Monitoring device and operation method thereof
The monitoring device employs AI models to analyze battery images and detect new defects, improving defect detection precision and preventing defective product leakage by retraining the model based on new defect types, addressing the limitations of existing inspection methods.
Patent Information
- Application Number
- PCT/KR2024/005947
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-02
- Filing Date
- 2024-05-02
- Publication Date
- 2024-11-07
AI Technical Summary
Current battery inspection methods struggle to detect new types of defects in the manufacturing and assembly process, leading to potential leakage of defective products, as they rely on pre-learned data and may misclassify new defects as good products due to low incidence of new defect types.
A monitoring device equipped with a communication circuit, processor, and memory that uses two artificial intelligence models to determine if a battery is defective by analyzing images, with the first model classifying normal or defective products and the second model identifying new defect types by comparing image distributions to learning data, allowing for retraining of the first model if necessary.
The solution effectively detects new types of defective products, increasing the precision of defect detection and preventing the leakage of defective batteries by learning images of new defective types, thereby enhancing the reliability and safety of battery inspection.
Smart Images

Figure KR2024005947_07112024_PF_FP_ABST
Abstract
Description
Monitoring device and method of operation thereof
[0001] Cross-citation with related applications
[0002] This invention claims the benefit of priority from Korean Patent Application No. 10-2023-0057243, filed on May 2, 2023, the entire disclosure of which is incorporated herein by reference.
[0003] Technology field
[0004] The embodiments disclosed in this document relate to a monitoring device and a method of operating the same.
[0005] Recently, as the demand for portable electronic products such as laptops, video cameras, and mobile phones has rapidly increased, and the development of electric vehicles, energy storage batteries, robots, and satellites has been in full swing, research into high-performance batteries capable of repeated charging and discharging is actively being conducted.
[0006] Currently commercialized batteries include nickel-cadmium batteries, nickel-hydrogen batteries, nickel-zinc batteries, and lithium batteries. Among these, lithium batteries are receiving attention for their advantages of being able to charge and discharge freely, having a very low self-discharge rate, and having a high energy density, as they have almost no memory effect compared to nickel-based batteries.
[0007] During the battery manufacturing and assembly process, the sealing portion of the battery exterior may become deformed, wrinkled, or damaged, resulting in cosmetic defects (e.g., corner dents, dents, scratches, wrinkles, electrolyte contamination, etc.) in the battery pouch. Since pouch appearance defects can impact the reliability and / or safety of the battery, it is important to detect defects in the battery (e.g., pouch) before distribution.
[0008] Commonly used visual inspection devices to detect external defects in batteries (e.g., pouches) collect and train images of both good and defective battery types. The devices then inspect batteries undergoing manufacturing and assembly processes for the presence of external defects. However, if a new type of defect, not one of the previously learned types, occurs, the battery can be judged as good. Furthermore, because the incidence of new types of defects in the pouch manufacturing and assembly process is low, collecting suitable images of defective products for visual inspection device training is challenging.
[0009] Therefore, the present invention aims to prevent defect leakage by compensating for the detection uncertainty of a new type of defect.
[0010] The technical problems of the embodiments disclosed in this document are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the descriptions below.
[0011] According to an embodiment disclosed in the present document, a monitoring device includes a communication circuit, a processor, and a memory storing instructions, wherein the instructions, when executed by the processor, cause the monitoring device to obtain an image of a battery through the communication circuit, input the image into a first artificial intelligence model to determine whether the battery is defective, and input the image into a second artificial intelligence model based on a result of determining whether the image is defective, thereby determining whether the image has a different distribution from learning data for classifying the battery as normal among learning data used for learning the first artificial intelligence model, and determining whether to retrain the first artificial intelligence model based on a result of determining whether the image has a different distribution from the learning data.
[0012] An operating method of a monitoring device according to an embodiment disclosed in this document may include an operation of acquiring an image of a battery, an operation of inputting the image into a first artificial intelligence model to determine whether the battery is defective, an operation of inputting the image into a second artificial intelligence model based on a result of determining whether the image is defective, thereby determining whether the image has a different distribution from learning data for classifying the battery as normal among learning data used for learning the first artificial intelligence model, and an operation of determining whether to retrain the first artificial intelligence model based on a result of determining whether the image has a different distribution from the learning data.
[0013] According to various embodiments disclosed in this document, the monitoring device can detect a new type of defective product that is not included in the stored learning data, thereby preventing the leakage of the defective product.
[0014] According to various embodiments disclosed in this document, a monitoring device can improve the accuracy of defect detection by learning images of new types of defective products.
[0015] The effects of the monitoring device and its operating method according to the disclosure of this document are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art according to the disclosure of this document.
[0016] FIG. 1 is a block diagram of a monitoring device according to an embodiment of the present disclosure.
[0017] Figure 2 illustrates an artificial intelligence model according to one embodiment disclosed in this document.
[0018] FIG. 3 is a diagram illustrating learning data according to an embodiment disclosed in this document.
[0019] Figure 4 illustrates an artificial intelligence model according to one embodiment disclosed in this document.
[0020] FIG. 5 illustrates a normal and defective judgment process of a battery unit according to one embodiment disclosed in this document.
[0021] FIG. 6A and FIG. 6B illustrate a process for making a final judgment on whether a battery unit is good or bad using artificial intelligence models according to an embodiment disclosed in this document.
[0022] FIG. 7 is a flowchart illustrating an operation method of a monitoring device according to an embodiment of the present disclosure.
[0023] In connection with the description of the drawings, the same or similar reference numerals may be used for identical or similar components.
[0024] Hereinafter, embodiments of the present invention will be described with reference to the attached drawings. However, this is not intended to limit the present invention to specific embodiments, and it should be understood that the present invention encompasses various modifications, equivalents, and / or alternatives of the embodiments.
[0025] The embodiments and terminology used in this document are not intended to limit the technical features described in this document to a specific embodiment, but should be understood to encompass various modifications, equivalents, or alternatives of the embodiment. In connection with the description of the drawings, similar reference numerals may be used to refer to similar or related components. The singular form of a noun corresponding to an item may include one or more of the item, unless the relevant context clearly indicates otherwise.
[0026] In this document, the phrases "A or B", "at least one of A and B", "at least one of A or B", "A, B, or C", "at least one of A, B, and C", and "at least one of A, B, or C" can each include any one of the items listed together in that phrase, or all possible combinations thereof. Terms such as "first", "second", "first", "second", "A", "B", "(a)", or "(b)" may be used merely to distinguish the corresponding component from other corresponding components, and do not limit the corresponding components in any other respect (e.g., importance or order) unless specifically stated otherwise.
[0027] In this document, when a component (e.g., a first component) is referred to as being “connected,” “coupled,” or “connected,” with or without the terms “functionally” or “communicatively,” or “coupled” or “connected,” it means that the component can be connected to the other component directly (e.g., wired or wirelessly), or indirectly (e.g., via a third component).
[0028] The methods according to various embodiments disclosed in this document may be provided as included in a computer program product. The computer program product may be traded as a commodity between a seller and a buyer. The computer program product may be distributed in the form of a machine-readable storage medium (e.g., compact disc read-only memory, CD-ROM), or may be distributed online (e.g., downloaded or uploaded) through an application store or directly between two user devices. In the case of online distribution, at least a portion of the computer program product may be temporarily stored or temporarily generated in a machine-readable storage medium, such as the memory of a manufacturer's server, an application store's server, or an intermediary server.
[0029] According to the embodiments disclosed in this document, each component (e.g., a module or a program) of the above-described components may include one or more entities, and some of the entities may be separated and placed in other components. According to the embodiments disclosed in this document, one or more components or operations of the aforementioned components may be omitted, or one or more other components or operations may be added. Alternatively or additionally, a plurality of components (e.g., a module or a program) may be integrated into a single component. In this case, the integrated component may perform one or more functions of each of the plurality of components identically or similarly to those performed by the corresponding component among the plurality of components prior to the integration. According to the embodiments disclosed in this document, the operations performed by a module, program, or other component may be executed sequentially, in parallel, iteratively, or heuristically, or one or more of the operations may be executed in a different order, omitted, or one or more other operations may be added.
[0030] Figure 1 is a block diagram of a monitoring device (101) according to one embodiment of the present disclosure.
[0031] Referring to FIG. 1, a monitoring device (101) according to one embodiment disclosed in this document can be connected to an image acquisition device (103) and a user terminal (105) via wires and / or wirelessly.
[0032] In one embodiment, the connection between the monitoring device (101) and the image acquisition device (103) may be a communication connection via a wired and / or wireless network. In one embodiment, the wired network may be based on a local area network (LAN) communication or a power line communication. In one embodiment, the wireless network may be based on a short-range communication network (e.g., Bluetooth, wireless fidelity (WiFi), or infrared data association (IrDA)) or a wide-range communication network (cellular network, 4G network, 5G network).
[0033] In another embodiment, the connection between the battery abnormality diagnosis device (101) and the image acquisition device (103) may be a connection via a device-to-device communication method (e.g., a bus, a general purpose input and output (GPIO), a serial peripheral interface (SPI), or a mobile industry processor interface (MIPI)).
[0034] In one embodiment, the connection between the battery abnormality diagnosis device (101) and the user terminal (105) may be a communication connection via a wired and / or wireless network.
[0035] In one embodiment, the image acquisition device (103) can acquire an image of a battery unit (115). According to an embodiment, the battery unit (115) can include a secondary battery. According to an embodiment, the secondary battery can include a pouch-type secondary battery, a square secondary battery, and / or a cylindrical secondary battery.
[0036] The image acquisition device (103) can capture an image of the battery unit (115) by photographing the battery unit (115). According to one embodiment, the image acquisition device (103) can be implemented as a camera.
[0037] In one embodiment, the user terminal (105) may be a mobile device (e.g., a mobile phone, a laptop computer, a smart phone, a smart pad), or a personal computer (PC). In one embodiment, the user terminal (105) may be a terminal used by an administrator of the monitoring device (101).
[0038] In one embodiment, the monitoring device (101) may include a communication circuit (120), a memory (140), and a processor (150). According to an embodiment, the monitoring device (101) illustrated in FIG. 1 may further include at least one component (e.g., a display, an input device, or an output device) other than the components illustrated in FIG. 1.
[0039] In one embodiment, the communication circuit (120) establishes a wired communication channel and / or a wireless communication channel between the monitoring device (101) and the image acquisition device (103) and / or the user terminal (105), and can transmit and receive data with the image acquisition device (103) and / or the user terminal (105) through the established communication channel.
[0040] In one embodiment, the memory (140) may include volatile memory and / or non-volatile memory.
[0041] In one embodiment, the memory (140) may store data used by at least one component (e.g., processor (150)) of the monitoring device (101). For example, the data may include a program (130) (or instructions related thereto), input data, or output data. In one embodiment, the instructions, when executed by the processor (150), may cause the monitoring device (101) to perform operations defined by the instructions.
[0042] In one embodiment, the program (130) may include one or more software (e.g., an artificial intelligence model learning unit (141), an image acquisition unit (143), a diagnosis unit (145), and one or more artificial intelligence models (161, 165)).
[0043] In one embodiment, the processor (150) may include a central processing unit, an application processor, a graphics processing unit, a neural processing unit (NPU), an image signal processor, a sensor hub processor, or a communication processor.
[0044] In one embodiment, the processor (150) may execute software (e.g., an artificial intelligence model learning unit (141), an image acquisition unit (143), a diagnosis unit (145), and one or more artificial intelligence models (161, 165)) to control at least one other component (e.g., a hardware or software component) of a monitoring device (101) connected to the processor (150) and perform various data processing or calculations.
[0045] Hereinafter, with reference to FIGS. 2 to 6b, a method for a monitoring device (101) to diagnose an abnormality in a battery unit (115) through an artificial intelligence model learning unit (141), an image acquisition unit (143), a diagnosis unit (145), and one or more artificial intelligence models (161, 165) will be described.
[0046] First AI model training
[0047] Figure 2 illustrates an artificial intelligence model (161) according to one embodiment disclosed in this document. Figure 3 is a diagram illustrating learning data according to one embodiment disclosed in this document.
[0048] In one embodiment, the artificial intelligence model learning unit (141) may learn the artificial intelligence model (161) based on images (201, 205) of the battery unit (115) that have been previously acquired. Here, the images (201) and the images (205) may be previously acquired learning data. The images (201) may refer to images for classifying the battery unit (115) as normal among the learning data. The images (205) may refer to images for classifying the battery unit (115) as defective among the learning data. In one embodiment, the images (201) and the images (205) may include an entire image of the battery unit (115) or images of each part of the battery unit (115). In one embodiment, when the images (201) and the images (205) are composed of full images, the images (201) may be composed of images of a battery unit (115) classified as a normal product, and the images (205) may be composed of images of a battery unit (115) classified as a defective product. In one embodiment, when the images (201) and the images (205) are composed of partial images, the images (201) may be composed of images classified as normal among images obtained by dividing the full image of the battery unit (115) by inspection items (e.g., stamping, pressing, printed text, welding), and the images (205) may be composed of images classified as defective among images obtained by dividing the full image of the battery unit (115) by inspection items.
[0049] For example, referring to FIG. 3, images (310, 320, 330, 340) may be images of battery units (115) classified as defective. For example, images (310, 340) may be images in which there are scratches (325, 335) that are outside the normal range, and images (320, 330) may be images in which there are scratches (325, 335) that are outside the normal range. Similarly, images (350, 360) may be images of battery units (115) classified as normal.
[0050] In one embodiment, the artificial intelligence model (161) may be a model capable of multinomial classification (e.g., a model based on a convolutional neural network (CNN)). In one embodiment, the artificial intelligence model (161) may be trained to classify whether the battery unit (115) is normal or defective based on images (201, 205) of the battery unit (115). In one embodiment, the artificial intelligence model (161) may be trained to classify whether each inspection item of the battery unit (115) is normal or defective based on images (201, 205) of the battery unit (115).
[0051] In one embodiment, the artificial intelligence model (161) may include an input layer (210), a hidden layer (220), and an output layer (230). Here, the input layer (210) may input images (201) and images (205) of previously acquired battery units (115). The hidden layer (220) may have a structure in which multiple layers are sequentially connected. The output layer (230) may be a layer for outputting a multinomial classification result (209) of each of the images (201) and the images (205). In one embodiment, the multinomial classification result (209) may include probability values for classes (241, 243) according to inspection items of the images (201) and the images (205). For example, class (241) may represent the probability that the battery unit (115) is normal, and class (243) may represent the probability that the battery unit (115) is defective. In one embodiment, more than two classes (241, 243) may be provided. For example, classes may be further included to represent normality and defects for each inspection item (e.g., dent, scratch, or press).
[0052] In one embodiment, the artificial intelligence model learning unit (141) may adjust the parameters of the hidden layer (220) so that the obtained multinomial classification result (209) and the pre-classification result of each of the images (201) and the images (205) are identical or less than a reference difference value by inputting the images (201) and the images (205) into the artificial intelligence model (161).
[0053] Second AI model training
[0054] Figure 4 illustrates an artificial intelligence model (165) according to one embodiment disclosed in this document.
[0055] In one embodiment, the artificial intelligence model learning unit (141) may train the artificial intelligence model (165) based on images (201) of the previously acquired battery unit (115). Here, the images used to train the artificial intelligence model (165) may be images for classifying the battery unit (115) as normal among the learning data.
[0056] In one embodiment, the artificial intelligence model (165) may be a model capable of multinomial classification (e.g., a CNN-based model). In one embodiment, the artificial intelligence model (165) may be trained to output a similarity between images (201) of the battery unit (115) and normal images for each inspection item. Here, the similarity may refer to the maximum probability value according to the multinomial classification results obtained by the artificial intelligence model (165).
[0057] In one embodiment, the artificial intelligence model (165) can be trained to compare the similarity based on images (201) of battery units (115) with a reference value to distinguish whether the image is in-distribution (i.e., corresponding to an image of a previously learned good product) or out-of-distribution (i.e., corresponding to an image of a new type of defective product).
[0058] In one embodiment, the artificial intelligence model (165) may include an input layer (410), a hidden layer (420), and an output layer (430). Here, the input layer (410) may input images (201) of previously acquired battery units (115). The hidden layer (420) may have a structure in which multiple layers are sequentially connected. The output layer (430) may be a layer for outputting a multinomial classification result (409) of each of the images (201). In one embodiment, the multinomial classification result (409) may include probability values for classes according to inspection items of the images (201). The classes set in the artificial intelligence model (165) may be fewer than the classes set in the artificial intelligence model (161). For example, only classes indicating normality may exist in the artificial intelligence model (165).
[0059] In one embodiment, the artificial intelligence model learning unit (141) can adjust the parameters of the hidden layer (420) so that the maximum value (441) of the probability value according to the obtained multinomial classification result (409) exceeds the reference value (449) by inputting images (201) into the artificial intelligence model (165).
[0060] In one embodiment, the artificial intelligence model learning unit (141) can test the artificial intelligence model (165) using the images (205). For example, the artificial intelligence model learning unit (141) can check whether the maximum value (445) of the probability value according to the multinomial classification result (409) of the images (205) is less than or equal to the reference value (449). In one embodiment, when the maximum value (445) of the probability value according to the multinomial classification result (409) of the images (205) exceeds the reference value (449), the artificial intelligence model learning unit (141) can adjust the reference value (449) of the artificial intelligence model (165). For example, the artificial intelligence model learning unit (141) can adjust the reference value (449) based on a baseline algorithm or an ODIN (out of distribution detector for neural networks) algorithm. Here, the adjustment of the reference value (449) can be distinguished from the learning process for adjusting the parameters of the hidden layer (420).
[0061] Detection of defective battery units
[0062] FIG. 5 illustrates a process for determining whether a battery unit (115) is normal or defective according to an embodiment disclosed in this document. FIG. 6a and FIG. 6b illustrate a process for determining whether a battery unit (115) is good or defective using an artificial intelligence model (161) and an artificial intelligence model (165) according to an embodiment disclosed in this document.
[0063] In one embodiment, the processor (150) can detect a good product and / or a defective product among the battery units (115) through the diagnostic unit (145). The processor (150) can classify the battery units (115) as normal or defective through the diagnostic unit (145) to detect a good product and / or a defective product. The processor (150) can classify the battery units (115) as normal or defective through the diagnostic unit (145) based on previously stored learning data. Hereinafter, unless otherwise stated, the operations of the processor (150) can be understood as being performed through the diagnostic unit (145).
[0064] In one embodiment, the processor (150) can determine whether the battery unit (115) is normal or defective by performing a rule-based inspection (503) on the battery unit (115) based on an image (501) of the battery unit (115). Here, the rule-based inspection (503) may mean an image-based inspection for each inspection item related to the appearance of the battery unit (115) (e.g., presence of a jelly roll, presence of a stamp, presence of a press) or presence of a pressing. For example, the processor (150) may distinguish the pressing or pressing of a jelly roll based on the area and magnitude of the pressure applied to the jelly roll included in the image (501). For example, the pressing or pressing of the jelly roll on the image (501) may be distinguished by a color difference in the image. Here, the inspection items may include inspection items related to the presence of a jelly roll, a stamp, and a pressing, as well as the outer diameter size and the roundness of the positive electrode. In one embodiment, the criteria for determining whether the battery unit (115) is good or defective by external inspection may be set by the specifications of the battery unit (115).
[0065] In one embodiment, if the battery unit (115) is determined to be defective in the rule-based inspection (503), the processor (150) may inspect whether the battery unit (115) is normal or defective based on the artificial intelligence model-based inspection (505) based on the image (501). In one embodiment, the artificial intelligence model used in the artificial intelligence model-based inspection (505) may be distinguished from the artificial intelligence model (161) and the artificial intelligence model (165).
[0066] In one embodiment, the processor (150) may determine that the battery unit (115) is normal in the rule-based inspection (503) or normal in the artificial intelligence model-based inspection (505). However, the processor (150) may determine that the battery unit (115) is defective in the rule-based inspection (503) and the artificial intelligence model-based inspection (505).
[0067] The rule-based inspection (503) and the artificial intelligence model-based inspection (505) included in the process of FIG. 5 are examples, and a normal judgment (510) or a defective judgment (520) can be performed based on one of the rule-based inspection (503) or the artificial intelligence model-based inspection (505).
[0068] In the case of a normal judgment (510), the processor (150) can perform a classification operation according to Fig. 6a. In the case of a defective judgment (520), the processor (150) can perform a classification operation according to Fig. 6b.
[0069] Referring to FIGS. 6A and 6B, the processor (150) inputs an image of the battery unit (115) into an artificial intelligence model (161) to determine whether the battery unit (115) is normal or defective.
[0070] In one embodiment, if the judgment result of the artificial intelligence model (161) is the same as the previous judgment result, the processor (150) may not perform a judgment based on the artificial intelligence model (165). For example, if the battery unit (115) is determined to be normal based on the rule-based inspection (503) and / or the artificial intelligence model-based inspection (505), and the artificial intelligence model (161) also determines that the battery unit (115) is normal, the processor (150) may determine a normal classification (610) for the battery unit (115). For another example, if the battery unit (115) is determined to be defective based on the rule-based inspection (503) and / or the artificial intelligence model-based inspection (505), and the artificial intelligence model (161) also determines that the battery unit (115) is defective, the processor (150) may determine a defective classification (620) for the battery unit (115).
[0071] In one embodiment, if the judgment result of the artificial intelligence model (161) is different from the previous judgment result, the processor (150) may perform a judgment based on the artificial intelligence model (165). For example, if the battery unit (115) is judged to be defective based on the rule-based inspection (503) and the artificial intelligence model-based inspection (505), and the battery unit (115) is judged to be normal based on the artificial intelligence model (161), the processor (150) may make a final judgment on whether the battery unit (115) is normal or defective through the artificial intelligence model (165). In another example, if the battery unit (115) is judged to be normal based on the rule-based inspection (503) and the artificial intelligence model-based inspection (505), and the battery unit (115) is judged to be defective based on the artificial intelligence model (161), the processor (150) may make a final judgment on whether the battery unit (115) is normal or defective through the artificial intelligence model (165).
[0072] In one embodiment, the processor (150) may determine, based on the artificial intelligence model (165), whether the image of the battery unit (115) has a different distribution (i.e., out of distribution) from the learning data for classifying the battery unit (115) as normal among the learning data used for learning the artificial intelligence model (161). In one embodiment, the processor (150) may detect a new type of defective product by comparing the similarity obtained by inputting the image of the battery unit (115) into the artificial intelligence model (165) with a reference value. Here, the similarity may mean the maximum value of the probability value according to the multinomial classification result obtained by the artificial intelligence model (165). For example, when the similarity exceeds the reference value, the processor (150) may determine the battery unit (115) as normal. For another example, when the similarity is less than the reference value, the processor (150) may determine the battery unit (115) as defective. In one embodiment, a battery unit (115) reclassified as defective among the battery units (115) may correspond to a new type of defective product.
[0073] In one embodiment, the processor (150) can retrain the artificial intelligence model (161) based on the image when the battery unit (115) is reclassified as defective. In one embodiment, the image when the battery unit (115) is reclassified as defective can be utilized as an image for classifying the battery unit (115) as defective.
[0074] FIG. 7 is a flowchart illustrating an operating method of a monitoring device according to an embodiment of the present disclosure. FIG. 7 may be described with reference to FIGS. 1 to 6b.
[0075] Referring to FIG. 7, in operation 710, the monitoring device (101) may input an image into an artificial intelligence model (161) to determine whether the battery unit (115) is defective. Here, the artificial intelligence model (161) may be a model trained to classify the battery unit (115) as normal or defective based on previously acquired learning data. The previously acquired learning data may include images for classifying the battery unit (115) as normal and images for classifying the battery unit (115) as defective.
[0076] In operation 720, the monitoring device (101) can determine whether the image has a different distribution than the learning data for classifying the battery unit (115) as normal.
[0077] In one embodiment, the monitoring device (101) may determine whether the image has a different distribution from the training data for classifying the battery unit (115) as normal based on the result of inputting the artificial intelligence model (161) to determine whether the battery unit (115) is defective. For example, if the judgment result of the artificial intelligence model (161) is the same as the previous judgment result, the monitoring device (101) may not perform the judgment based on the artificial intelligence model (165). Here, the previous judgment may be based on the rule-based inspection (503) and / or the artificial intelligence model-based inspection (505).
[0078] In one embodiment, if the judgment result of the artificial intelligence model (161) is different from the previous judgment result, the monitoring device (101) can perform a judgment based on the artificial intelligence model (165).
[0079] In one embodiment, the monitoring device (101) can determine, based on the artificial intelligence model (165), whether the image of the battery unit (115) has a different distribution (i.e., out of distribution) from the learning data for classifying the battery unit (115) as normal among the learning data used to train the artificial intelligence model (161). In one embodiment, the monitoring device (101) can detect a new type of defective product by comparing the similarity obtained by inputting the image of the battery unit (115) into the artificial intelligence model (165) with a reference value. Here, the similarity may mean the maximum value of the probability value according to the multinomial classification result obtained by the artificial intelligence model (165). For example, when the similarity exceeds the reference value, the monitoring device (101) can determine the battery unit (115) as normal. For another example, when the similarity is less than the reference value, the monitoring device (101) can determine the battery unit (115) as defective. In one embodiment, a battery unit (115) reclassified as defective among the battery units (115) may correspond to a new type of defective product.
[0080] In operation 730, the monitoring device (101) may determine whether to retrain the artificial intelligence model (161) based on the judgment result. In one embodiment, the monitoring device (101) may determine that retraining the artificial intelligence model (161) is necessary if the image has a different distribution (i.e., out of distribution) from the training data.
Claims
1. In the monitoring device, communication circuit, processor, A memory for storing instructions, wherein the instructions, when executed by the processor, cause the monitoring device to: Through the above communication circuit, an image of the battery is acquired, By inputting the above image into the first artificial intelligence model, it is determined whether the battery is defective or not, By inputting the image into a second artificial intelligence model based on the result of determining whether the image is defective, it is determined whether the image has a different distribution from the learning data for classifying the battery as normal among the learning data used for learning the first artificial intelligence model, and Based on the result of determining whether the above learning data has a different distribution, it is determined whether to retrain the first artificial intelligence model. Monitoring device.
2. In claim 1, The above instructions, when executed by the processor, cause the monitoring device to: Determine whether the battery is defective based on an algorithm different from the first artificial intelligence model and the second artificial intelligence model, If the judgment result based on the above different algorithm is different from the judgment result through the first artificial intelligence model, the second artificial intelligence model is used to determine whether the image has the different distribution. Monitoring device.
3. In claim 2, The above instructions, when executed by the processor, cause the monitoring device to: Determine whether the image is defective based on a rule-based algorithm, and If the judgment result based on the rule-based algorithm indicates that the battery is defective, the image is judged to be defective through an artificial intelligence model other than the first artificial intelligence model and the second artificial intelligence model. Monitoring device.
4. In claim 1, The above instructions, when executed by the processor, cause the monitoring device to: If the image has the same distribution as the training data, the battery is finally determined to be good, and if the image has a different distribution from the training data, the battery is finally determined to be defective. Monitoring device.
5. In claim 1, The second artificial intelligence model is trained based on the training data to classify the battery as normal, The first artificial intelligence model is learned based on the learning data for classifying the battery as normal and the learning data for classifying the battery as defective. Monitoring device.
6. In the method of operating the monitoring device, The action of acquiring an image of the battery, An action of inputting the above image into a first artificial intelligence model to determine whether the battery is defective; An operation of inputting the image into a second artificial intelligence model based on the result of determining whether the image is defective, thereby determining whether the image has a different distribution from the learning data for classifying the battery as normal among the learning data used for learning the first artificial intelligence model, and An operation for determining whether to retrain the first artificial intelligence model based on the result of determining whether the first artificial intelligence model has a different distribution from the learning data is included. How it works.
7. In claim 6, Further comprising an operation of determining whether the battery is defective based on an algorithm different from the first artificial intelligence model and the second artificial intelligence model, The action of determining whether the above different distributions are present is If the judgment result based on the other algorithm is different from the judgment result through the first artificial intelligence model, an operation of determining whether the image has the different distribution through the second artificial intelligence model is included. How it works.
8. In claim 7, the operation of determining whether the battery is defective based on the other algorithm comprises: An operation for determining whether the image is defective or not based on a rule-based algorithm, and If the judgment result based on the rule-based algorithm indicates that the battery is defective, an operation of judging whether the image is defective is included through an artificial intelligence model other than the first artificial intelligence model and the second artificial intelligence model. How it works.
9. In claim 6, If the image has the same distribution as the training data, it is finally determined that the battery is a good product, and if the image has a different distribution from the training data, it is finally determined that the battery is a defective product. How it works.
10. In claim 6, The second artificial intelligence model is trained based on the training data to classify the battery as normal, The first artificial intelligence model is learned based on the learning data for classifying the battery as normal and the learning data for classifying the battery as defective. How it works.
Citation Information
Patent Citations
Monitoring device and operating method thereof
KR102932303B1
Visual inspection method for circumferential pit of cylindrical lithium battery
CN109191421A
System of token issuance by reflecting position information for recombination of pieces
KR1020240019897A
Copper wire for alternating current having a continuous hollow in the longitudinal direction, and methods for manufacturing the same
KR1020240049665A
Electronic device and operation method of electronic device for detecting and classifying object
KR102507273B1
