Machine vision unit with easy customization

Abstract

The present invention relates to a machine vision unit that is easy to customize and adjust, and more specifically, to an intelligent machine vision unit that is easy to customize and adjust, and a control method thereof, which quantifies the mechanical, electronic, and operational status of equipment and the instability of the inspection environment in real time, and dynamically optimizes image acquisition conditions and defect judgment criteria based thereon to maximize inspection reliability and operational efficiency.

Description

Technology Field

[0001] The present invention relates to a machine vision unit that is easy to customize and adjust, and more specifically, to an intelligent machine vision unit that is easy to customize and adjust, and a control method thereof, which quantifies the mechanical, electronic, and operational status of equipment and the instability of the inspection environment in real time, and dynamically optimizes image acquisition conditions and defect judgment criteria based thereon to maximize inspection reliability and operational efficiency. Background Technology

[0002] With the recent rapid growth of the electric vehicle and energy storage system (ESS) markets, the demands for the quality and stability of batteries, which are core components, are higher than ever before. Consequently, machine vision-based automated inspection equipment is being introduced as an essential component in the production process to automatically detect minute scratches, foreign substances, and deformations on battery surfaces.

[0003] Conventional machine vision units, which are easy to customize, primarily utilized a method of capturing images using preset camera parameters (exposure, resolution, etc.) under fixed lighting conditions and determining the presence or absence of defects through static algorithms. While this approach could guarantee consistent performance in ideal environments, it faced various problems in actual mass production lines.

[0004] First, minute vibrations occurring on the production line or mechanical play within the equipment itself caused image blur at the moment of capture, leading to the missing of minute defects or over-detection where normal surfaces were mistaken for defects. Second, factors such as reduced lighting intensity due to prolonged operation and the aging of camera sensors undermined the consistency of inspection results, and periodic manual calibration to compensate for this resulted in decreased productivity. Third, the failure to effectively account for the surface reflection characteristics that differ by battery model led to a problem where the misreading rate of batteries made of specific materials surged.

[0005] In conclusion, conventional technology had a fundamental limitation in guaranteeing the reliability of inspection results because it lacked the ability to actively detect and respond to dynamic changes in the inspection environment or the deterioration of the equipment itself. The problem to be solved

[0006] The present invention was devised to solve the problems of the prior art as described above, and aims to solve the following problems.

[0007] The first objective is to provide an easily customizable machine vision unit capable of detecting physical and temporal instabilities, such as micro-vibrations and system timing errors occurring in actual production environments, in real time and actively responding to them to acquire clear and reliable images without data distortion even in unstable environments.

[0008] The second task is to provide a machine vision unit that is easy to customize and adjust, which can quantitatively evaluate measurement uncertainty caused by sensor aging due to long-term operation of equipment or the unique optical characteristics of the inspection target, and reflect this in real-time in the final defect judgment criteria, thereby preventing potential defect leakage and over-detection that may occur due to system degradation.

[0009] The third objective is to provide an easily customizable machine vision unit that maximizes equipment uptime and operational efficiency by comprehensively diagnosing various aspects of the equipment's condition, such as mechanical wear, electronic component degradation, and poor operational management, to prevent unexpected breakdowns in advance, minimize production stoppages even in the event of partial component failure, and self-restoring functions.

[0010] The problems of the present invention are not limited to those mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0012] According to an embodiment of the present invention for solving the above problem

[0013] A battery mounting unit for mounting a battery from the outside;

[0014] A machine vision unit for capturing a surface image of the above battery;

[0015] A control unit for adjusting the position or angle of the machine vision unit; and

[0016] A control unit that controls the operation of the battery mounting unit, the machine vision unit, and the adjustment unit; is included.

[0017] The above control unit is,

[0018] A step of determining the stability grade at the time of inspection based on sensor data received from the battery mounting unit or the control unit;

[0019] A step of dynamically determining the number of image acquisitions of the machine vision unit according to the above-determined stability grade;

[0020] A step of determining the uncertainty level of the system based on images acquired a number of times determined above and accumulated operational history data of the equipment; and

[0021] The method is characterized by performing a step of determining whether there is a final defect by correcting the initial defect score calculated through image analysis according to the uncertainty grade determined above, and

[0022] The above control unit is,

[0023] A micro-sleep probability (MSP) is calculated based on vibration data received from an accelerometer sensor embedded in the battery mounting unit, and

[0024] The camera-actuator delay jitter (CALJ) is calculated based on positioning feedback data received from the above-mentioned control unit, and

[0025] Characterized by determining the stability grade as 'caution' when either the above MSP or the above CALJ exceeds a preset 'caution threshold', and determining the number of image acquisitions as a plurality of first acquisitions.

[0026] The above control unit is,

[0027] When the above stability rating is determined to be 'caution', a clarity score is calculated for each of the multiple images acquired in the above first number of acquisitions, and

[0028] The method is characterized by minimizing the effect of image blur that may occur due to the micro-slip or jitter by selecting the single image with the highest calculated sharpness score as the final analysis target.

[0029] The above control unit is,

[0030] The automatic calibration module embedded in the above equipment calculates the Accumulated Calibration Deviation (ACD) based on the accumulated calibration amount log data, and

[0031] Check the pre-set Reflective Anisotropic Dispersion (RAV) value according to the unique surface characteristics of the above battery model, and

[0032] It is characterized by determining the uncertainty grade as 'normal' if either the above ACD or the above RAV exceeds a preset 'normal threshold', and

[0033] The above control unit is,

[0034] If the above uncertainty level is determined to be 'moderate',

[0035] A final defect score that is adjusted upward is calculated by multiplying the above initial defect score by a preset first correction factor greater than 1, and

[0036] It is characterized by determining whether there is a final defect by comparing the upwardly adjusted final defect score with a preset defect judgment threshold, thereby preventing defects from being underestimated and leaked due to the correction cumulative deviation or the reflection anisotropic variance.

[0037] The above control unit is,

[0038] If the above ACD and the above RAV simultaneously exceed the 'normal threshold', or if either the above ACD or the above RAV exceeds a 'risk threshold' higher than the 'normal threshold', the uncertainty class is determined as 'suspicious', and

[0039] It may be characterized by determining whether a measurement result is defective under complex or severe uncertainty based on a more conservative standard by significantly increasing the final defect score by multiplying the initial defect score by a second correction factor larger than the first correction factor. Effects of the invention

[0041] According to the present invention for solving the above-mentioned problems, the following effects can be expected.

[0042] Inspection reliability can be dramatically improved. The present invention detects various instability factors that impair inspection quality, such as micro-vibrations and system jitter, in real time, and dynamically applies an optimal shooting method, such as multi-frame acquisition, accordingly. Through this, consistent and accurate inspection results can be obtained even amidst environmental noise, which was a cause of errors in conventional technology.

[0043] It is possible to prevent the leakage of defective products and increase production yield. This invention quantifies even the aging status of equipment or the diffuse reflection characteristics of the battery surface to evaluate measurement uncertainty and conservatively reflects this in the final judgment score. As a result, it fundamentally prevents critical errors where boundary defects are misjudged as good products due to system deterioration, thereby enhancing product quality stability.

[0044] The total cost of ownership (TCO) of the equipment can be reduced and the utilization rate maximized. By comprehensively diagnosing the mechanical, electronic, and operational status of the equipment to enable predictive maintenance based on actual conditions, the present invention reduces unnecessary parts replacement costs and minimizes production downtime caused by unexpected breakdowns. Furthermore, by allowing the system to automatically reconfigure its functions to continue production in the event of a failure in some lighting units, it significantly improves the resilience of the equipment and the operational efficiency of the entire production line.

[0045] The effects according to the present invention are not limited to those exemplified above, and a wider variety of effects are included within the present invention. Brief explanation of the drawing

[0047] Figure 1 illustrates an overall relationship diagram according to the present invention. Figure 2 illustrates a flowchart between all components according to the present invention. Figure 3 illustrates a flowchart for calculating the number of dynamic frames acquired according to the present invention. Figure 4 illustrates a flowchart for calculating an adjustment fault score reflecting system uncertainty according to the present invention. FIG. 5 illustrates a flowchart for calculating the maintenance requirement index according to the present invention. Figure 6 illustrates a flowchart of the lighting system reconstruction optimization according to the present invention. Specific details for implementing the invention

[0048] Hereinafter, various embodiments are described in more detail with reference to the attached drawings. The embodiments described in this specification may be modified in various ways. Specific embodiments may be depicted in the drawings and described in detail in the detailed description. However, specific embodiments disclosed in the attached drawings are intended only to facilitate understanding of various embodiments. Accordingly, the technical concept is not limited by specific embodiments disclosed in the attached drawings, and it should be understood that it includes all equivalents or substitutions that fall within the spirit and scope of the invention.

[0049] Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but these components are not limited by the aforementioned terms. The aforementioned terms are used solely for the purpose of distinguishing one component from another.

[0050] Functions related to artificial intelligence according to the present disclosure are operated through a processor and memory. The processor may be composed of one or more processors. In this case, the one or more processors may be general-purpose processors such as CPUs, APs, and DSPs (Digital Signal Processors), graphics-dedicated processors such as GPUs and VPUs (Vision Processing Units), or artificial intelligence-dedicated processors such as NPUs. The one or more processors control the processing of input data according to predefined operation rules or artificial intelligence models stored in memory. Alternatively, if the one or more processors are artificial intelligence-dedicated processors, the artificial intelligence-dedicated processors may be designed with a hardware structure specialized for processing a specific artificial intelligence model.

[0051] The predefined rules of operation or artificial intelligence models are characterized by being created through learning. Here, being created through learning means that a basic artificial intelligence model is trained using a number of training data by a learning algorithm, thereby creating predefined rules of operation or artificial intelligence models configured to perform desired characteristics (or objectives). Such learning may be performed on the device itself where the artificial intelligence according to the present disclosure is executed, or it may be performed through a separate server and / or system. Examples of learning algorithms include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but are not limited to the examples described above.

[0052] An artificial intelligence model can be composed of multiple neural network layers. Each of the multiple neural network layers has multiple nodes and weight values, and performs neural network operations through calculations between the results of previous layers and the multiple weights. The multiple weights possessed by the multiple neural network layers can be optimized based on the learning results of the artificial intelligence model. For example, multiple weights can be updated so that the loss value or cost value obtained by the artificial intelligence model during the learning process is reduced or minimized. Additionally, to minimize the loss value or cost value, multiple weights can be updated in a direction that minimizes the gradient associated with the loss value or cost value. Artificial neural networks may include deep neural networks (DNNs), such as Convolutional Neural Networks (CNNs), Deep Neural Networks (DNNs), Recurrent Neural Networks (RNNs), Restricted Boltzmann Machines (RBMs), Deep Belief Networks (DBNs), Bidirectional Recurrent Deep Neural Networks (BRDNNs), or Deep Q-Networks, but are not limited to the examples mentioned above.

[0053] A network is a network that serves as a transmission path for web pages; it may be a closed network such as a LAN (Local Area Network) or WAN (Wide Area Network), but it is desirable for it to be an open network such as the Internet. The Internet refers to a global open computer network structure that provides the TCP / IP protocol and various services existing at its upper layers, namely HTTP (HyperText Transfer Protocol), Telnet, FTP (File Transfer Protocol), DNS (Domain Name System), SMTP (Simple Mail Transfer Protocol), SNMP (Simple Network Management Protocol), NFS (Network File Service), and NIS (Network Information Service).

[0054] Terminals can be implemented in various forms. For example, the terminals described in this specification may include mobile terminals such as smartphones, tablet PCs, PDAs, portable multimedia players, and MP3 players, as well as fixed terminals such as smart TVs and desktop computers.

[0055] In this specification, terms such as “comprising” or “having” are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. When a component is described as being “connected” or “connected” to another component, it should be understood that it may be directly connected to or connected to that other component, or that there may be other components in between. On the other hand, when a component is described as being “directly connected” or “directly connected” to another component, it should be understood that there are no other components in between.

[0056] Meanwhile, a "module" or "part" for a component as used in this specification performs at least one function or operation. Furthermore, a "module" or "part" may perform a function or operation by hardware, software, or a combination of hardware and software. Additionally, a plurality of "modules" or a plurality of "parts," excluding a "module" or "part" that must be performed on specific hardware or on at least one processor, may be integrated into at least one module. A singular expression includes a plural expression unless the context clearly indicates otherwise.

[0057] In addition, power, power transmission, and control therefor for the following assembly configurations and embodiments, including "by control," follow conventional technology including terminals, applications, hardware control modules, etc., so they are omitted to avoid redundancy.

[0058] In addition, the operation embodiments and configurations described in a general manner without being explained in detail below follow the prior art and are omitted in order to focus on describing the purpose of the present invention and the resulting effects.

[0059] Furthermore, in describing the present invention, if it is determined that a detailed description of related known functions or configurations may unnecessarily obscure the essence of the invention, such detailed description is abbreviated or omitted.

[0060] The machine vision unit with easy customizable control according to the present invention is equipment for determining good or defective products by photographing and analyzing the surface condition of a battery to be inspected, and comprises a machine vision unit (100), a control unit (200), a battery mounting unit (300), and a control unit (400). Specifically, the image acquisition unit (110) may be configured as a high-speed industrial camera that supports a shooting speed of at least 100 frames per second, so as to be able to complete 3 or 5 multi-frame shots required for determining a 'caution' or 'danger' grade, as described later, within a short time in milliseconds according to the command of the control unit (400). In addition, the lighting unit (120) includes a function in which each light source receives an individual control signal from the control unit (400) to independently adjust its light intensity and irradiation angle in order to perform the 'lighting reconstruction optimization' process described later.

[0061] The machine vision unit (100) is a hardware configuration for capturing images to inspect the surface condition of a battery and includes an image acquisition unit (110) and a lighting unit (120). In addition to capturing the battery, which is the subject, and generating image data, the image acquisition unit (110) supports a high-speed continuous shooting mode of tens of frames per second or more according to the command of the control unit (400), and includes the function of sequentially acquiring multiple frames synchronized with a precise trigger signal. In addition to assisting in the acquisition of a clear image by irradiating uniform light onto the subject, the lighting unit (120) is composed of a plurality of physically separated independent light sources, and each light source receives an individual control signal from the control unit (400) and includes the function of independently adjusting its light intensity and irradiation angle.

[0062] The above-described control unit (200) is a hardware configuration that changes the physical position, angle, or focal length of the machine vision unit (100). In addition to simply changing the physical position, it includes a high-precision servo motor and encoder with positioning precision in micrometers, and includes a function to transmit a feedback signal indicating its current position and a 'positioning complete' status to the control unit (400) in real time with millisecond precision. This high-precision feedback function is an essential component of the present invention. This is intended to solve the technical problem of guaranteeing the temporal reliability of inspection data by going beyond simply moving to a target position as in conventional technology, and instead quantifying the temporal irregularity (jitter) itself between the time of movement completion and the actual shooting time as a key judgment criterion called 'Camera-Actuator Delay Jitter (CALJ)'.

[0063] The battery mounting unit (300) is a hardware configuration that stably fixes the battery to be inspected at a designated location. In addition to simply fixing the battery at a designated location, it includes a 3-axis accelerometer sensor to measure micro-vibrations of the mounted battery in real time, includes multiple pressure sensors to detect the contact pressure distribution between the battery and the jig, and transmits the measured sensor data to the control unit (400) to provide basic data for calculating the micro-slip probability and the mounting degree of freedom instability coefficient. Here, the 'multiple pressure sensors' are configured to be placed at least five times at the four corners and the center of the battery to detect the state in which the battery tilts to one side or makes uneven contact during mounting. The pressure distribution data measured from the pressure sensors can be used as a weight when calculating the mounting degree of freedom instability coefficient (MFI). For example, if the contact pressure in a specific direction is measured to be abnormally low, it is determined that there is a high probability that mechanical play has occurred in that direction, and a higher weight is assigned to the position error of that axis to improve the reliability of the MFI value.

[0064] The control unit (400) is a central processing unit that controls the overall operation of the machine vision unit that is easy to customize and includes an inspection condition control module (410), an automatic correction module (420), a data collection module (430), and an analysis judgment module (440). The inspection condition control module (410) is a software module that actively controls image shooting conditions in response to an inspection environment that changes in real time. In addition to simply controlling shooting conditions, it receives a 'number of frames acquired' dynamically determined from the analysis judgment module (440) to control multiple shots of the image acquisition unit (110), and receives new light quantity and angle setting values ​​derived as a result of the 'lighting reconstruction optimization' process to transmit a command to reconstruct the lighting unit (120) in real time. The above automatic correction module (420) is a software module that automatically corrects hardware state changes that may occur due to the long-term operation of the equipment. In addition to simply automatically correcting hardware state changes, it records and stores the types and amounts of all correction actions performed as time-series log data in an internal database, and transmits a command to provide the stored log data to the analysis judgment module (440) as original data for calculating the 'accumulated correction deviation'. The above data collection module (430) is a software module that receives and stores image data generated from the image acquisition unit (110) of the machine vision unit (100). In addition to simply receiving and storing image data, it transmits a command to the analysis judgment module (440) to integrally collect all heterogeneous data necessary for calculating the core judgment criteria of the invention, including micro-vibration and pressure sensor data from the battery mounting unit (300), position determination feedback data from the control unit (200), and correction log data from the automatic correction module (420).The analysis judgment module (440) is a software module that analyzes image data received from the data collection module (430) to make a final determination of whether the battery is good or not. In addition to simply analyzing image data to determine whether the battery is good or not, it is configured to perform all stepwise logical processing procedures described later in processes 1, 2, and 3, which are the main computational logic of the present invention. Specifically, the analysis judgment module (440) receives a heterogeneous sensor data stream synchronized in a time series from the data collection module (430). For the data stream, it performs operations sequentially by assigning priorities in the order of process 3 (maintenance), process 1 (frame acquisition), and process 2 (defect determination). This is based on a logical causal relationship in which the fundamental soundness of the equipment is diagnosed first, optimal shooting conditions are determined under that state, and finally a judgment with corrected reliability is made. The data collection module performs a preprocessing function by assigning a common timestamp to sensor data collected at different intervals and storing it in a buffer, thereby synchronizing all data at a specific point in time and providing it to the analysis judgment module. Based on various data transmitted from the data collection module (430), it calculates key judgment criteria in real time, such as micro-slip probability, highlight variation coefficient, camera-actuator delay jitter, calibration cumulative deviation, seating degree of freedom instability coefficient, and maintenance debt accumulation index; determines the 'number of frames acquired' by comparing with a pre-set rule table and dynamic threshold values; adjusts the initial defect score by reflecting 'system uncertainty'; and finally determines the 'maintenance requirement grade' and transmits a command to output the result to the inspection condition control module (410) or an external interface, respectively.

[0065] The control unit (400) of the present invention calculates and evaluates three judgment criteria in real time that indicate physical, optical, and temporal stability at the time of inspection in order to prevent potential data distortion that may occur with a single-frame shooting method. The first judgment criterion includes a micro-slip probability, which is an indicator that quantifies the possibility that a battery fixed to a battery mounting unit (300) may move slightly at the moment of inspection due to external vibration or internal stress. The micro-slip probability is intended to prevent the loss of fine features of defects or the creation of false defects due to image motion blur, and a high micro-slip probability means that physical stability has deteriorated. The second judgment criterion includes a highlight variation coefficient, which indicates how much the location or area of ​​a strong light reflection region within the image fluctuates due to curvature of the battery surface, fine foreign matter, or fine flickering of lighting. The highlight variation coefficient is intended to prevent optical misinterpretation in which defects are obscured or the highlight itself is mistaken for a defect due to irregular changes in the highlight area, and a high highlight variation coefficient means that optical stability has deteriorated. The third judgment criterion includes camera-actuator delay jitter, which indicates how irregularly the time delay between the signal that the control unit (200) has reached the target position and the moment the image acquisition unit (110) begins actual shooting varies. The purpose of the camera-actuator delay jitter is to ensure that data is acquired at the correct position and timing, thereby preventing a decrease in reproducibility during repeated inspections caused by minute positional errors, and a high camera-actuator delay jitter means that the temporal synchronization stability of the system has decreased.

[0066] The present invention overcomes the limitations of a static single-frame acquisition method by integrating the three judgment criteria mentioned above. The technical basis lies in fundamentally improving the reliability of the original data by acquiring multiple frames when instability is detected and selecting the best data among them or fusing the data, unlike a single shot which is valid only under ideal conditions. To this end, the control unit (400) includes internal logic for determining an inspection stability grade divided into three stages: 'stable', 'caution', and 'danger', compares the values ​​of each judgment criterion with a preset threshold value, determines the grade based on the combination of the comparison results, and performs a procedure to differentially allocate the number of frames acquired according to the final grade.

[0067] The above judgment criteria are physically measured and quantified through a specific method. The micro-slip probability is calculated by collecting acceleration data at 1 kHz for 100 milliseconds from a 3-axis MEMS accelerometer sensor attached to the bottom of the jig of the battery mounting unit (300) until just before shooting. The micro-slip probability value is calculated through a 'vibration-image quality correlation model' established during the equipment calibration phase. To establish the 'vibration-image quality correlation model,' vibrations of various intensities are intentionally applied to a standard specimen using a controllable micro-vibration generator during the calibration phase, the standard deviation of acceleration data is measured for each vibration intensity, and the clarity of the captured image is simultaneously measured quantitatively. Subsequently, regression analysis is performed based on the measured multiple data pairs to generate an empirical function model that predicts the rate of image clarity degradation according to the acceleration value. The function model is automatically selected as the optimal model based on the distribution characteristics of the data pairs collected during the calibration phase. For example, a linear regression model is used when the data distribution is close to linear; however, if a non-linear relationship is observed where image quality degradation rapidly intensifies beyond a certain vibration intensity, a polynomial regression model or an exponential function model is applied to more accurately model the 'vibration-image quality correlation.' Based on these models, when the real-time measured acceleration standard deviation is 0.05G, the expected image sharpness degradation rate is calculated as 15%, which is normalized to a value between 0 and 1 to finally yield a micro-slip probability value of 0.6. The above highlight coefficient of variation is obtained by performing five consecutive low-resolution preliminary shots prior to the main shoot. In each preliminary image, pixels with a luminance value exceeding 240 (based on a 255 standard) are clustered as highlight areas, and the total area of ​​the highlight region for each frame is calculated. For example, if the average of the area measured across five frames is 1046.8 pixels and the standard deviation is 47.6 pixels, the coefficient of variation is 0.045 is calculated. The camera-actuator delay jitter is calculated by recording the difference between the timestamp at which the servo motor of the control unit (200) generated a 'positioning complete' flag and the timestamp at which the image sensor of the image acquisition unit (110) started exposure, through the system internal log for the last 10 inspections. For example, if the standard deviation of the measured time delay values ​​is 0.76 milliseconds, it is quantitatively determined that the jitter has increased.

[0068] Specifically, the use of regression analysis to construct a 'vibration-image quality correlation model' in this invention is intended to generate a generalized model capable of predicting image quality degradation even for arbitrary, unmeasured vibration values. Conventional methods that simply store experimental data in the form of lookup tables have limitations in that they can only correspond to specific vibration intensities used in the experiment and cannot accurately predict values ​​in between. On the other hand, regression analysis identifies the mathematical relationships inherent in multiple collected data pairs and expresses them as a single continuous function model.

[0069] The table below is an example comparing the prediction results of the lookup table method and the regression analysis model method for new, unmeasured vibration values ​​based on 10 pairs of experimental data. It is assumed that the lookup table predicts the nearest experimental value.

[0070]

[0071] As is clearly shown in the results of the above examples, the lookup table method predicts new vibration values ​​(e.g., 0.025G) existing between experimental data by substituting them with the result of the nearest experimental value (0.02G or 0.03G), resulting in a significant error compared to the actual value. On the other hand, the functional model generated through regression analysis (e.g., quality degradation rate) 160 × vibration - 0.5) performs continuous interpolation predictions that reflect the overall trend of the data even for these unmeasured values, thereby producing results that are very close to the actual measured values. This provides a technical advantage that enables a more precise and flexible response to vibrations in diverse and unpredictable actual production environments by deriving a continuous prediction model from discontinuous data points.

[0072] According to the 'inspection stability grade' determined through the above logical processing procedure, the frame acquisition method is specifically changed as follows. If the grade is determined to be 'Stable,' it means that the system is stable and therefore no additional data acquisition is necessary, so a command is transmitted to proceed with shooting with the basic frame acquisition count of 1. If the grade is determined to be 'Caution,' high-speed burst shooting is performed with the first acquisition count of 3, a sharpness score is calculated for each of the 3 captured frames, and a command is transmitted to select the single frame with the highest score as the final analysis target. If the grade is determined to be 'Dangerous,' high-speed burst shooting is performed with the second acquisition count of 5, and after aligning the 5 frames by pixel, image stacking and averaging techniques are applied to generate a single composite image with reduced noise and enhanced fine features, which is then used as the final analysis target.

[0073] Taking a scenario of a production line where micro-vibrations occur intermittently as an example, first, after the battery to be inspected is fixed to the battery mounting unit (300), the control unit (400) collects judgment criteria data for 100 milliseconds before shooting. When the measurement results show that the micro-slip probability is 0.7, the highlight variation coefficient is 0.03, and the camera-actuator delay jitter is 1.2 milliseconds, the control unit (400) refers to an internal judgment rule table. Since the micro-slip probability value exceeds the 'caution threshold' of 0.5, the overall 'inspection stability grade' is determined as 'caution' even if other indicators are stable. Based on the 'caution' grade, a command is transmitted to dynamically increase the number of frame acquisitions from the default 1 to 3, and the image acquisition unit (110) continuously captures 3 images. The analysis judgment module (440) selects the frame with the highest clarity score among the 3 images, which has the least influence from micro-vibration, and performs a final defect analysis based on the selected frame.

[0074] If valid data is not received from the accelerometer sensor required to calculate the micro-slip probability, the system considers the indicator to be in the most conservative 'danger' state or automatically sets it to a 'caution' level according to a predefined safety default value to perform control to prepare for potential danger. Additionally, if the 'danger' level is determined to be 10 or more times consecutively, which is a preset number, it is highly likely that this is a fundamental problem with the equipment rather than a temporary issue, so the system automatically sends a maintenance request signal to the upper system or outputs an inspection warning to the operator.

[0075] When comparing the operational results of the prior art and the present invention, both technologies yield a normal judgment in ideal environments; however, in environments where micro-vibrations occur, the prior art faces a problem where the probability of non-detection or over-detection increases. In contrast, the present invention maintains an accurate judgment by determining a 'caution' grade and selecting the clearest frame after capturing 3 frames, thereby analyzing the data unaffected by blur. Furthermore, in environments where lighting flicker occurs, the prior art causes non-detection failures as momentary highlights obscure micro-defects; however, the present invention succeeds in defect detection by determining a 'danger' grade, capturing 5 frames, and averaging them to analyze a stable image in which momentary changes in highlights are offset.

[0076] [Table 1]

[0077]

[0078] The operation of the present invention can be specifically exemplified through simulation. In the first embodiment, when the micro-slip probability is 0.21, the highlight variation coefficient is 0.02, and the camera-actuator delay jitter is 1.1 milliseconds, all indicators are below the 'caution threshold,' so it is determined to be 'stable' and 1 frame is acquired. In the second embodiment, when the micro-slip probability is 0.55 and exceeds the 'caution threshold,' it is determined to be 'caution' and 3 frames are acquired. In the fifth embodiment, when the micro-slip probability is 0.61 and the highlight variation coefficient is 0.07, and both indicators simultaneously exceed the 'caution threshold,' it is determined to be 'dangerous' and 5 frames are acquired. In the sixth embodiment, when the micro-slip probability is 0.82 and exceeds the 'dangerous threshold,' it is determined to be 'dangerous' solely and 5 frames are acquired.

[0079] [Table 2]

[0080]

[0081] In the above rule table, if conditions R-03 and R-04 are satisfied, they take precedence over R-02.

[0082] The control unit (400) operates according to an internal judgment rule table. As a first rule, if all conditions are satisfied—that the micro-slip probability is less than 0.5, the highlight variation coefficient is less than 0.05, and the camera-actuator delay jitter is less than 2.5 milliseconds—it is determined to be a 'stable' grade and a command to set the frame acquisition count to 1 is transmitted. As a second rule, if any one of the conditions is satisfied—that the micro-slip probability is 0.5 or higher, the highlight variation coefficient is 0.05 or higher, or the camera-actuator delay jitter is 2.5 milliseconds or higher—it is determined to be a 'caution' grade and a command to set the frame acquisition count to 3 is transmitted. As a third rule, if multiple conditions corresponding to the 'caution' grade conditions are satisfied simultaneously, it is determined to be a 'dangerous' grade and a command to set the frame acquisition count to 5 is transmitted. As the fourth rule, if any of the following conditions are met—that is, the micro-slip probability is 0.8 or higher, the highlight variation coefficient is 0.1 or higher, or the camera-actuator delay jitter is 4.0 milliseconds or higher—the case is classified as 'Dangerous' and the number of frames acquired is set to 5, with the third and fourth rules applied in priority over the second rule. The priority of the rule table above is based on the following technical judgment. When a single metric exceeds the 'Dangerous Threshold' (R-04), it has the highest priority as it signifies that a clear and serious problem has occurred in the system. When multiple 'Caution' level problems occur simultaneously (R-03), it is reasonable to classify this as a higher danger level than a single 'Caution' problem (R-02), as even if each problem is minor, their combined interaction can cause unpredictable data distortion (e.g., resonance). This rule design is based on experimental results from the analysis of past misreading data, which confirmed that complex causes correlate with a higher misreading rate than single causes.

[0083] The judgment rules of the present invention guarantee reproducibility because they are based on objective and measurable physical quantities, such as sensor data and system logs, rather than relying on the subjective judgment or senses of the operator. Furthermore, in order to maintain consistent inspection quality despite the unpredictable instability of the actual production environment, the introduction of a dynamic inspection method that adapts to the environment is inevitable, and the present invention provides a specific solution for this. By adopting these judgment criteria and rules, the present invention fundamentally avoids data distortion that may occur due to physical, optical, and temporal instability, thereby reducing the misreading rate; improves inspection reliability by basing it on data that guarantees all inspections were performed under optimal conditions; and increases the efficiency and yield of the entire production line by preventing unnecessary re-inspections and the leakage of defective products.

[0084] The definitions of the terms used above are as previously described. The 'caution threshold' and 'risk threshold' used in the judgment rule table are not fixed constants, but are set during the calibration process performed during the initial installation of the equipment or when changing the battery model. The calibration process involves performing more than 100 repeated measurements using standard specimens, analyzing the statistical distribution of each judgment criterion data collected at that time, and automatically setting them according to objective criteria. Furthermore, the present invention includes a procedure for dynamically updating thresholds to adapt to changes in the system state due to long-term operation. When a preset cycle arrives, the moving average of each judgment criterion data collected from tests judged as 'stable' during the previous cycle is calculated. If this value continues to deviate from the preset rate of change relative to the average value at the time of initial setup, it is determined that the basic state of the system has changed. When the above condition is satisfied, the system performs an automatic readjustment process along with a notification to the operator, reanalyzes the statistical distribution based on the 100 most recent 'stable' state data points, and transmits a command to automatically update and apply the 'caution threshold' and 'risk threshold' suitable for the new environment.

[0085] The 'caution threshold' and 'danger threshold' used in this invention are established through simulations and analyses based on actual data as follows, thereby eliminating arbitrariness and ensuring objectivity. First, to secure reference data for a normal state during the initial calibration phase of the equipment, 500 inspection simulations using standard specimens are performed to collect Micro-Slip Probability (MSP) data. As a result of statistical analysis, the mean of the data was calculated to be 0.20 and the standard deviation 0.10; based on this, the 'caution threshold' is set to 0.50, which is the 'mean + 3 x standard deviation' point capable of detecting statistically significant deviations. Additionally, MSP values ​​that cause severe distortion in image quality are analyzed through past operational data and experiments involving the intentional application of vibration. The table below shows an example of 10 simulation data sets used to identify the MSP value at which the image distortion rate begins to exceed a specific threshold (e.g., 5%).

[0086] [Table 3]

[0087]

[0088] As shown in the simulation results above, a tendency was consistently observed where the image distortion rate increased sharply around an MSP value of 0.80 and exceeded the allowable threshold (5%). Based on this practical data analysis, the 'risk threshold' to prevent critical inspection errors is set to 0.80.

[0089] To verify the validity of the set threshold, a simulation was performed to compare the results of applying the dynamic frame acquisition logic of the present invention to 10 virtual scenarios with various levels of instability with the prior art of a fixed 1-frame method. The 'expected misinterpretation rate' is the potential error probability that may occur when capturing a fixed 1-frame under the level of instability of each scenario, and the 'actual misinterpretation rate' is the final error probability after applying the present invention.

[0090] [Table 4]

[0091]

[0092] As clearly shown in the verification simulation table above, the logic of the present invention maintains the same efficiency as the prior art without unnecessary additional shooting when the MSP value is below the 'stable' grade threshold. However, in scenarios where the MSP value exceeds the 'caution' or 'danger' threshold (S1-02, S1-04, S1-05, S1-07, S1-08, S1-09), it demonstrates a clear technical effect of significantly reducing the expected misreading rate by dynamically increasing the number of frame acquisitions and applying an appropriate post-processing method. This proves that the threshold set based on data directly contributes to the improvement of actual inspection quality.

[0093] The control unit (400) of the present invention includes three judgment criteria for evaluating the overall uncertainty of the system in order to reflect in the final judgment the 'measurement reliability' that cannot be obtained solely from the analysis results of the image data itself. The first judgment criterion includes the micro-slip probability, which is an indicator representing physical stability at the moment of inspection as defined in Process 1 above. In the context of calculating a defect score, the micro-slip probability is intended to evaluate the possibility that the shape or size of the measured defect may be distorted by micro-vibration, and a high value indicates that the reliability of the measured defect data is low. The second judgment criterion includes the accumulated correction deviation, which indicates how much the correction values, such as light intensity, focus, and sensitivity, performed by the automatic correction module (420) during the long-term operation of the equipment have changed cumulatively compared to the initial state. The accumulated correction deviation is intended to evaluate the potential error level that the gradual performance degradation of the system, such as sensor aging, lens contamination, or a decrease in light intensity, affects the current measurement value, and a high value indicates that the measurement reliability of the system itself has decreased. The third judgment criterion includes reflective anisotropic dispersion, which indicates the degree to which the brightness or contrast ratio of the same defect is measured inconsistently depending on the lighting angle or shooting position due to fine curvature or hairline finish on the battery surface. The reflective anisotropic dispersion is intended to evaluate the possibility that the severity of the measured defect may be overestimated or underestimated due to the optical characteristics of the battery surface, and a high value indicates that the reliability of the defect data under specific measurement conditions is low.

[0094] The present invention includes a procedure for adjusting the 'final defect score' by linking the 'system uncertainty grade' determined by synthesizing the three judgment criteria to the 'initial defect score' calculated through image analysis. The technical basis for this is that even if scratches of the same size are observed, the result captured by an aged sensor under extreme vibration is less reliable than the result captured by the best sensor in a stable environment, so a more conservative judgment must be made considering potential risks. To this end, the control unit (400) determines the system uncertainty grade in three stages: 'trustworthy', 'average', and 'doubtful', and performs a procedure for applying a predefined differential 'defect score correction rule' according to each grade.

[0095] The above judgment criteria are physically measured and quantified through specific methods. The micro-slip probability is calculated in the same manner as described in Process 1 above. The calibration cumulative deviation is calculated based on the internal logs recorded by the automatic calibration module (420) as it fine-tunes the illumination intensity, camera exposure time, and digital gain value to obtain the optimal image for each inspection. The calibration cumulative deviation is calculated by normalizing the standard deviations of each calibration value recorded during the most recent 1,000 inspections and summing them by applying automatically set weights. The weights are automatically set during the equipment calibration phase, in which a reference image is first taken using a standard defect specimen, and an image is taken with only the illumination intensity correction value intentionally changed by +10% relative to the reference value while keeping other conditions fixed, and the sensitivity, which is the rate of change of the measured defect score, is calculated. In the same way, the rate of change of the defect score is measured for the exposure time and digital gain value, respectively, and the three measured sensitivity values ​​are normalized so that the highest weight is automatically assigned to the factor that has the greatest influence on the defect score. Furthermore, the present invention goes beyond simply setting weights based on sensitivity alone and enables the dynamic optimization of weights by utilizing long-term operational data of the equipment. Specifically, through the analysis of historical data, the 'variability (standard deviation)' during actual operation of each correction element (lighting, exposure, gain) is additionally calculated, and the final weight is recalculated to be proportional to 'sensitivity × variability'. This is a rational method that assigns the highest weight to the element that has a significant impact on the defect score (high sensitivity) and fluctuates frequently in practice (high variability), thereby serving as the primary cause of uncertainty. For example, assuming that through the above procedure, the weights for digital gain (which is most sensitive to the defect score), lighting (0.3), and exposure time (0.2) are automatically set, and the normalized standard deviations of each correction amount during real-time operation are 0.7, 0.4, and 0.2, respectively, the cumulative correction deviation is 0.It is calculated as 51. Reflective anisotropic variance is obtained by photographing the same standard defect area from five different angles during the initial setup phase for each battery model. The coefficient of variation of the contrast values ​​of the defects measured in each image is calculated and stored as the 'basic reflective anisotropic variance' value for the corresponding battery model. Furthermore, whenever a production lot changes or at preset cycles, the above measurement procedure is simplified and performed for batteries sampled from the production line. If the measured 'real-time reflective anisotropic variance' value deviates from the preset tolerance relative to the stored 'basic reflective anisotropic variance' value, the control unit (400) determines that the surface characteristics of the lot have changed and transmits a command to update the measured 'real-time reflective anisotropic variance' value to a new reference value to be applied to the inspection of the lot.

[0096] Based on the 'system uncertainty grade' determined through the above logical processing procedure, the initial defect score is adjusted as follows and used for the final judgment. If the grade is determined to be 'trustworthy,' since both the measurement environment and the system state are stable, a command is transmitted to set the final defect score to be the same as the initial defect score by trusting the image analysis results as they are. If the grade is determined to be 'average,' since there is some uncertainty, a command is transmitted to increase the final defect score by multiplying the initial defect score by 1.2, which is the 'first correction factor' set in advance, in order to prepare for potential risks by slightly strengthening the judgment criteria. If the grade is determined to be 'suspicious,' since the reliability of the measurement results is very low and the risk is high to make a judgment as is, a command is transmitted to significantly increase the final defect score by multiplying the initial defect score by 1.5, which is the 'second correction factor' larger than the 'first correction factor,' or to automatically classify it as 're-inspection required' regardless of the defect score.

[0097] For example, in a scenario where a battery with severe surface reflection is inspected in equipment where sensor aging has progressed due to long-term operation, the analysis judgment module (440) first analyzes the captured image to detect fine scratches and calculates an initial defect score of 60 points. Subsequently, the control unit (400) evaluates the system state at the time of inspection and calculates the micro-slip probability as 0.2, the calibration cumulative deviation as 0.7, and the reflection anisotropic variance as 0.08. The control unit (400) refers to an internal judgment rule table and determines the overall 'system uncertainty grade' as 'suspicious' because the calibration cumulative deviation and the reflection anisotropic variance values ​​each exceed the 'normal threshold'. Based on the 'suspicious' grade, a 'second correction factor' of 1.5 is applied to the initial defect score to recalculate the final defect score to 90 points. Finally, since the recalculated final defect score of 90 points exceeds the defect judgment threshold of 70 points, a command is transmitted to finally determine the battery as 'defective'.

[0098] If the final defect score with the correction factor applied becomes abnormally high, it may indicate a serious system error beyond mere uncertainty; therefore, a system inspection warning is triggered simultaneously with the defect determination to prompt verification by an operator. Additionally, if an item for which a 're-inspection' was requested due to a 'Suspicious' rating receives a 'Suspicious' rating consecutively during the re-inspection, automatic re-inspection is no longer performed. Instead, it is classified as 'requiring manual verification' to prevent infinite loops and request operator intervention.

[0099] When comparing the operational results of the prior art and the present invention, both technologies determine the product to be good in stable environments with boundary defects. However, in environments with boundary defects and aging sensors, the prior art determines the product to be good, creating a risk of leaking defects that could actually be serious flaws. In contrast, the present invention successfully prevents the leakage of potential defects by determining the product as 'suspicious,' adjusting the final score upward, and then determining it as defective. Furthermore, in environments without defects but with extreme vibration, the prior art determines the product to be good, whereas the present invention determines it as 'suspicious,' adjusts the final score upward to determine it as defective, or requests a re-inspection, thereby recognizing the possibility of false defects caused by vibration and preventing mis-diagnosis.

[0100] [Table 5]

[0101]

[0102] In the above rule table, if conditions R-13 and R-14 are satisfied, they take precedence over R-12.

[0103] The control unit (400) operates according to an internal judgment rule table. As a first rule, if all conditions are satisfied—that the micro-slip probability is less than 0.3, the correction cumulative deviation is less than 0.4, and the reflection anisotropic variance is less than 0.05—it is determined to be a 'trusted' grade and a command is sent to set the final score to be the same as the initial score. As a second rule, if any one of the conditions is satisfied—that the micro-slip probability is 0.3 or higher, the correction cumulative deviation is 0.4 or higher, or the reflection anisotropic variance is 0.05 or higher—it is determined to be a 'normal' grade and a command is sent to multiply the final score by 1.2. As a third rule, if multiple conditions corresponding to the 'normal' grade condition are satisfied simultaneously, it is determined to be a 'doubtful' grade and a command is sent to multiply the final score by 1.5. As a fourth rule, if any of the conditions are satisfied—that is, the micro-slip probability is 0.7 or higher, the correction cumulative deviation is 0.8 or higher, or the reflection anisotropic variance is 0.1 or higher—it is classified as 'suspected' and 're-examination required,' and the third and fourth rules are applied in priority over the second rule.

[0104] The present invention eliminates subjectivity that relies on a single information source, such as image analysis results, and ensures the transparency and reproducibility of the judgment process based on objective data, such as sensor data, system logs, and physical measurements. Furthermore, since a perfect inspection environment does not exist in reality, measuring and managing such 'uncertainty' is an inevitable requirement for high-precision inspection systems, and the present invention provides a specific solution for this. By adopting these judgment criteria and rules, the present invention effectively blocks the leakage of defects by preventing defects from being underestimated due to system deterioration, reduces over-detection by inducing re-inspection for false defects caused by an unstable measurement environment, and analyzes the adjustment pattern of the final defect score over the long term to trace back where reliability degradation mainly occurs in the system and utilizes this as predictive maintenance data.

[0105] The definitions of the terms used above are as previously described, and the threshold values ​​used in the judgment rule table and the 'first and second correction coefficients' are set during the calibration process. The calibration process collects data by intentionally reproducing various levels of uncertainty using multiple good and defective master specimens, and automatically derives objective threshold values ​​and correction coefficients by using statistical techniques to search for a point where the probability of defect leakage and the probability of over-detection are simultaneously minimized based on the collected data. The specific method for 'reproducing uncertainty' is as follows. To reproduce ACD, the correction limit range of the automatic correction function is intentionally reduced via software, or data is collected from a test bed equipped with aged lighting and sensors. To reproduce RAV, data is collected by preparing multiple battery specimens with slightly different surface finishes (hairline direction, glossiness, etc.). Through this systematic experimental design, an extensive dataset for various uncertainty scenarios that may occur in reality is constructed, and the optimal threshold values ​​and correction coefficients are set based on this. In addition, as described in Process 1 above, a command to dynamically update the values ​​is automatically transmitted through a periodic automatic readjustment process.

[0106] The threshold value and 'defect score correction coefficient' for determining the 'system uncertainty grade' used in the present invention are optimized through the following data-based simulation. First, using a 'borderline defect' specimen (a sample with actual defects but an initial defect score close to the acceptance criterion), inspections are repeatedly performed under various levels of Accumulated Calibration Deviation (ACD) values, and the actual defect leakage rate is measured. The table below is an example of 10 simulation data runs.

[0107] [Table 6]

[0108]

[0109] As a result of the above data analysis, it was observed that the defect leakage rate exceeded the management standard (2%) starting from the point where the ACD value exceeded 0.40. Therefore, the ACD threshold for the 'Normal' grade, where the risk of potential defect leakage must begin to be managed, is set to 0.40. The thresholds for other indicators and the threshold for the 'Suspicious' grade are also set through data-based analysis in the same manner.

[0110] To verify the validity of the set threshold and correction coefficient, the results of applying the adjustment defect score calculation logic of the present invention to 10 different scenarios were compared with the prior art. The 'initial judgment' is the result without considering uncertainty, and the 'final judgment' is the result of applying the present invention.

[0111] [Table 7]

[0112]

[0113] Through the verification simulation above, it can be confirmed that the logic of the present invention does not perform unnecessary score adjustments when system uncertainty is low ('trust' grade). However, in borderline defect scenarios (S2-02, S2-04, S2-07, S2-09) where system uncertainty is determined to be 'moderate' or 'suspicious' grade, it demonstrates a clear technical effect of effectively preventing the leakage of critical defects by ultimately determining potential defects as defective, even though they could have been judged as good based on the initial score. This proves that data-based thresholds and correction coefficients contribute to actual quality control.

[0114] The control unit (400) of the present invention determines the comprehensive maintenance requirement through three judgment criteria that evaluate the actual condition of the equipment from various angles, rather than simple operating time, in order to prevent sudden failure of the equipment and continuously maintain optimal inspection performance. The first judgment criterion includes the mounting degree of freedom instability coefficient, which is an indicator that quantifies mechanical play or non-linearity of positioning caused by wear or loosening of the driving and fixing parts of the battery mounting unit (300) due to repetitive use. The mounting degree of freedom instability coefficient is intended to detect physical performance degradation of the mechanism that impedes the positional reproducibility of the inspection target, and a high value indicates that the mechanical precision of the equipment has deteriorated. The second judgment criterion includes the calibration cumulative deviation, which indicates the gradual aging state of optical and electronic components such as sensors, lenses, and lighting, as defined in Process 2 above. In the context of determining the maintenance requirement, the calibration cumulative deviation is intended to evaluate whether the fundamental performance limit of the hardware has been reached, which is no longer compensated for by the automatic calibration function alone, and a high value suggests that replacement or precise readjustment of the parts is required. The third judgment criterion includes the Maintenance Debt Accumulation Index, an indicator that quantifies the extent to which planned regular maintenance activities have accumulated without being performed within the recommended cycle. The purpose of the aforementioned Maintenance Debt Accumulation Index is to detect procedural and operational issues that are accumulating potential failure risks, even if they do not appear to be problems at present; a high index indicates that the reliability of the equipment is gradually decreasing due to inadequate preventive maintenance.

[0115] In order to overcome the inefficiency of simple time-based maintenance and the fatal disadvantages of repair after failure, the present invention determines the actual health of the equipment by integrating the three heterogeneous data, namely mechanical, electronic, and operational data. The technical basis for this is that since equipment failure occurs due to the accumulation of complex factors rather than a single cause, failure can only be effectively predicted and prevented by comprehensively monitoring condition indicators from various aspects. To this end, the control unit (400) determines the maintenance requirement level in three stages: 'normal', 'concern', and 'warning', and performs a procedure to provide differential levels of notification and specific action guides to the operator according to each level.

[0116] The above judgment criteria are physically measured and quantified through specific methods. The settling degree of freedom instability coefficient is calculated by periodically commanding the control unit (400) to move a specific path 10 times repeatedly, and recording the error between the actual movement position reported by the motor encoder and the commanded position at every moment. The settling degree of freedom instability coefficient is calculated as the standard deviation of the recorded position error data; for example, if the standard deviation of the position error is 3 micrometers, it is calculated as 3.0. The calibration cumulative deviation is calculated in the same manner as described in Process 2 above. The maintenance debt accumulation index is calculated based on each maintenance item, recommended cycle, and 'importance grade' set according to objective criteria, which are pre-registered in the system's internal maintenance management module. The 'importance grade' is initially set according to the recommended maintenance guidelines provided by the equipment manufacturer, and furthermore, can be objectively re-verified and adjusted through the FMEA technique, which evaluates the severity of the potential impact of non-compliance with each maintenance item on equipment sudden shutdown, deterioration of inspection quality, and the occurrence of safety issues. Specifically, for each maintenance item, a risk priority number (RPN = S × O × D) is calculated based on the Severity, Occurrence, and Detection scores (1!10 points each) derived through FMEA analysis. A rule is applied to normalize the calculated RPN value and divide it into intervals (e.g., RPN 1–100 → 1 point, 101–200 → 2 points, etc.) to convert it into a final 'importance grade' of 15 points. The above maintenance debt accumulation index is calculated by summing the importance grades of all non-fulfilled items that have exceeded the recommended cycle at the current point in time, multiplying the excess time or frequency by the importance grade. For example, if two items with importance levels of 3 and 5, respectively, set through FMEA analysis, are delayed by 1,000 hours and 2,000 hours, respectively, the maintenance debt accumulation index is calculated as 13,000.

[0117] The use of 'standard deviation' in this invention to evaluate dynamic variability, such as Camera-Actuator Delay Jitter (CALJ) and Mounting Degrees of Freedom Instability Factor (MFI), is based on the following technical necessity. Instability does not simply mean that the average value of the error is large, but is defined as the degree to which the measured values ​​are 'irregularly scattered' around the mean. Standard deviation is a representative and reliable statistical indicator for quantifying the dispersion of such data.

[0118] The following 10 examples comparing the 'maximum-minimum error range' method that can be used in conventional technology with the 'standard deviation' method of the present invention clearly demonstrate its technical superiority.

[0119]

[0120] In the results of the above examples, the 'max-minimum range' method shows a very large value (20) in the 'one-time noise' scenario (ID 4, 8) which includes a single sudden large error, posing a risk of misjudging a system that may actually be stable as excessively unstable. On the other hand, in the more dangerous scenario (ID 5, 9) where the error range is narrow but continuous irregular vibrations occur, it shows a relatively small value (4), which may lead to an underestimation of the severity of the problem. In contrast, the 'standard deviation' method comprehensively reflects the distribution of all data, so it reacts relatively less sensitively to one-time noise (ID 4, 8), while accurately detecting continuous instability (ID 5, 9) as a significant value that is clearly distinguishable from the stable state (ID 1, 7). As a result, the present invention has the technical effect of evaluating the dynamic stability of a system more accurately and reliably compared to conventional range-based judgment methods.

[0121] Based on the 'Maintenance Requirement Level' determined through the aforementioned logical processing procedure, the following specific information and recommendations are delivered to the operator. If determined to be 'Normal,' it signifies that the equipment is in optimal condition, so the equipment is operated normally without separate notification. If determined to be 'Concern,' a "Maintenance Recommended" message is displayed on the equipment's user interface screen along with a yellow icon, and a command is sent to guide recommended actions related to status indicators approaching a threshold. If determined to be 'Warning,' a red warning window pops up on the screen and an audible alert is generated; along with a "Immediate Inspection Required" message, the core cause of exceeding the threshold and essential corrective actions are specified, and an equipment status anomaly code is automatically transmitted to the higher-level production management system.

[0122] For example, in a scenario where regular lens cleaning is omitted and wear of the X-axis drive unit progresses slowly, the control unit (400) first calculates each status indicator. When the calculation results show that the settling degree of freedom instability coefficient is 4.5 micrometers, the calibration cumulative deviation is 0.78, and the maintenance debt cumulative index is 15,000, the control unit (400) refers to an internal judgment rule table. Since the values ​​of the calibration cumulative deviation and the maintenance debt cumulative index each exceed the 'interest threshold,' both conditions are satisfied simultaneously, and the overall 'maintenance requirement grade' is determined as 'Warning.' The control unit (400) immediately displays a red warning window on the screen and provides detailed information such as "Cause 1: Approaching calibration cumulative deviation threshold" and "Cause 2: Maintenance debt accumulation," along with a message stating "Immediate inspection required: Suspected complex performance degradation," and a specific guide stating "Essential action: Perform optical system precision correction and lens cleaning," while simultaneously automatically transmitting a status code to the upper system.

[0123] The change trends of all status indicators are stored as time-series data in an internal database, and the operator analyzes the cause of performance degradation by viewing the past trend graph of each indicator. When the operator performs maintenance and records completion in the system, the control unit (400) automatically remeasures the corresponding status indicator immediately, and if the remeasured value does not return to the normal range, it generates a "maintenance incomplete" warning to induce re-verification.

[0124] When comparing the operational results of the prior art and the present invention, the prior art results in sudden failure when premature wear occurs, whereas the present invention prevents such failure by detecting an excess of the seating degree of freedom instability coefficient value and issuing a warning in advance. When component life is good, the prior art results in wasted costs due to unnecessary replacement, whereas the present invention reduces operating costs by extending the replacement cycle since all condition indicators are normal. When minor maintenance is omitted, the prior art shows a gradual increase in the rate of unexplained misreading, whereas the present invention detects an excess of the maintenance debt accumulation index value and guides the maintenance of optimal inspection quality through simple measures.

[0125] [Table 8]

[0126]

[0127] In the above rule table, if conditions R-23 and R-24 are satisfied, they take precedence over R-22.

[0128] The control unit (400) operates according to an internal judgment rule table. As a first rule, if all conditions are satisfied—that the seating degree of freedom instability coefficient is less than 3.5 micrometers, the correction cumulative deviation is less than 0.6, and the maintenance debt cumulative index is less than 8,000—it is determined to be a 'normal' grade and no separate action is taken. As a second rule, if any of the above indicators exceed the 'concern threshold,' it is determined to be a 'concern' grade and a command is sent to display a yellow icon and recommended action. As a third rule, if multiple conditions corresponding to the 'concern' grade condition are satisfied simultaneously, it is determined to be a 'warning' grade and a command is sent to display a red warning window and specify mandatory action. As a fourth rule, if any of the above indicators exceed the 'warning threshold,' it is determined to be a 'warning' grade and a command is sent to display a red warning window and automatically report to the upper system, and the third and fourth rules are applied in priority over the second rule.

[0129] The present invention excludes subjective and empirical judgments and provides objective grounds for maintenance requirements based on quantitative and traceable data. Furthermore, organically combining the three elements of mechanical, electronic, and operational management for monitoring is an inevitable technical solution for failure prevention and performance maintenance. By adopting these components, the present invention maximizes equipment uptime by preventing unexpected failures in advance, optimizes maintenance costs by preventing both over-maintenance and under-maintenance, and maintains consistency in inspection quality by detecting and addressing equipment performance degradation early.

[0130] The definitions of the terms used above are as previously described, and the 'interest threshold' and 'warning threshold' used in the 'judgment rule table' are set based on the equipment manufacturer's recommended specifications, long-term operational data statistics, or the analysis of failure history data of similar equipment. Furthermore, the present invention may further include a dynamic learning mechanism that self-optimizes thresholds based on actual operational data. Whenever an actual failure occurs during equipment operation and is recorded in the maintenance history, the system stores the time-series pattern of the status indicators measured immediately before the failure in the database as a 'failure sign pattern'. The 'failure sign pattern' is defined not as a simple numerical value immediately before the failure, but as multidimensional vector data including the average value, standard deviation, and the trend of change (slope) of each status indicator calculated through linear regression during the relevant period. By learning this multidimensional pattern, the system can detect not only simple threshold exceedances but also potential risks that show a trend of increasing at an abnormal rate even when the value is within the normal range. At preset intervals, the system analyzes stored 'failure symptom pattern' data and 'normal operation pattern' data using machine learning algorithms, and based on the learning results, sends a command to automatically update the 'interest threshold' and 'warning threshold' by resetting the optimal threshold boundaries that can most clearly distinguish between failure and normal conditions.

[0131] The threshold value for determining the 'maintenance requirement grade' used in this invention is set through correlation analysis with actual failure data. One year of operational data from 100 identical pieces of equipment is analyzed to aggregate the number of actual drive unit failures within a specific range of Mounting Degrees of Freedom Instability (MFI) values. The table below shows an example of data for analyzing the relationship between MFI values ​​and failure rates.

[0132] [Table 9]

[0133]

[0134] Based on the analysis of the data above, an inflection point was observed where the failure rate increased sharply to over 40% starting from the range where the MFI value exceeded 4.5 μm. Therefore, to prevent production stoppages caused by sudden failures, the 'warning threshold,' which is the point where the risk of failure is very high and immediate action is required, is set to 5.0 μm. The 'concern threshold' can be set to 3.5 μm, where the failure rate begins to increase significantly.

[0135] To verify the validity of the set threshold, the results of applying the maintenance judgment logic of the present invention to 10 virtual equipment status scenarios were compared with the prior art of a fixed time-based maintenance method (TBM, e.g., maintenance every 2000 hours).

[0136] [Table 10]

[0137]

[0138] The verification simulation described above demonstrates the clear technical effects of the present invention. Prior art time-based maintenance performs maintenance regardless of the actual condition of the equipment, resulting in wasted costs due to over-maintenance (S3-01, S3-06) or causing sudden failures by failing to respond to unexpected problems such as premature wear (S3-02, S3-05, S3-09). In contrast, the present invention demonstrates excellent effects in reducing unnecessary maintenance and preventing sudden failures in advance by accurately determining 'Normal,' 'Concern,' and 'Warning' grades based on the actual condition data of the equipment.

[0139] The control unit (400) of the present invention includes judgment criteria for evaluating the state of the lighting environment and setting restoration goals in order to prevent a rapid deterioration in inspection quality and minimize production interruption when an unexpected situation, such as a failure of some units, occurs in a lighting system composed of a plurality of independent lighting units (120). The first judgment criterion includes the lighting unit (120) operation state, which is binary state information indicating whether each individual lighting unit (120) is normally lit and stably outputting light intensity according to control commands. The lighting unit (120) operation state is intended to identify the system's currently available resources in real time, thereby clearly distinguishing between normal lighting units (120) to be controlled and faulty lighting units (120) to be excluded from control. The second judgment criterion includes the lighting uniformity deviation, which is an indicator that quantifies the difference between the brightness distribution of the actual lighting currently being irradiated in the inspection area and a predefined 'ideal lighting distribution map'. The aforementioned illumination uniformity deviation is intended to evaluate the achievement of the restoration goal for the lighting environment; minimizing this deviation value is the ultimate goal of lighting reconstruction and serves as a key indicator for evaluating performance during the iterative control process.

[0140] The present invention includes an automated recovery procedure that goes beyond a passive response of merely generating a warning when some lighting units (120) fail, and actively readjusts the output and angles of the remaining normal lighting units (120) to restore the original ideal lighting environment as closely as possible. The technical basis for this lies in the optical principle that in a lighting system using multiple light sources, the parts of each light source contributing to the overall lighting distribution overlap; therefore, even if some light sources are lost, overall uniformity can be restored to a certain level by appropriately adjusting the contribution of the remaining light sources. To this end, the control unit (400) switches to a 'lighting reconstruction mode' immediately upon detecting a failure and executes a procedure that performs an 'iterative feedback control loop' to progressively search for and optimize the setting values ​​of the normal lighting units (120) with the goal of an 'ideal lighting distribution map'.

[0141] The above judgment criteria are physically measured and quantified through specific methods. The operating status of the lighting unit (120) is obtained by a self-diagnostic circuit built into each lighting unit (120) monitoring the output current and voltage in real time. The control unit (400) requests a status report from each lighting unit (120) at every inspection cycle, and if there is no response or the response value is outside the normal range, the status of the unit is determined to be 'faulty'. For example, if there is no response from unit 3 out of a total of 8 lighting units (120), the system generates a status vector indicating a faulty state. The lighting uniformity deviation is obtained by the control unit (400) placing a reference white plate in the inspection area, taking an image to obtain a 'real-time lighting distribution map', and calculating the difference between this map and an 'ideal lighting distribution map' stored in memory. For example, for a 100x100 pixel area, the root mean square error is calculated by adding up the squares of the differences in brightness values ​​for each pixel, dividing by the number of pixels, and taking the square root. If the root mean square error in an ideal state is less than 5.0 but the value measured immediately after a failure is 45.8, it is quantitatively determined that the lighting uniformity is severely compromised.

[0142] The use of the Root Mean Square Error (RMSE) to calculate the illumination uniformity deviation in the present invention is based on the following technical necessity.

[0143] The process of 'squaring' errors has the effect of imposing greater penalties on specific areas with large deviations (e.g., severe shading or excessive highlights). Compared to the Mean Absolute Error (MAE) method, which simply averages the absolute values ​​of errors, this allows for more sensitive detection of localized lighting defects, thereby enabling stricter management of overall lighting quality.

[0144] Taking the 'square root' of the total error makes the unit of the final deviation value the same as the original data (pixel brightness value), making it easier to intuitively interpret the physical meaning of the value and compare it with the threshold value.

[0145] The technical superiority of the above RMSE method is clearly demonstrated through the following 10 comparative examples. These examples assume various lighting failure conditions and are the results of calculating both MAE and RMSE for each condition. In the 'condition description,' 'local' refers to cases where severe deviations are concentrated only in specific areas, while 'global' refers to cases where low levels of deviation are spread throughout the image.

[0146]

[0147] As is clearly shown in the results of the above examples, when noise is spread throughout (ID 1, 3, 4, 6, 8, 9), the MAE and RMSE values ​​tend to increase proportionally. However, when localized and severe lighting defects (ID 2, 5, 7, 10) that have a critical impact on inspection quality occur, the MAE value increases relatively gradually, whereas the RMSE value is calculated to be much larger because it amplifies the deviation by squaring it. This demonstrates that the RMSE method has a technical effect of more accurately assessing the 'risk' of the lighting environment and enabling early response than the conventional simple average method.

[0148] When a new set of optimized lighting setting values ​​is derived as a result of performing the above logical processing procedure, the control unit (400) immediately transmits and applies the derived optimal light intensity and angle setting values ​​to each normally operating lighting unit (120). In addition, to compensate for the minute difference between the reconfigured lighting environment and the ideal state, the control unit (400) transmits a command to slightly correct inspection parameters, such as the camera's exposure time or gain value, accordingly. When the lighting reconfiguration is completed, a notification containing detailed information, including the details of the failure and the fact that automatic reconfiguration is complete, is displayed to the operator, and a command is transmitted to record all processes in a log.

[0149] For example, in a scenario where unit 3 of a total of 8 LED lighting units (120) suddenly fails, first, during the state check before the next inspection, the control unit (400) detects the failure of lighting unit 3 (120), immediately pauses the inspection process, and automatically switches to 'lighting reconfiguration mode'. The control unit (400) photographs a reference plate to confirm that the root mean square error, which is the current lighting uniformity deviation, is 45.8, and loads an 'ideal lighting distribution map' from memory. Subsequently, the control unit (400) performs a greedy search-based optimization algorithm. In the search phase, first, lighting units 2 and 4 (120), which are closest to the failed unit 3, are selected as primary compensation units, and the change in root mean square error is measured by changing the light intensity of each unit by a predetermined unit. This is an initial search strategy to increase computational efficiency. If the improvement in the RMSE value is minimal with only adjacent units, the algorithm performs a multi-stage search procedure to find the optimal compensation unit by gradually expanding the search range to include secondary adjacent units or all normal units. Even if a faulty unit is located at the edge and there is only one adjacent unit, the search range is automatically expanded to find the optimal solution. In the decision and adjustment stage, if the root mean square error value decreases most significantly when the light intensity of unit 4 is increased, unit 4 is determined as the optimal adjustment target for this iteration, and the corresponding setting value is actually applied. In the next iteration, the same search and adjustment stage is performed again for all normal lighting units (120) based on the current state. The above iteration process is performed until the root mean square error value no longer decreases or the amount of decrease is below a preset negligible level. After dozens of iterations, optimization is terminated at the point where the root mean square error value decreases to 8.5, and the finally derived setting value is applied to each lighting unit (120).Finally, after increasing the camera exposure time by 5% to compensate for the reduced total light intensity, a notification is displayed to the operator and a command is sent to automatically resume the inspection process.

[0150] If the illumination uniformity deviation fails to converge below an acceptable threshold within the preset maximum number of iterations, it implies that it is difficult to secure even the minimum inspection quality with the remaining lighting alone; therefore, the system abandons automatic recovery and transmits a command to safely stop equipment operation, accompanied by a warning stating, "Illumination system fatal error. Manual inspection required."

[0151] When comparing the operation results of the prior art and the present invention, if a single lighting unit (120) fails, the prior art immediately stops the operation of the equipment, causing the entire production line to stop. In contrast, the present invention automatically completes the reconfiguration of the lighting environment within seconds, displays a warning, and continues inspection without interrupting production, thereby minimizing the loss of operating rate. Furthermore, if shading occurs due to a failure, the prior art causes a fatal defect leakage in which all defects in the affected area are not detected, whereas the present invention significantly reduces the possibility of fatal defect leakage by repositioning the remaining lights to compensate for the shading area, thereby greatly mitigating the shading and ensuring minimum inspection quality.

[0152] [Table 11]

[0153]

[0154] The logic flow of the control unit (400) is as follows. In the first step, if the status report value of the lighting unit (120) is not normal, a fault is detected, the inspection is paused, and the system switches to 'lighting reconfiguration mode'. In the second step, if the current root mean square error exceeds the target value, it is determined that optimization is in progress, the setting value of the normal lighting unit (120) is adjusted, and the second step is performed again. In the third step, if the current root mean square error converges to below the target value or the number of iterations exceeds the maximum value, it is determined that optimization is complete or reconfiguration has failed, and respective measures are performed.

[0155] The reconfiguration process of the present invention completely eliminates manual adjustment methods that rely on the operator's senses or experience, and guarantees perfect reproducibility by being automatically performed according to a predefined algorithm toward the clear goal of minimizing the quantitative and objective indicator known as 'illumination uniformity deviation' measured in real time. Furthermore, since component failures are an unavoidable phenomenon in automated production lines operating 24 hours a day and stopping the production line every time a failure occurs causes massive losses, it is an inevitable requirement for advanced automation systems to possess 'resilience'—that is, the ability to recognize problems on their own and continue operations by making maximum use of available resources—and the present invention is a specific technical solution to this requirement.

[0156] By adopting these components, the present invention significantly reduces line downtime caused by minor component failures, thereby maximizing equipment utilization and overall productivity; ensures quality stability by fundamentally preventing a large number of undetected defects that may occur due to uneven lighting immediately after a failure; and increases the efficiency of maintenance activities by allowing an operator to plan repairs in time for the next downtime while the equipment is operating in standby mode.

[0157] The definitions of the terms used above are as previously described, and the 'target root mean square error' and 'tolerance' used in the logic are set during the initial calibration stage of the equipment. After setting the optimal lighting environment while all lighting units (120) are in a normal state, the lighting distribution of the reference image captured is saved as an 'ideal lighting distribution map,' and the root mean square error value at this time is set as the 'target root mean square error.' Furthermore, the present invention may include a procedure for dynamically updating the 'ideal lighting distribution map' itself, which is the target of optimization, in order to adapt to the gradual aging of the equipment or changes in the characteristics of the inspection target. When there is a preset cycle or a command from an operator, the system first confirms that all lighting units (120) are in a normal state and performs a global optimization algorithm that searches again from the beginning for the lighting environment most suitable for the current optical system state and the surface characteristics of the inspection target. As an example of the above 'global optimization algorithm,' a simulated annealing technique may be used. This technique prevents getting stuck in local optima by accepting solutions that are slightly worse than the current optimal solution (settings where RMSE increases slightly) with a certain probability, and can derive a new 'ideal lighting distribution map' that approximates the global optimal solution by efficiently exploring a wide search space. A command is then transmitted to update the existing 'ideal lighting distribution map' with the new optimal lighting distribution derived through this process, using it as a new standard.

[0158] In this invention, the 'reconstruction failure' threshold for determining the success of lighting reconstruction is set based on actual data analyzing the relationship between the illumination uniformity deviation (RMSE) and the actual defect detection rate. The RMSE value is varied while intentionally making the lighting environment non-uniform, and the defect detection rate of a standard defective specimen is repeatedly measured at each RMSE value. The table below shows an example of 10 simulation data trials.

[0159] [Table 12]

[0160]

[0161] As a result of analyzing the simulation data above, a critical phenomenon was clearly observed where the defect detection rate drops sharply to below 99.9%, which is the production quality control standard, starting from the point where the RMSE value exceeds 15.0. Therefore, as a bottom line to guarantee minimum inspection quality and prevent defect leakage, the RMSE threshold for determining 'reconfiguration failure' is set to 15.0.

[0162] To verify the validity of the set threshold, the results of applying the automatic reconfiguration logic of the present invention to 10 different lighting failure scenarios were compared with a prior art that continues inspection without recognizing the failure.

[0163] [Table 13]

[0164]

[0165] The verification simulation table above clearly demonstrates the unique effects of the present invention. The prior art continues inspection without detecting lighting failures, causing a massive outflow of defects due to severe shading in all failure scenarios. In contrast, the present invention successfully restores the lighting environment when the degree of failure is minor (S4-01, 02, 04, 05, 06, 08, 09), thereby ensuring inspection quality without interrupting production. More importantly, when the degree of failure is severe and even minimum quality cannot be guaranteed (S4-03, 07, 10), it accurately detects this through a data-based 'reconfiguration failure' threshold and safely stops the equipment, thereby having the effect of fundamentally preventing the worst-case scenario in which a massive amount of defects are produced.

[0166] A comprehensive operation method of a machine vision unit that is easy to customize and adjust according to one embodiment of the present invention is described in detail. This embodiment assumes a complex situation in which a battery with boundary defects is inspected in an environment where surface characteristics are somewhat irregular and fine vibrations occur, while the equipment has been operated for a long time and some performance degradation has accumulated.

[0167] Just before the battery to be inspected is transported via a conveyor belt and placed at a predetermined position on the battery mounting unit (300), the control unit (400) performs a periodic system health diagnosis. First, the analysis judgment module (440) analyzes the past positioning error log received from the control unit (200) and calculates the mounting degree of freedom instability coefficient as 4.5 micrometers. At the same time, it analyzes the correction log of the automatic correction module (420) to calculate the correction cumulative deviation as 0.78 and calculates the maintenance debt cumulative index as 15,000 based on the data of the internal maintenance management module. The analysis judgment module (440) compares these three indicators with an internal 'judgment rule table' and, since the correction cumulative deviation and the maintenance debt cumulative index values ​​exceed the 'interest threshold' values ​​of 0.6 and 8,000, respectively, both conditions are simultaneously satisfied, and the overall 'maintenance requirement grade' is determined as 'Warning'. The control unit (400) records the corresponding 'warning' grade status in internal memory and sends a command to set up a notification to the operator after the inspection cycle is completed.

[0168] When the battery is fully seated in the battery mounting unit (300), the data collection module (430) collects micro-vibration data from the accelerometer sensor within the battery mounting unit (300), position determination feedback data from the control unit (200), and highlight variation data through low-resolution preliminary shooting. Based on the collected data, the analysis judgment module (440) calculates the micro-slip probability as 0.55, the highlight variation coefficient as 0.03, and the camera-actuator delay jitter as 1.3 milliseconds. The analysis judgment module (440) compares these indicators with another 'judgment rule table' and determines the overall 'inspection stability grade' as 'caution' because the micro-slip probability value exceeds the 'caution threshold' of 0.5. Accordingly, it transmits a command to the inspection condition control module (410) to dynamically increase the number of frame acquisitions from the default 1 to 3.

[0169] Upon receiving the above command, the inspection condition control module (410) transmits a command to control the image acquisition unit (110) to continuously capture three images at high speed. The three acquired frames are transmitted to the analysis judgment module (440) through the data collection module (430). The analysis judgment module (440) calculates the clarity score of each frame and selects the clearest frame, frame 2, which had the least influence from micro-vibrations, as the final analysis target.

[0170] The analysis judgment module (440) performs a precise analysis of the selected second frame to detect fine scratches on the battery surface and calculates an initial defect score of 60 points. Subsequently, to correct the reliability of the final judgment, the analysis judgment module (440) combines previously calculated system uncertainty indicators, namely the micro-slip probability (0.55) calculated in the previous step, the correction cumulative deviation (0.78) calculated in the system health diagnosis, and the reflection anisotropic variance (0.08), which is an intrinsic value of the battery model. The analysis judgment module (440) compares these indicators again with the 'judgment rule table' and determines the overall 'system uncertainty grade' as 'suspicious' because the micro-slip probability and the correction cumulative deviation values ​​each exceed the 'normal threshold', thus satisfying both conditions simultaneously. Based on the 'suspicious' grade, it internally executes a command to apply a 'second correction coefficient' of 1.5 to the initial defect score, and as a result, recalculates the final defect score to 90 points. Since the recalculated final defect score exceeds 70 points, which is a preset defect judgment threshold, the analysis judgment module (440) finally determines the battery as 'defective' and transmits the result to an external system.

[0171] Meanwhile, while the above equipment continues to operate, the control unit (400) detects a failure in unit 3 of the eight lighting units (120) during the next inspection cycle. The control unit (400) immediately pauses the inspection process and automatically switches to 'lighting reconfiguration mode'. The control unit (400) photographs a reference plate to confirm that the root mean square error, which is the current lighting uniformity deviation, is 45.8, and loads an 'ideal lighting distribution map' from memory. Subsequently, the control unit (400) performs an 'iterative optimization algorithm'. While repeatedly fine-tuning the light intensity and angle of the remaining seven normal lighting units (120), including lighting units 2 and 4 (120) adjacent to the failed unit 3, the change in the root mean square error value is measured at each step. After tens of iterations, optimization is terminated at the point of 8.5, where the root mean square error value no longer decreases significantly, and a command is sent to apply the newly derived settings to each normal lighting unit (120). At the same time, to compensate for the reduced total light intensity, the camera exposure time is set to increase by 5% through the inspection condition control module (410).

[0172] When the lighting reconfiguration is complete, the control unit (400) displays two combined notifications on the operator interface. First, based on the initially determined 'Warning' grade, it displays a message stating, "Immediate inspection required: Accumulated calibration deviation and approaching maintenance debt threshold." Second, based on the lighting reconfiguration result, it additionally displays a message stating, "Lighting unit 3 (120) failed. Automatic reconfiguration of lighting environment completed in standby mode. Production continues, but please inspect as soon as possible." After all notifications are displayed, the control unit (400) automatically resumes the inspection process under the reconfigured lighting environment.

[0173] Through the comprehensive embodiment described above, it is clearly evident that the present invention goes beyond simple defect detection to comprehensively diagnose the mechanical, electronic, and operational status of equipment, actively responds to unstable inspection environments, self-evaluates the reliability of measurement results to correct judgments, and performs organic operations to minimize production interruptions and self-recover even in the event of hardware failure, thereby providing significantly improved inspection reliability and operational efficiency compared to conventional technology.

[0174] Although preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above. Various modifications are possible by those skilled in the art without departing from the essence of the invention as claimed in the claims, and such modifications should not be understood individually from the technical spirit or perspective of the present invention. Explanation of the symbols

[0175] Machine vision unit (100) Image acquisition unit (110) Lighting unit (120) Control unit (200) Battery mounting unit (300) Control unit (400) Inspection condition control module (410) Automatic correction module (420) Data collection module (430) Analysis judgment module (440)

Claims

Claim 1 In a machine vision unit used in battery automation inspection equipment, the machine vision unit comprises: an image acquisition unit that captures a surface image of a battery, which is a subject; The system includes a lighting unit that irradiates light onto the subject; and operates in conjunction with a control unit that receives sensor data from a sensor provided in a battery mounting unit that fixes the battery and a control unit that adjusts the position of the machine vision unit; the image acquisition unit receives a number of frames dynamically determined according to the stability grade at the time of inspection from the control unit, and performs single shooting or multi-frame shooting according to the received number of frames acquired; the control unit calculates a micro-slip probability based on vibration data received from an accelerometer sensor embedded in the battery mounting unit, and calculates a camera-actuator delay jitter by calculating the standard deviation of the difference between the positioning completion flag timestamp and the exposure start timestamp included in the positioning feedback data received from the control unit; if either the calculated micro-slip probability or the camera-actuator delay jitter exceeds a preset 'caution threshold', the stability grade is determined as 'caution' and the number of frames acquired is determined as a plurality of first acquisition numbers; and the image acquisition unit changes the shooting parameters A machine vision unit that is easy to customize and adjust, characterized by performing continuous multi-frame shooting as many times as the first number of acquisitions without, and the control unit calculating a sharpness score for all of the acquired multiple images and selecting the single image with the highest score as the final analysis target, thereby minimizing the effect of image blur. Claim 2 delete

Images