Diagnostic system using artificial intelligence model and operation method thereof

Abstract

A diagnostic system according to one embodiment of the present invention may comprise: a first device for training a machine learning-based diagnostic model by using a training image; and a second device for determining whether an object to be diagnosed is defective by using the diagnostic model provided from the first device and an inference image of the object to be diagnosed. Here, the second device may determine whether the inference image is a retraining target on the basis of whether a predefined condition is satisfied, and the first device may store, in a predefined storage space, the original of the inference image determined as the retraining target by the second device.

Description

Diagnostic system using artificial intelligence model and its operation method

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

[0002] The present invention relates to an abnormal diagnosis method, a diagnosis system, and an operation method thereof, and more specifically, to a diagnosis system that diagnoses a diagnosis target using an artificial intelligence-based diagnosis model, and an operation method thereof.

[0003] Secondary batteries are batteries that can be reused by charging even after discharge, and can be used as an energy source for small devices such as mobile phones, tablet PCs, and vacuum cleaners, and are also used as medium- to large-scale energy sources such as personal mobility, automobiles, and ESS (Energy Storage Systems) for smart grids.

[0004] Secondary batteries are used in the form of assemblies 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, depending on the requirements of the system.

[0005] Producing battery cells or battery assemblies requires a variety of processes. Here, automated equipment is installed on the production line to perform each process, and the organic connection of these automated equipment enables mass production of battery cells or battery assemblies.

[0006] To ensure product quality meets required standards, diagnostics are performed at key processing points on the production line. Typically, a video-based monitoring system (FMVS; Factory Monitoring Visual System) utilizing camera equipment installed at key processing points on the production line can be utilized.

[0007] Recently, FMVS, which incorporates an AI model, has been proposed to improve the diagnostic accuracy of monitoring systems. This AI-based diagnostic system inputs images of the target object into an AI-based diagnostic model and, based on the output from the model, determines whether the target object is defective.

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

[0009] However, since the original images collected during the diagnostic process are high-resolution, high-volume data, it requires enormous costs to build a storage facility to store a large amount of original images.

[0010] As a related prior literature, there is KR 10-2018-0009563 A.

[0011] The purpose of the present invention to solve the above problems is to provide a diagnosis system using an artificial intelligence model.

[0012] Another object of the present invention to solve the above problems is to provide a method for operating such a diagnostic system.

[0013] According to one embodiment of the present invention for achieving the above object, a diagnostic system may include a first device that trains a machine learning-based diagnostic model using a learning image; and a second device that determines whether a diagnostic target is defective using the diagnostic model provided from the first device and an inference image of the diagnostic target. Here, the second device determines whether the inference image is a target for re-learning based on whether a predefined condition is satisfied, and the first device may store the original of the inference image determined as a target for re-learning by the second device in a predefined storage space.

[0014] The first device can compress an inference image determined not to be a relearning target according to a predefined image processing process, and store the compressed image of the inference image separately from the original.

[0015] The first device may include a storage device that stores the inference image. Here, the storage device may include a first storage space that stores a first image, which is the original of the inference image; and a second storage space that stores a second image, which is the original converted to a low-capacity or low-quality image.

[0016] The first device can retrain the diagnostic model by using a plurality of first images stored in the first storage space as learning data.

[0017] The first device can restore at least some of the second images stored in the second storage space according to a predefined image processing process and retrain the diagnostic model using the restored images.

[0018] The first device can convert a first image, among a plurality of first images stored in the first storage space, for which a preset storage period has elapsed, into a second image and then store the converted first image in the second storage space.

[0019] The second device may determine the inference image as a target for relearning if it satisfies both a first condition, which is determined to be defective by the diagnostic model, and a second condition, which is determined to be defective by a defect detection model defined separately from the diagnostic model.

[0020] The second device may determine the inference image as a target for relearning if it satisfies a third condition, in which an abnormal area is detected by an abnormality detection model defined separately from the diagnosis model and the defect detection model, even if the first condition is not satisfied.

[0021]

[0022] According to one embodiment of the present invention for achieving the above other object, a method for operating a diagnostic system includes a first device for training a machine learning-based diagnostic model using a training image; and a second device for determining whether a diagnostic target is defective using the diagnostic model provided from the first device and an inference image for the diagnostic target, the method including a step of the second device determining whether the inference image is a target for re-training based on whether a predefined condition is satisfied; and a step of the first device storing an original of an inference image determined by the second device to be a target for re-training in a predefined storage space.

[0023] The method of operating the above diagnostic system may further include a step of the first device compressing an inference image determined not to be a relearning target according to a predefined image processing process; and a step of the first device storing the compressed image of the inference image separately from the original.

[0024] The first device may include a storage device that stores the inference image, and the storage device may include a first storage space that stores a first image, which is an original of the inference image; and a second storage space that stores a second image, which is an original converted to a low-capacity or low-quality image.

[0025] The operating method of the above diagnostic system may further include a step of retraining the diagnostic model by using a plurality of first images stored in the first storage space as learning data.

[0026] The method of operating the above diagnostic system may further include a step of the first device restoring at least some of the second images stored in the second storage space according to a predefined image processing process; and a step of the first device retraining the diagnostic model using the restored images.

[0027] The operating method of the above diagnostic system may further include a step of the first device converting a first image, among a plurality of first images stored in the first storage space, for which a preset storage period has elapsed, into a second image and then storing the converted first image in the second storage space.

[0028] The step of determining whether the above-mentioned inference image is a target for relearning may include a step of determining the inference image as a target for relearning if both a first condition, which is determined to be defective by the diagnostic model; and a second condition, which is determined to be defective by a defect detection model defined separately from the diagnostic model, are satisfied.

[0029] The step of determining whether the above-mentioned inference image is a target for relearning may include a step of determining the inference image as a target for relearning if the third condition, in which an abnormal area is detected by an abnormality detection model defined separately from the diagnosis model and the defect detection model, is satisfied, even if the first condition is not satisfied.

[0030] According to the above-described embodiment of the present invention, the diagnostic accuracy of the diagnostic system can be improved, and the cost of constructing a storage device included in the diagnostic system can be minimized.

[0031] Figure 1 is a block diagram of a factory monitoring system according to an embodiment of the present invention.

[0032] Figure 2 is a block diagram of a diagnostic system according to an embodiment of the present invention.

[0033] Figure 3 is an operation flowchart of a diagnostic system operation method by a second device according to an embodiment of the present invention.

[0034] FIG. 4 is an operational flowchart of a diagnostic system operation method by a second device according to another embodiment of the present invention.

[0035] Figure 5 is an operation flowchart of a diagnostic system operation method by a first device according to an embodiment of the present invention.

[0036] FIG. 6 is an operational flowchart of a diagnostic system operation method by a first device according to another embodiment of the present invention.

[0037] Figure 7 is a block diagram of a first device according to an embodiment of the present invention.

[0038] Figure 8 is a block diagram of a second device according to an embodiment of the present invention.

[0039] 10: Diagnostic target

[0040] 20: Image sensor

[0041] 30: Diagnostic device

[0042] 40: Diagnostic model learning device

[0043] 100, 700: First device

[0044] 200, 800: Second device

[0045] The present invention is susceptible to various modifications and embodiments. Specific embodiments are illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present invention to specific embodiments, but rather to encompass all modifications, equivalents, and alternatives falling within the spirit and technical scope of the present invention. Throughout the description of each drawing, similar reference numerals have been used to designate similar components.

[0046] Terms such as "first," "second," "A," and "B" may be used to describe various components, but these components should not be limited by these terms. These terms are used solely to distinguish one component from another. For example, without departing from the scope of the present invention, the first component could be referred to as the "second component," and similarly, the second component could also be referred to as the "first component." The term "and / or" includes any combination of multiple related items listed or any one of multiple related items listed.

[0047] When a component is referred to as being "connected" or "connected" to another component, it should be understood that it may be directly connected or connected to that other component, but that there may be other components intervening. Conversely, when a component is referred to as being "directly connected" or "connected" to another component, it should be understood that there are no other components intervening.

[0048] The terminology used in this application is only used to describe specific embodiments and is not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, it should be understood that the terms "comprise" or "have" indicate the presence of a feature, number, step, operation, component, part, or combination thereof described in the specification, but do not exclude in advance the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0049] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Terms defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and shall not be interpreted in an idealized or overly formal sense unless explicitly defined herein.

[0050]

[0051] Hereinafter, various embodiments of the present invention will be described in detail with reference to the attached drawings. The description will exemplify a diagnostic system according to the present invention installed in a battery manufacturing or assembly plant to diagnose batteries or battery processing equipment. However, the scope of the present invention is not limited to such types of inspection objects.

[0052]

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

[0054] Referring to FIG. 1, a 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] An image sensor (20) is a device that generates image data for one or more diagnostic targets (10) located within a factory. For example, the image sensor (20) may correspond to a camera that is fixed and configured at a specific location.

[0056] The image sensor (20) is connected to a diagnostic device (30) through a network and can transmit image data on a diagnostic target (10) to the diagnostic device (30).

[0057] The image sensor (20) can generate image data for a specific area of ​​the diagnostic target (10). For example, the image sensor (20) can be installed at a specific location in the assembly equipment and configured to photograph the fastening area of ​​a specific bolt. As another example, the image sensor (20) can be installed at a facility adjacent to the electrolyte injection device and configured to photograph the interior of the electrolyte injection device.

[0058] The image sensor (20) can generate image data for the diagnosis target (10) at preset unit times and transmit the data to the diagnosis device (30). For example, the image sensor (20) can generate 60 image data per second (60 fps) and transmit the data to the diagnosis device (30).

[0059] The factory monitoring system may include a plurality of image sensors (20). Here, each of the image sensors may be positioned at a predetermined location, configured to generate image data for a corresponding diagnostic target, and transmit the generated image data to a diagnostic device (30).

[0060] The diagnostic device (30) is a device that receives image data from the image sensor (20) and diagnoses whether there is an abnormality in the diagnosis target based on the image data. Here, the abnormality diagnostic device (30) can determine whether there is a defect in the diagnosis target using the image data received from the image sensor (20) and a predefined diagnostic model. For example, the diagnostic device (30) can determine whether there is a defect such as a bolt fastening defect, a cell assembly defect, or a cell specification defect.

[0061] The diagnostic model learning device (40) is 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 learn a machine learning-based diagnostic model using learning data and provide the learned diagnostic model to the diagnostic device (30).

[0062] The diagnostic device (30) inputs an image of a diagnosis target transmitted from an image sensor into a diagnostic model provided from a diagnostic model learning device (40), and can determine whether the diagnosis target is defective based on an inference result output by the diagnostic model. That is, the diagnostic device (30) can perform a diagnosis on the diagnosis target using an inference image transmitted from the image sensor (20) and a learned diagnostic model provided from the diagnostic model learning device (40).

[0063] The diagnostic model learning device (40) can store the inference image in a storage device. Here, the diagnostic model learning device (40) can store the original inference image and the compressed image of the inference image separately in the storage device.

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

[0065]

[0066] Figure 2 is a block diagram of a diagnostic system according to an embodiment of the present invention.

[0067] Referring to FIG. 2, a 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 the diagnostic model learning device (40) of FIG. 1, and the second device (200) may correspond to the diagnostic device (30) of FIG. 1.

[0068] The first device (100) can train a machine learning-based diagnostic model (110) for diagnosis of a diagnostic target and provide the trained diagnostic model to the second device (200).

[0069] The diagnostic model (110) can be trained to output whether or not an image of a diagnostic target is defective (normal or defective) as output data when the image is input as input data.

[0070] The first device (100) may include a storage device (120) in which learning images are stored, and may train a diagnostic model (110) using the learning images stored in the storage device (120).

[0071] The second device (200) can determine whether the diagnosis target is defective or not by using the diagnosis model (210) received from the first device (100) and the inference image for the diagnosis target. For example, the second device (200) can receive an image of the diagnosis target generated by the image sensor as an inference image and input the inference image into the diagnosis model (210). Here, the second device (200) can determine whether the diagnosis target is defective or not based on the inference result output by the diagnosis model (210).

[0072] The second device (200) can determine whether the inference image is a target for relearning based on whether a predefined condition is satisfied.

[0073] For example, if the inference result by the diagnostic model (210) is [bad], the second device (200) can determine the corresponding inference image as a target for relearning.

[0074] For another example, the second device (200) may determine the inference image as a target for relearning when the inference result by the diagnostic model (210) is [bad] and the inference result by the machine learning-based defect detection model defined separately from the diagnostic model is [bad].

[0075] For another example, the second device (200) may determine the inference image as a target for relearning if the inference result by the diagnostic model (210) is [normal], but the inference result by the machine learning-based abnormality detection model defined separately from the diagnostic model is [abnormal area detection].

[0076] The first device (100) can store the inference image in the storage device (120). Here, the first device (100) can receive the inference image from the second device (200) or from an image sensor.

[0077] The first device (100) can store the inference image in its original form in the storage device (120), or can store the inference image in the storage device (120) after converting it into a low-capacity or low-quality image according to a predefined image processing process.

[0078] Here, if a specific inference image is determined to be a re-learning target by the second device (200), the first device (100) may store the original of the inference image in the storage device. In addition, if a specific inference image is determined not to be a re-learning target by the second device (200), the first device (100) may compress the inference image according to a predefined image processing process and store the compressed inference image in the storage device.

[0079] The storage device (120) of the first device (100) may include a first storage space for storing a first image, which is the original of the inference image, and a second storage space for storing a second image, which is the original of the inference image converted to a low-capacity or low-quality image.

[0080] For example, the first device (100) can store the original (e.g., BMP file) of the inference image determined to be a target for relearning in the first storage space of the storage device (120). As another example, the first device (100) can compress the inference image determined not to be a target for relearning according to a predefined compression process and store the compressed image (e.g., JPEG file) in the second storage space. That is, the first device can store the inference images separately depending on whether they are targets for relearning, but can store the images that are targets for relearning as originals and convert and store the images that are not targets for relearning into low-capacity images.

[0081] The first device (100) may convert first images stored in the first storage space, for which a preset storage period has elapsed, into second images and then store them in the second storage space. For example, the first device (100) may select images for which a storage period has elapsed from among original images stored in the first storage space, compress the selected original images, and then move them to the second storage space.

[0082] The first device (100) can retrain the diagnostic model (110) using images stored in the storage device (120). Here, the first device (100) can retrain the diagnostic model (110) using the first image stored in the first storage space of the storage device (120) as training data.

[0083] The first device (100) can restore at least some of the second images stored in the second storage space of the storage device (120) according to a predefined image processing process, and retrain the diagnostic model (110) using the restored images as learning data. For example, the first device (100) can select one or more second images from among the second images stored in the second storage space based on conditions set by a user (e.g., images stored for a specific period of time, or images determined to have a specific defect type). Thereafter, the first device (100) can restore the selected second images into high-resolution or high-quality images using a predefined image restoration algorithm, and then retrain the diagnostic model (110) using the restored images.

[0084] That is, the diagnostic system according to the present invention can classify inference images input during the diagnostic process into those subject to retraining and those not subject to retraining, store the images subject to retraining as originals, and utilize them in the subsequent retraining process of the diagnostic model. In addition, the diagnostic system can compress and store images not subject to retraining, select some of the compressed images necessary for retraining, restore them, and utilize them in the retraining process. Accordingly, the cost of constructing a storage device within the diagnostic system can be minimized, and at the same time, the quality of training data for retraining the diagnostic model can be prevented from deteriorating.

[0085]

[0086] Figure 3 is an operation flowchart of a diagnostic system operation method by a second device according to an embodiment of the present invention.

[0087] The second device can receive an inference image generated by the image sensor (S310).

[0088] Thereafter, the second device can input the inference image into the diagnostic model received from the first device and determine whether the diagnosis target is defective based on the inference result output by the diagnostic model (S320).

[0089] The second device can determine whether the inference image is a target for relearning based on whether a predefined condition is satisfied (S330).

[0090] In an embodiment, the second device may determine the inference image as a target for relearning if both a first condition, which is determined to be defective by the diagnostic model, and a second condition, which is determined to be defective by a defect detection model defined separately from the diagnostic model, are satisfied.

[0091] The second device may receive a defect detection model trained by the first device. The defect detection model may be a machine learning-based model pre-trained using inference images determined to be defective as training data. For example, the defect detection model may be defined to learn features extracted from images determined to be defective and output whether or not an inference image is defective when input.

[0092] That is, according to the present embodiment, if an inference image is judged as defective by a diagnostic model applied to a second device and is also judged as defective by a defect detection model provided separately from the diagnostic model, the image may be determined as a target image for relearning.

[0093] In another embodiment, the second device may determine the inference image as a target for relearning if it satisfies a third condition, that is, an abnormal area is detected by an abnormality detection model defined separately from the diagnosis model and the defect detection model, even if the first condition is not satisfied.

[0094] The second device may receive a trained anomaly detection model from the first device. Here, the anomaly detection model may be a machine learning-based model pretrained to determine whether an input image contains a predefined abnormal region.

[0095] That is, according to the present embodiment, even if the inference image is determined to be normal by the diagnostic model applied to the second device, if it is determined to include an abnormal area by an abnormality detection model provided separately from the diagnostic model, it can be determined as an image to be relearned.

[0096] The second device can transmit the judgment result by S330 to the first device (S340). Here, the second device can transmit the original file of the inference image along with the judgment result for the inference image to the first device.

[0097]

[0098] FIG. 4 is an operational flowchart of a diagnostic system operation method by a second device according to another embodiment of the present invention.

[0099] The second device can receive an inference image generated by the image sensor (S410).

[0100] Thereafter, the second device can determine whether the diagnosis target is defective or not using the inference image and the diagnosis model received from the first device (S420).

[0101] If the diagnosis model determines that the diagnosis object is defective (Y of S430), the second device inputs the inference image into a defect detection model defined separately from the diagnosis model, and can determine whether the diagnosis object is defective based on the inference result output by the defect detection model (S440).

[0102] If a defect is detected in the diagnosis target by the defect detection model (Y in S450), the second device may determine the corresponding inference image as a re-learning target (S460). If a defect is not detected in the diagnosis target by the defect detection model (N in S450), the second device may determine the corresponding inference image as not a re-learning target (S490).

[0103] If the diagnosis model determines that the image is normal (N of S430), the second device inputs the inference image into an anomaly detection model defined separately from the diagnosis model, and can determine whether an abnormal area is detected in the diagnosis target based on the inference result output by the anomaly detection model (S470).

[0104] If an abnormal area is detected in the diagnosis target by the anomaly detection model (Y in S480), the second device may determine the corresponding inference image as a re-learning target (S460). If an abnormal area is not detected in the diagnosis target by the anomaly detection model (N in S480), the second device may determine the corresponding inference image as not a re-learning target (S490).

[0105] The second device can transmit the judgment result for the inference image to the first device (S340). Here, the second device can transmit the original file of the inference image along with the judgment result for the inference image to the first device.

[0106]

[0107] Figure 5 is an operation flowchart of a diagnostic system operation method by a first device according to an embodiment of the present invention.

[0108] The first device can receive, from the second device, the original inference image and a judgment result including whether the inference image is a target for relearning (S510).

[0109] If the inference image is a target for relearning (Y in S520), the first device can store the original of the inference image in the storage device (S530). Here, the first device can store the original (e.g., a BMP file) of the inference image determined to be a target for relearning in the first storage space of the storage device.

[0110] If the inference image is not a target for relearning (N of S520), the first device may convert the inference image into a low-capacity or low-quality image according to a predefined image processing process and store the converted inference image in a storage device. Here, the first device may compress the inference image, which is determined not to be a target for relearning, according to a predefined compression process and store the compressed image (e.g., a JPEG file) in a second storage space of the storage device.

[0111] That is, the first device can store inference images by distinguishing them according to whether they are targets for relearning or not, and can store images that are targets for relearning as originals, and convert images that are not targets for relearning into low-capacity images and store them.

[0112] The first device can compress images among the original images stored in the first storage space, for which a preset storage period has elapsed, and then move the compressed images to the second storage space.

[0113]

[0114] FIG. 6 is an operational flowchart of a diagnostic system operation method by a first device according to another embodiment of the present invention.

[0115] The first device can initiate a mode for relearning the diagnostic model when a preset period of time has elapsed or a user request corresponding to initiating the relearning mode is input (S610).

[0116] When the relearning mode is initiated, the first device can collect learning data from the storage device (S620).

[0117] In an embodiment, the first device may determine an original image stored in a first storage space of the storage device as learning data.

[0118] In another embodiment, my device can restore at least some of the compressed images stored in the second storage space of the storage device according to a predefined image processing process and determine the restored images as learning data.

[0119] Thereafter, the first device can retrain the pre-stored diagnostic model using the learning data collected in S620 (S630).

[0120] Once retraining is complete, the first device can provide the retrained diagnostic model to the second device. The second device can then update the existing diagnostic model with the retrained model and use the updated diagnostic model to determine whether the diagnostic target is defective.

[0121]

[0122] Figure 7 is a block diagram of a first device according to an embodiment of the present invention.

[0123] The first device (700) according to an embodiment of the present invention corresponds to a diagnostic model learning device of a factory monitoring system, or can be implemented by being included in a diagnostic model learning device.

[0124] The first device (700) may include at least one processor (710), a memory (720) that stores at least one command executed through the processor, and a transmission / reception device (730) that is connected to a network and performs communication.

[0125] The at least one command may include a command to store the original of the inference image determined to be a relearning target by the second device in a predefined storage space.

[0126] The at least one command may further include a command for compressing an inference image determined not to be a target of relearning according to a predefined image processing process, and a command for storing the compressed image of the inference image separately from the original.

[0127] The at least one command may further include a command to retrain the diagnostic model using a plurality of first images stored in the first storage space of the storage device as training data.

[0128] The at least one command may further include a command to restore at least some of the second images stored in the second storage space of the storage device according to a predefined image processing process, and a command to retrain the diagnostic model using the restored images.

[0129] The at least one command may further include a command to convert a first image, among a plurality of first images stored in the first storage space, for which a preset storage period has elapsed, into a second image and then store the converted first image in the second storage space.

[0130] The first device (700) may further include an input interface device (740), an output interface device (750), a storage device (760), etc. Each component included in the first device (700) may be connected by a bus (770) and communicate with each other. The storage device (760) may store a diagnostic model, a defect detection model, and an anomaly detection model, and may include a first storage space for storing a first image, and a second storage space for storing a second image.

[0131] Here, the processor (710) may refer to a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to embodiments of the present invention are performed. The memory (or storage device) may be comprised of at least one of a volatile storage medium and a non-volatile storage medium. For example, the memory may be comprised of at least one of a read-only memory (ROM) and a random access memory (RAM).

[0132]

[0133] Figure 8 is a block diagram of a second device according to an embodiment of the present invention.

[0134] The second device (800) according to an embodiment of the present invention corresponds to a diagnostic device of a factory monitoring system, or can be implemented as included in a diagnostic device.

[0135] The second device (800) may include at least one processor (810), a memory (820) that stores at least one command executed through the processor, and a transmission / reception device (830) that is connected to a network and performs communication.

[0136] The at least one command may include a command for determining whether the diagnosis target is defective using the diagnosis model transmitted from the first device and the inference image for the diagnosis target when an inference image for the diagnosis target is received, a command for determining whether the inference image is a target for relearning based on whether a predefined condition is satisfied, and a command for transmitting the determination result to the first device.

[0137] The command for determining whether the above-mentioned inference image is a target for relearning may include a command for determining the inference image as a target for relearning if both a first condition, which is determined to be defective by the diagnostic model; and a second condition, which is determined to be defective by a defect detection model defined separately from the diagnostic model, are satisfied.

[0138] The command for determining whether the above-mentioned inference image is a target for relearning may include a command for determining the above-mentioned inference image as a target for relearning if the above-mentioned third condition, in which an abnormal area is detected by an abnormality detection model defined separately from the above-mentioned diagnosis model and the above-mentioned defect detection model, is satisfied, even if the above-mentioned first condition is not satisfied.

[0139] The second device (800) may further include an input interface device (840), an output interface device (850), a storage device (860), etc. Each component included in the second device (800) may be connected by a bus (870) to communicate with each other. The storage device (860) may store a diagnostic model, a defect detection model, and an anomaly detection model.

[0140] Here, the processor (810) may refer to a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to embodiments of the present invention are performed. The memory (or storage device) may be composed of at least one of a volatile storage medium and a non-volatile storage medium. For example, the memory may be composed of at least one of a read-only memory (ROM) and a random access memory (RAM).

[0141]

[0142] The operations of the method according to an embodiment 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 any type of recording device that stores data readable by a computer system. Furthermore, a computer-readable recording medium can be distributed across network-connected computer systems, allowing the computer-readable program or code to be stored and executed in a distributed manner.

[0143]

[0144] While some aspects of the present invention have been described in the context of a device, they may also represent a description of a corresponding method, wherein a block or device corresponds to a method step or a feature of a method step. Similarly, aspects described in the context of a method may also be described as a corresponding block or item or a feature of a corresponding device. Some or all of the method steps may be performed by (or using) a hardware device, such as, for example, a microprocessor, a programmable computer, or an electronic circuit. In some embodiments, one or more of the most significant method steps may be performed by such a device.

[0145] Although the present invention has been described with reference to preferred embodiments thereof, it will be understood by those skilled in the art that various modifications and changes may be made to the present invention without departing from the spirit and scope of the present invention as set forth in the claims below.

Claims

1. A first device that trains a machine learning-based diagnostic model using learning images; and A second device is included that determines whether the diagnosis target is defective by using the diagnosis model provided from the first device and an inference image for the diagnosis target, The second device above, Based on whether the predefined conditions are satisfied, it is determined whether the above inference image is a target for relearning, The above first device, A diagnostic system that stores the original of the inference image determined to be a relearning target by the second device in a predefined storage space.

2. In claim 1, The above first device, A diagnostic system that compresses an inference image determined not to be a subject of relearning according to a predefined image processing process and stores the compressed image of the inference image separately from the original.

3. In claim 1, The above first device, A storage device for storing the above inference image is included, The above storage device, A first storage space for storing a first image, which is the original of the above inference image; and A diagnostic system comprising a second storage space for storing a second image converted from the original to a low-capacity or low-quality image.

4. In claim 3, The above first device, A diagnostic system that retrains the diagnostic model by using a plurality of first images stored in the first storage space as learning data.

5. In claim 3, The above first device, A diagnostic system that restores at least some of the second images stored in the second storage space according to a predefined image processing process and retrains the diagnostic model using the restored images.

6. In claim 3, The above first device, A diagnostic system that converts a first image among a plurality of first images stored in the first storage space, for which a preset storage period has elapsed, into a second image and then stores the converted first image in the second storage space.

7. In claim 1, The second device above, A diagnostic system that determines the inference image as a target for relearning when both a first condition, which is determined to be defective by the diagnostic model, and a second condition, which is determined to be defective by a defect detection model defined separately from the diagnostic model, are satisfied.

8. In claim 7, The second device above, A diagnostic system that determines the inference image as a target for relearning if the third condition is satisfied, in which an abnormal area is detected by an abnormality detection model defined separately from the diagnostic model and the defect detection model, even if the first condition is not satisfied.

9. A method for operating a diagnostic system, comprising: a first device that trains a machine learning-based diagnostic model using a learning image; and a second device that determines whether a diagnostic target is defective using the diagnostic model provided from the first device and an inference image of the diagnostic target. A step in which the second device determines whether the inference image is a target for relearning based on whether a predefined condition is satisfied; and A method for operating a diagnostic system, comprising a step of storing the original of an inference image determined to be a relearning target by the second device in a predefined storage space.

10. In claim 9, A step in which the first device compresses the inference image determined not to be the relearning target according to a predefined image processing process; and A method of operating a diagnostic system, wherein the first device further includes a step of storing a compressed image of the inference image separately from the original.

11. In claim 9, The above first device, A storage device for storing the above inference image is included, The above storage device, A first storage space for storing a first image, which is the original of the above inference image; and A method of operating a diagnostic system, comprising a second storage space for storing a second image converted from the original to a low-capacity or low-quality image.

12. In claim 11, A method for operating a diagnostic system, further comprising a step of retraining the diagnostic model by using a plurality of first images stored in the first storage space as learning data.

13. In claim 11, A step in which the first device restores at least some of the second images stored in the second storage space according to a predefined image processing process; and A method of operating a diagnostic system, wherein the first device further includes a step of retraining the diagnostic model using the restored image.

14. In claim 11, A method of operating a diagnostic system, wherein the first device further comprises a step of converting a first image, among a plurality of first images stored in the first storage space, for which a preset storage period has elapsed, into a second image and then storing the converted first image in the second storage space.

15. In claim 9, The step of determining whether the above inference image is a target for relearning is as follows: A method for operating a diagnostic system, comprising a step of determining the inference image as a target for relearning when both a first condition, which is determined to be defective by the diagnostic model, and a second condition, which is determined to be defective by a defect detection model defined separately from the diagnostic model, are satisfied.

16. In claim 15, The step of determining whether the above inference image is a target for relearning is as follows: A method for operating a diagnostic system, comprising a step of determining the inference image as a target for relearning when a third condition is satisfied, in which an abnormal area is detected by an abnormality detection model defined separately from the diagnostic model and the defect detection model, even if the first condition is not satisfied.

Images