image, and the storage device may include: a first storage space for storing a first image, which is the original of the inferred image; and a second storage space for storing a second image, which is a converted image of the original with lower capacity or quality.

[0026] The operation method of the diagnostic system may further include: retraining the diagnostic model by the first device using multiple first images stored in the first storage space as training data.

[0027] The operation method of the diagnostic system may further include: reconstructing at least some of the second images stored in the second storage space by the first device according to a predefined image processing procedure; and retraining the diagnostic model by the first device using the reconstructed images.

[0028] The operation method of the diagnostic system may further include: the first device converting a first image that has been stored for a preset period of time into a second image from among a plurality of first images stored in a first storage space, and storing the second image in a second storage space.

[0029] Determining whether an inferred image is a target for retraining may include: identifying an inferred image as a target for retraining if two conditions are met, including: a first condition in which the inferred image is determined to be defective by a diagnostic model; and a second condition in which the inferred image is determined to be defective by a defect detection model defined separately from the diagnostic model.

[0030] Determining whether an inferred image is a target for retraining may include: even if the first condition is not met, if the third condition is met, the inferred image is identified as a target for retraining, in which an anomaly detection model, which is defined separately from the diagnostic model and the defect detection model, detects an anomaly region.

[0031] [Beneficial Effects]

[0032] According to embodiments of this disclosure, the diagnostic accuracy of a diagnostic system can be improved, and the cost of constructing a storage device included in the diagnostic system can be minimized. Attached Figure Description

[0033] Figure 1 This is a block diagram of a factory monitoring system according to an embodiment of the present invention.

[0034] Figure 2 This is a block diagram of a diagnostic system according to an embodiment of the present invention.

[0035] Figure 3 This is a flowchart illustrating a method for operating a diagnostic system using a second device according to an embodiment of the present invention.

[0036] Figure 4 This is a flowchart illustrating a method for operating a diagnostic system using a second device according to another embodiment of the present invention.

[0037] Figure 5 This is a flowchart illustrating a method for operating a diagnostic system using a first device according to an embodiment of the present invention.

[0038] Figure 6 This is a flowchart illustrating a method for operating a diagnostic system using a first device according to another embodiment of the present invention.

[0039] Figure 7 This is a block diagram of a first device according to an embodiment of the present invention.

[0040] Figure 8 This is a block diagram of a second device according to an embodiment of the present invention.

[0041] 10: Diagnostic Goals

[0042] 20: Image sensor

[0043] 30: Diagnostic device

[0044] 40: Diagnostic model training device

[0045] 100, 700: First equipment

[0046] 200, 800: Second equipment Detailed Implementation

[0047] This invention can be modified in various forms and has various embodiments, and specific embodiments thereof are shown by way of example in the accompanying drawings and will be described in detail below. However, it should be understood that the invention is not intended to be limited to the specific embodiments, but rather, the invention is intended to cover all modifications, equivalents, and substitutions falling within the spirit and scope of the invention. Throughout the description of the accompanying drawings, the same reference numerals refer to the same elements.

[0048] It will be understood that although terms such as first, second, A, B, etc., may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used only to distinguish one element from another. For example, a first element may be referred to as a second element without departing from the scope of the invention, and similarly, a second element may be referred to as a first element. As used herein, the term "and / or" includes a combination of a plurality of associated listed items or any one of a plurality of associated listed items.

[0049] It will be understood that when a component is described as "coupled" or "connected" to another component, it can be directly coupled or connected to the other component, or there may be intermediate components. Conversely, when a component is described as "directly coupled" or "directly connected" to another component, there are no intermediate components.

[0050] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It will be further understood that the terms “comprising,” “including,” “containing,” “covering,” and / or “having” as used herein specify the presence of stated features, integers, steps, operations, constituent elements, components, and / or combinations thereof, but do not preclude the presence or addition of one or more other features, integers, steps, operations, constituent elements, components, and / or combinations thereof.

[0051] Unless otherwise defined, all terms used herein, including technical and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It will also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant field, and will not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0052] Various embodiments of the invention will be described in detail below with reference to the accompanying drawings. The description will also illustrate, by way of example, the diagnostic system according to the invention installed in a battery manufacturing or assembly plant to diagnose batteries or battery processing equipment. However, the scope of the invention is not limited to such testing targets.

[0053] Figure 1 This is a block diagram of a factory monitoring system according to an embodiment of the present invention.

[0054] refer to Figure 1 The factory monitoring system according to an embodiment of the present invention may include an image sensor 20, a diagnostic device 30, and a diagnostic model learning device 40.

[0055] Image sensor 20 is a device that generates image data of one or more diagnostic targets 10 located within the factory. For example, image sensor 20 may correspond to a camera fixed in a specific location.

[0056] The image sensor 20 can be connected to the diagnostic device 30 via a network and transmit the image data of the diagnostic target 10 to the diagnostic device 30.

[0057] Image sensor 20 can generate image data of a specific area of ​​diagnostic target 10. For example, image sensor 20 can be mounted at a specific location in an assembly facility and configured to capture the fastening area of ​​a specific bolt. As another example, image sensor 20 can be mounted near an electrolyte injection device and configured to capture the interior of the electrolyte injection device.

[0058] The image sensor 20 can generate image data of the diagnostic target 10 at preset time intervals and transmit the image data to the diagnostic device 30. For example, the image sensor 20 can generate 60 image data per second (60 frames per second) and transmit the image data to the diagnostic device 30.

[0059] The factory monitoring system may include multiple image sensors 20. Each image sensor can be positioned at a preset location, generate image data of its corresponding diagnostic target, and transmit the generated image data to the diagnostic device 30.

[0060] The diagnostic device 30 can receive image data from the image sensor 20 and diagnose whether the target has an anomaly based on the image data. Here, the diagnostic device 30 for detecting anomalies can use the image data received from the image sensor 20 and a predefined diagnostic model to determine whether the target has a defect. For example, the diagnostic device 30 can determine whether there is a defect such as a bolt tightening failure, a single component failure, or a single specification failure.

[0061] The diagnostic model learning device 40 can be a computing device that generates a diagnostic model and provides the diagnostic model to the diagnostic device 30. Here, the diagnostic model learning device 40 can use learning data to train a machine learning-based diagnostic model and provide the trained diagnostic model to the diagnostic device 30.

[0062] The diagnostic device 30 can input one or more images of the diagnostic target transmitted from the image sensor into a diagnostic model provided by the diagnostic model learning device 40, and determine whether the diagnostic target has a defect based on the inference results output by the diagnostic model. In other words, the diagnostic device 30 can use the inferred image received from the image sensor 20 and the trained diagnostic model provided by the diagnostic model learning device 40 to perform a diagnosis on the diagnostic target.

[0063] The diagnostic model learning device 40 can store the inferred image in a storage device. Here, the diagnostic model learning device 40 can store the original inferred image and the compressed image of the inferred image in the storage device respectively.

[0064] The diagnostic model learning device 40 can retrain the diagnostic model after a preset time period or upon user request. Here, the diagnostic model learning device 40 can use images stored in a storage device to retrain the diagnostic model. Upon completion of retraining, the diagnostic model learning device 40 can provide the retrained diagnostic model to the diagnostic device 30.

[0065] Figure 2 This is a block diagram of a diagnostic system according to an embodiment of the present invention.

[0066] refer to Figure 2 The diagnostic system according to an embodiment of the present invention may include a first device 100 and a second device 200. Here, the first device 100 may correspond to... Figure 1 The diagnostic model learning device 40, and the second device 200 can correspond to Figure 1 Diagnostic device 30.

[0067] The first device 100 can train a machine learning-based diagnostic model 110 for diagnosing diagnostic targets and provide the trained diagnostic model to the second device 200.

[0068] When an image of the diagnostic target is received as input data, a diagnostic model 110 can be trained to output whether the diagnostic target is defective (normal or defective) as output data.

[0069] The first device 100 may include a storage device 120 in which training images are stored, and the training images stored in the storage device 120 are used to train the diagnostic model 110.

[0070] The second device 200 can use the diagnostic model 210 received from the first device 100 and the inferred image of the diagnostic target to determine whether the diagnostic target is defective. For example, the second device 200 can receive an image of the diagnostic target generated by an image sensor as an inferred image and input the inferred image into the diagnostic model 210. Here, the second device 200 can determine whether the diagnostic target is defective based on the inference result output by the diagnostic model 210.

[0071] The second device 200 can determine whether the inferred image is a target for retraining based on whether predefined conditions are met.

[0072] For example, if the inference result of the diagnostic model 210 is [fault], the second device 200 can determine the corresponding inferred image as the target for retraining.

[0073] For example, if the inference result of the diagnostic model 210 is [fault], and the inference result of the machine learning-based defect detection model defined separately from the diagnostic model is also [fault], then the second device 200 can determine the corresponding inferred image as the target for retraining.

[0074] As another example, even if the inference result of the diagnostic model 210 is [normal], if the inference result of the machine learning-based anomaly detection model defined separately from the diagnostic model is [anomaly region detection], then the second device 200 can identify the corresponding inferred image as the target for retraining.

[0075] The first device 100 can store the inferred image in the storage device 120. Here, the first device 100 can receive the inferred image from the second device 200 or from the image sensor.

[0076] The first device 100 can store the inferred image in its original form in the storage device 120, or it can use a predefined image processing procedure to convert the inferred image into a low-capacity or low-quality image and then store the converted image in the storage device 120.

[0077] Here, if a specific inferred image is determined by the second device 200 to be a retraining target, the first device 100 may store the original inferred image in a storage device. Furthermore, if the second device 200 determines that a specific inferred image is not a retraining target, the first device 100 may use a predefined image processing procedure to compress the inferred image and store the compressed inferred image in a storage device.

[0078] The storage device 120 of the first device 100 may include a first storage space for storing a first image as the original of the inferred image, and a second storage space for storing a second image as a converted image with lower capacity or quality as the original inferred image.

[0079] For example, the first device 100 may store the original inferred image (e.g., a BMP file) that is identified as a retraining target in a first storage space of the storage device 120. In another example, the first device 100 may use a predefined compression process to compress inferred images that are not identified as retraining targets and store the compressed images (e.g., JPEG files) in a second storage space. In other words, the first device may classify and store inferred images separately based on whether they are retraining targets, and thus, images that are retraining targets can be stored as originals, while images that are not retraining targets can be converted to a lower capacity and stored.

[0080] The first device 100 can convert a first image stored in a first storage space that has undergone a preset storage period into a second image and then store it in the second storage space. For example, the first device 100 can select an image stored for more than one year from the original images stored in the first storage space, compress the selected original image, and then move it to the second storage space.

[0081] The first device 100 can use the images stored in the storage device 120 to retrain the diagnostic model 110. Here, the first device 100 can use the first image stored in the first storage space of the storage device 120 as training data to retrain the diagnostic model 110.

[0082] The first device 100 can recover at least some of the second images stored in the second storage space of the storage device 120 according to a predefined image processing procedure, and use the recovered images as training data to retrain the diagnostic model 110. For example, the first device 100 can select one or more second images from the second images stored in the second storage space based on conditions set by the user (e.g., storing images for a specific time period or identifying images with a specific defect type). Thereafter, the first device 100 can use a predefined image recovery algorithm to recover the selected second images into high-resolution or high-quality images, and then use the recovered images to retrain the diagnostic model 110.

[0083] In other words, the diagnostic system according to embodiments of the present invention can classify the inferred images input during the diagnostic process into retraining targets and non-retraining targets, store the retraining target images as originals, and utilize them in subsequent retraining of the diagnostic model. Furthermore, the diagnostic system can compress and store images that are not retraining targets, select a subset of these compressed images required for retraining, recover them, and utilize them during retraining. This minimizes the cost of building storage within the diagnostic system while also preventing degradation of the quality of training data used to retrain the diagnostic model.

[0084] Figure 3 This is a flowchart illustrating a method for operating a diagnostic system by a second device according to an embodiment of the present invention.

[0085] The second device can receive the inferred image generated by the image sensor (S310).

[0086] Subsequently, the second device can input the inferred image into the diagnostic model received from the first device, and determine whether the diagnostic target is defective based on the inference results output by the diagnostic model (S320).

[0087] The second device can determine whether the inferred image is the target for retraining based on whether predefined conditions are met (S330).

[0088] In an embodiment, if both a first condition and a second condition are met, the second device can determine the inferred image as the target for retraining, wherein the inferred image is determined to be defective by a diagnostic model in the first condition and the inferred image is determined to be defective by a defect detection model defined separately from the diagnostic model in the second condition.

[0089] The second device can receive a trained defect detection model from the first device. Here, the defect detection model can be a machine learning-based model pre-trained using inferred images identified as defective as training data. For example, a defect detection model can be defined as learning features extracted from images identified as defective and outputting whether the inferred image is defective when given an inferred image as input.

[0090] In other words, according to an embodiment of this example, if the inferred image is determined to be defective by the diagnostic model applied to the second device, and is also determined to be defective by a defect detection model provided separately from the diagnostic model, then the inferred image can be determined as the target image for retraining.

[0091] In another embodiment, even if the first condition is not met, the second device can also determine the inferred image as the target for retraining when the third condition is met, which is that an abnormal region is detected by an anomaly detection model defined separately from the diagnostic model and the defect detection model.

[0092] The second device can receive a trained anomaly detection model from the first device. Here, the anomaly detection model can be a pre-trained machine learning-based model to determine whether the input image contains a predefined anomalous region.

[0093] In other words, according to this embodiment, if an anomaly detection model provided separately from the diagnostic model determines that an abnormal region is contained, the inferred image can be determined as an image for retraining even if the inferred image is determined to be normal by the diagnostic model applied to the second device.

[0094] The second device can transmit the determination result from step S330 (step S340) to the first device. Here, the second device can transmit the original inferred image along with the determination result of the inferred image to the first device.

[0095] Figure 4 This is a flowchart illustrating a method for operating a diagnostic system by a second device according to another embodiment of the present invention.

[0096] The second device can receive the inferred image generated by the image sensor (S410).

[0097] Subsequently, the second device can use the inferred image and the diagnostic model received from the first device to determine whether the diagnostic target has a defect (S420).

[0098] If the diagnostic model determines that the target is defective (Yes in S430), the second device can input the inferred image into a defect detection model defined separately from the diagnostic model, and determine whether the target is defective based on the inference result output by the defect detection model (S440).

[0099] If the defect detection model detects a defect in the diagnostic target (Yes in S450), the second device can identify the corresponding inferred image as the retraining target (S460). If the defect detection model does not detect a defect in the diagnostic target (No in S450), the second device can identify the inferred image as not undergoing retraining (S490).

[0100] If the diagnostic model determines that the diagnostic target is normal (No in S430), the second device can input the inferred image into an anomaly detection model defined separately from the diagnostic model, and determine whether an abnormal region is detected in the diagnostic target based on the inference result output by the anomaly detection model (S470).

[0101] If the anomaly detection model detects an abnormal region in the diagnostic target (Yes in S480), the second device can determine the inferred image as the target for retraining (S460). If the anomaly detection model does not detect an abnormal region in the diagnostic target (No in S480), the second device can determine the inferred image as not undergoing retraining (S490).

[0102] The second device can transmit the determination result of the inferred image to the first device (S340). Here, the second device can transmit the original inferred image together with the determination result of the inferred image to the first device.

[0103] Figure 5 This is a flowchart illustrating a method for operating a diagnostic system by a first device according to an embodiment of the present invention.

[0104] The first device can receive the original inferred image and a determination result including whether the inferred image has undergone retraining from the second device (S510).

[0105] If the inferred image undergoes retraining (as in S520), the first device may store the original of the inferred image in the storage device (S530). Here, the first device may store the original of the inferred image (e.g., a BMP file) that has been determined to have undergone retraining in the first storage space of the storage device.

[0106] If the inferred image does not undergo retraining (No in S520), the first device can use a predefined image processing procedure to convert the inferred image into a low-size or low-quality image and store the converted inferred image in a storage device (S540). Here, the first device can use a predefined compression procedure to compress the inferred image determined not to be retrained and store the compressed image (e.g., a JPEG file) in a second storage space of the storage device.

[0107] In other words, the first device can classify and store inferred images separately based on whether the inferred image needs to be retrained. Images that have undergone retraining can be stored in their original form, while images that have not undergone retraining can be converted to a lower capacity and stored.

[0108] The first device can compress images that have exceeded a preset storage period from the original images stored in the first storage space, and then move the compressed images to the second storage space.

[0109] Figure 6 This is a flowchart illustrating a method for operating a diagnostic system by a first device according to another embodiment of the present invention.

[0110] When the preset time period has passed or a user request corresponding to the retraining mode is received, the first device can initiate a mode for retraining the diagnostic model (S610).

[0111] Once the retraining mode is initiated, the first device can collect training data from the storage device (S620).

[0112] According to an embodiment, the first device can determine the original image stored in the first storage space of the storage device as training data.

[0113] In another embodiment, the first device can recover at least some compressed images stored in a second storage space of the storage device according to a predefined image processing procedure, and determine the recovered images as training data.

[0114] Subsequently, the first device can use the training data collected in S620 to retrain the pre-stored diagnostic model (S630).

[0115] Once retraining is complete, the first device can provide the retrained diagnostic model to the second device (S640). The second device can then update the existing diagnostic model with the retrained model and use the updated model to determine if the diagnostic target has defects.

[0116] Figure 7 This is a block diagram of a first device according to an embodiment of the present invention.

[0117] The first device 700 according to an embodiment of the present invention may correspond to a diagnostic model learning device for a factory monitoring system, or may be implemented as included in a diagnostic model learning device.

[0118] The first device 700 may include at least one processor 710, a memory 720 storing at least one instruction executed by the processor, and a transceiver 730 connected to a network connection and performing communication.

[0119] At least one instruction may include instructions for storing the original of the inferred image, which has been determined by the second device as the target for retraining, in a predefined storage space.

[0120] At least one instruction may also include instructions for compressing an inferred image that is determined not to be a target for retraining according to a predefined image processing procedure; and instructions for storing the compressed image of the inferred image separately from the original inferred image.

[0121] At least one instruction may also include an instruction to retrain the diagnostic model using multiple first images stored in the first storage space as training data.

[0122] At least one instruction may also include instructions for reconstructing at least some of the second images stored in the second storage space according to a predefined image processing procedure, and instructions for retraining the diagnostic model using the reconstructed images.

[0123] At least one instruction may further include an instruction for converting a first image that has been stored for a preset period of time among a plurality of first images stored in the first storage space into a second image, and an instruction for storing the second image in the second storage space.

[0124] The first device 700 according to an embodiment of the present invention may further include an input interface device 740, an output interface device 750, a storage device 760, etc. The corresponding components included in the first device 700 can be connected and communicate with each other via a bus 770. The storage device 760 can store diagnostic models, defect detection models, and anomaly detection models, and may include a first storage space for storing a first image and a second storage space for storing a second image.

[0125] Here, processor 710 may refer to a central processing unit (CPU), graphics processing unit (GPU), or dedicated processor on which the methods according to embodiments of the present invention are executed. The memory (or storage unit) may include at least one of volatile storage media and non-volatile storage media. For example, the memory may include at least one of read-only memory (ROM) and random access memory (RAM).

[0126] Figure 8 This is a block diagram of a second device according to an embodiment of the present invention.

[0127] The second device 800 according to an embodiment of the present invention may correspond to a diagnostic device of a factory monitoring system, or may be implemented as being included in a diagnostic device.

[0128] The second device 800 may include at least one processor 810, a memory 820 storing at least one instruction executed by the processor, and a transceiver 830 connected to a network to perform communication.

[0129] At least one instruction may include: an instruction for determining whether a diagnostic target is defective when an inferred image of a diagnostic target is received, using a diagnostic model transmitted from a first device and the inferred image of the diagnostic target; an instruction for determining whether the inferred image has undergone retraining based on whether predefined conditions are met; and an instruction for transmitting the determination result to the first device.

[0130] Instructions for determining whether an inferred image has undergone retraining may include: instructions for identifying the inferred image as a target for retraining if both a first condition and a second condition are met, wherein the inferred image is determined to be defective by a diagnostic model under the first condition, and the inferred image is determined to be defective by a defect detection model defined separately from the diagnostic model under the second condition.

[0131] Instructions for determining whether an inferred image has undergone retraining may include: instructions for identifying the inferred image as a target for retraining even if a first condition is not met but a third condition is met, in which an anomaly detection model, which is defined separately from the diagnostic model and the defect detection model, detects an anomaly region.

[0132] The second device 800 according to an embodiment of the present invention may further include an input interface device 840, an output interface device 850, a storage device 860, etc. The corresponding components included in the second device 800 can be connected and communicate with each other via a bus 870. The storage device 860 can store diagnostic models, defect detection models, and anomaly detection models.

[0133] Here, processor 810 may refer to a central processing unit (CPU), graphics processing unit (GPU), or dedicated processor on which the methods according to embodiments of the present invention are executed. The memory (or storage unit) may include at least one of volatile storage media and non-volatile storage media. For example, the memory may include at least one of read-only memory (ROM) and random access memory (RAM).

[0134] The operation of the method according to embodiments of the present invention can be implemented as a computer-readable program or code on a computer-readable recording medium. A computer-readable recording medium includes all types of recording devices in which computer systems store data readable by the computer. Furthermore, the computer-readable recording medium can be distributed across network-connected computer systems to store and execute computer-readable programs or code in a distributed manner.

[0135] Although some aspects of the invention have been described in the context of a device, they may also refer to, according to the description of the corresponding method, a block or device corresponding to a method step or feature of a method step. Similarly, aspects described in the context of a method may also refer to features of a corresponding block or item or corresponding device. Some or all of the method steps may be performed by (or using) hardware devices, such as, for example, a microprocessor, a programmable computer, or electronic circuitry. In some embodiments, one or more of the most important method steps may be performed by such devices.

[0136] In the foregoing, the present invention has been described with reference to exemplary embodiments thereof. However, those skilled in the art will understand that various corrections and modifications may be made to the present invention within the scope of the appended claims without departing from the spirit and scope of the invention as described therein.

Claims

1. A diagnostic system, comprising: A first device is configured to train a machine learning-based diagnostic model using training images. as well as A second device is configured to determine whether the diagnostic target is defective using an inferred image of the diagnostic target provided by the first device and a diagnostic model. The second device determines whether the inferred image is a target for retraining based on whether one or more predefined conditions are met. The first device stores the original inferred image, which has been determined by the second device as the target for retraining, in a predefined storage space.

2. The diagnostic system according to claim 1, wherein, The first device compresses inferred images that are determined not to be targets for retraining according to a predefined image processing procedure, and stores the compressed images of the inferred images separately from the original inferred images.

3. The diagnostic system according to claim 1, wherein, The first device includes a storage device for storing the inferred image, and The storage device includes: A first storage space for storing a first image, the first image being the original of the inferred image; and A second storage space for storing a second image, which is a converted image of the original with lower capacity or quality.

4. The diagnostic system according to claim 3, wherein, The first device uses multiple first images stored in the first storage space as training data to retrain the diagnostic model.

5. The diagnostic system according to claim 3, wherein, The first device reconstructs at least some of the second images stored in the second storage space according to a predefined image processing procedure, and uses the reconstructed images to retrain the diagnostic model.

6. The diagnostic system according to claim 3, wherein, The first device converts a first image that has been stored for a preset period of time into a second image from among a plurality of first images stored in the first storage space, and stores the second image in the second storage space.

7. The diagnostic system according to claim 1, wherein, The second device identifies the inferred image as the target for retraining if two conditions are met, including: The first condition is that the inferred image is determined to be defective by the diagnostic model; and The second condition is that the inferred image is determined to be defective by a defect detection model defined separately from the diagnostic model.

8. The diagnostic system according to claim 7, wherein, Even if the first condition is not met, if the third condition is met, the second device identifies the inferred image as the target for retraining, wherein an abnormal region is detected by an anomaly detection model that is defined separately from the diagnostic model and the defect detection model in the third condition.

9. A method of operating a diagnostic system, the diagnostic system comprising: A first device is configured to train a machine learning-based diagnostic model using training images. And a second device, the second device being configured to determine whether the diagnostic target is defective using the diagnostic model and an inferred image of the diagnostic target provided by the first device, the method comprising: The second device determines whether the inferred image is a target for retraining based on whether one or more predefined conditions are met. and The first device stores the original of the inferred image, which has been determined by the second device as the target for retraining, in a predefined storage space.

10. The method of claim 9, further comprising: The first device compresses inferred images that are determined not to be targets for retraining, according to a predefined image processing procedure. and The first device stores the compressed image of the inferred image separately from the original image of the inferred image.

11. The method according to claim 9, wherein, The first device includes a storage device for storing the inferred image, and The storage device includes: A first storage space for storing a first image, the first image being the original of the inferred image; and A second storage space for storing a second image, which is a converted image of the original with lower capacity or quality.

12. The method of claim 11, further comprising: The diagnostic model is retrained using the first device and multiple first images stored in the first storage space as training data.

13. The method of claim 11, further comprising: The first device reconstructs at least some of the second images stored in the second storage space according to a predefined image processing procedure; and The diagnostic model is retrained using the reconstructed images from the first device.

14. The method of claim 11, further comprising: The first device converts a first image that has been stored for a preset period of time into a second image from among a plurality of first images stored in the first storage space, and stores the second image in the second storage space.

15. The method according to claim 9, wherein, Determining whether the inferred image is the target for retraining includes: The inferred image is identified as a target for retraining if two conditions are met, the conditions being: a first condition in which the inferred image is determined to be defective by the diagnostic model; and a second condition in which the inferred image is determined to be defective by a defect detection model defined separately from the diagnostic model.

16. The method according to claim 15, wherein, Determining whether the inferred image is the target for retraining includes: Even if the first condition is not met, if the third condition is met, the inferred image is identified as the target for retraining, in which an anomaly detection model, which is defined separately from the diagnostic model and the defect detection model, detects an anomaly region.

Images