Image and mechanical signal fusion-based online detection method, system and device for back adhesive defects and medium

Abstract

This invention discloses an online detection method, system, device, and medium for adhesive backing defects based on image and mechanical signal fusion, relating to the field of material defect detection technology. The method includes acquiring adhesive backing images by obtaining the adhesive viscosity and limiting the coating speed; establishing a mechanical spatial cycle by combining production line mechanical signals; extracting periodic flow mark textures and performing order spectrum analysis to identify normal textures and candidate defects; constructing a pre-processed suppression template based on normal textures for defect detection; and simultaneously selecting the most stable coating speed by adjusting the coating speed to perform secondary acquisition and confirmation of candidate defect areas. This solves the problem of difficulty in distinguishing normal flow mark textures from real stripe defects and the high false detection rate during adhesive backing coating.

Description

Technical Field

[0001] This invention relates to the field of material defect detection technology, and more specifically, to a method, system, device, and medium for online detection of adhesive defects based on the fusion of image and mechanical signals. Background Technology

[0002] In roll-to-roll adhesive backing production, pure vision inspection systems typically rely on high-speed linear scan cameras to acquire continuous images of the material surface. Image preprocessing methods such as grayscale normalization, histogram equalization, or filtering are used to enhance the visibility of abnormal areas, enabling automatic identification and location of stripes, scratches, or other defects. The advantage of this type of system is that it does not require contact with the material, allowing for online inspection and rapid response to defect changes on the production line. However, pure vision inspection systems have inherent limitations when dealing with complex process textures: when normal process features exhibit periodicity in the image or have similar grayscale changes to defect textures, the system struggles to accurately distinguish between real defects, easily leading to false positives and false negatives.

[0003] During roll-to-roll adhesive coating, the fluid spreading between the coating roller and the substrate creates flow marks on the material surface, distributed along the conveying direction. This texture is a normal process characteristic, caused by factors such as slight eccentricity of the coating roller, variations in adhesive viscosity, and periodic fluctuations in coating pressure. Under conventional image enhancement processing, normal flow marks and genuine stripe defects exhibit similar periodic variations in image grayscale, causing the enhanced image to simultaneously highlight both process textures and potential defect areas. Due to the similar frequency characteristics of both in grayscale space, pure visual inspection systems struggle to distinguish between normal process textures and genuine defects, leading to a significantly increased false detection rate for stripe-type defects.

[0004] In the field of rotating machinery vibration signal analysis, order tracking analysis is a mature method for identifying periodic vibration components. This method collects mechanical vibration signals and shaft rotation speed signals, converts the vibration signals into an order spectrum, and then identifies normal and abnormal vibrations synchronized with the rotation speed. In the context of adhesive coating inspection, the flow mark texture formed during the coating process can be analogized to a "normal order texture" generated by the periodic movement of the coating roller. An attempt is made to distinguish normal process textures from abnormal defects through periodic analysis, achieving a separation of normal structures and abnormal signals similar to that in rotating machinery fault diagnosis.

[0005] However, the direct application of analog order tracking analysis in adhesive backing image processing faces significant obstacles. First, the image data itself lacks a mechanical reference signal, making it impossible to directly establish a correspondence between texture frequency and the mechanical cycle of the coating roller. Second, coating flow mark textures are not strictly periodic signals; their spatial spacing varies with coating speed, adhesive viscosity, and material tension, resulting in unstable spatial frequencies and making it difficult to reliably identify normal and defective textures using traditional order analysis. To address these problems, this invention proposes a solution. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an online detection method, system, device, and medium for adhesive defects based on the fusion of image and mechanical signals. By fusing periodic texture analysis and order suppression of image and mechanical signals, the problem of difficulty in distinguishing normal flow marks from real stripe defects and high false detection rate during adhesive coating is solved.

[0007] To achieve the above objectives, the present invention provides the following technical solution: An online detection method for adhesive defects based on image and mechanical signal fusion includes the following steps: obtaining the current adhesive viscosity and limiting the coating speed; acquiring a linear array image sequence of the adhesive surface and preprocessing it using histogram equalization to obtain a first preprocessed sequence; establishing a mechanical spatial period based on the production line mechanical signals and extracting the grayscale change sequence in the conveying direction based on the first preprocessed sequence, and obtaining a periodic flow mark texture sequence through spatial autocorrelation; performing mechanical periodic normalization on the periodic flow mark texture sequence based on the coating roller speed signal, constructing a texture order sequence and performing order spectrum analysis to identify the normal flow mark texture order. The order of candidate stripe defects is determined; a preprocessing suppression template is constructed based on the normal flow mark texture order, and the preprocessing is subjected to order suppression to obtain a second preprocessing sequence, which is then used for machine vision adhesive defect detection; while keeping the adhesive viscosity constant, the coating speed is changed, and texture extraction and order analysis are repeated. The flow mark texture periodic performance under different coating speeds is enumerated and compared, and the coating speed with the most stable periodic performance is selected as the optimal coating speed corresponding to the adhesive viscosity; when a candidate stripe defect is detected, the coating speed is controlled to the optimal coating speed, and the corresponding area is re-acquired and a second detection is performed for confirmation.

[0008] An online defect detection system for adhesive backing based on image and mechanical signal fusion includes a material acquisition module, a mechanical spatial period establishment module, a texture order tracking module, a preprocessing suppression module, a period stability control module, and a defect confirmation module. The material acquisition module obtains the current adhesive viscosity and limits the coating speed, acquires a linear array image sequence of the adhesive backing surface, and preprocesses it using histogram equalization to obtain a first preprocessed sequence. The mechanical spatial period establishment module establishes a mechanical spatial period by combining the production line's mechanical signals and extracts the grayscale change sequence in the conveying direction based on the first preprocessed sequence, obtaining a periodic flow mark texture sequence through spatial autocorrelation. The texture order tracking module performs mechanical period normalization on the periodic flow mark texture sequence based on the coating roller speed signal to construct... The system employs a texture order sequence and performs order spectrum analysis to identify the order of normal flow mark textures and candidate stripe defects. A preprocessing suppression module constructs a preprocessing suppression template based on the normal flow mark texture order, performs order suppression on the preprocessing to obtain a second preprocessing sequence, and performs machine vision-based adhesive defect detection. A periodic stability control module changes the coating speed while maintaining constant adhesive viscosity, repeatedly performing texture extraction and order analysis. It enumerates and compares the periodic performance of flow mark textures at different coating speeds, selecting the coating speed with the most stable periodic performance as the optimal coating speed corresponding to the adhesive viscosity. A defect confirmation module controls the coating speed to the optimal coating speed when a candidate stripe defect is detected, re-acquires images of the corresponding area, and performs secondary detection for confirmation.

[0009] An electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform an online detection method for adhesive backing defects based on image and mechanical signal fusion.

[0010] A computer-readable storage medium storing a computer program that, when executed by a processor, implements an online detection method for adhesive backing defects based on the fusion of image and mechanical signals.

[0011] The technical effects and advantages of this invention, which relates to an online detection method, system, equipment, and medium for adhesive backing defects based on image and mechanical signal fusion, are as follows: 1. This invention achieves precise periodic analysis and order tracking of flow mark textures on the adhesive backing surface by fusing image data with mechanical signals from the coating process. In the image preprocessing stage, normal process textures are distinguished from potential defect textures through grayscale enhancement, spatial autocorrelation, and mechanical periodic normalization, constructing a normal flow mark texture order sequence. Normal textures are then suppressed in the preprocessing suppression template, thereby significantly reducing the false detection rate of stripe-like defects while preserving true defect characteristics. This method fully utilizes mechanical signals such as coating roller speed and material conveying speed as references, ensuring high synchronization between periodic textures and mechanical motion, thus improving the stability and reliability of machine vision inspection.

[0012] 2. This invention enumerates the periodic behavior of flow mark textures at different coating speeds to screen the optimal coating speed for the corresponding adhesive viscosity. When candidate defects are detected, the optimal coating speed is used for secondary data acquisition and confirmation, achieving dynamic feedback optimization. This technology can automatically adapt to changes in adhesive properties and processes under actual production conditions, making online detection of adhesive defects more accurate while ensuring production efficiency and product consistency. It provides a reliable quality monitoring method for high-precision roll-to-roll coating processes. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the online detection method for adhesive defects based on the fusion of image and mechanical signals according to the present invention.

[0014] Figure 2 This is a schematic diagram of the online detection system for adhesive defects based on the fusion of image and mechanical signals according to the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1, Figure 1 The present invention provides an online detection method for adhesive defects based on image and mechanical signal fusion, comprising the following steps: S1, obtain the current viscosity of the adhesive and limit the coating speed, acquire the linear array image sequence of the adhesive surface, and preprocess it through histogram equalization to obtain the first preprocessed sequence.

[0017] In this embodiment, the steps of obtaining the current adhesive viscosity and limiting the coating speed, acquiring a linear array image sequence of the adhesive surface, and preprocessing it using histogram equalization to obtain the first preprocessed sequence are as follows: The viscosity signal of the current batch of adhesive solution is acquired by collecting the viscosity signal of the adhesive solution through an online rotational viscometer. The viscosity signal is continuously sampled and the average viscosity value of the stable range is extracted to obtain the viscosity parameter of the current adhesive solution. Based on the viscosity parameters of the adhesive, the corresponding coating speed range is matched in the preset coating process parameter table, and the current coating speed of the production line is set within the coating speed range; Under the current coating speed conditions, start the roll-to-roll adhesive coating production line, and use a line scan camera set downstream of the coating section to scan and collect data on the surface of the continuously conveyed adhesive material line by line. Record each scan line according to the material conveying order and construct the original image sequence. The original image sequence is processed frame by frame to perform grayscale conversion, and the RGB pixel values ​​of each frame are mapped to single-channel grayscale values ​​according to the grayscale conversion relationship, and arranged in the acquisition order to form a continuous grayscale sequence. For each frame of the continuous grayscale sequence, the grayscale distribution is statistically analyzed, a grayscale histogram is established according to the grayscale range, and the corresponding cumulative grayscale distribution function is calculated based on the grayscale histogram. The gray-level cumulative distribution function is used to map and transform the gray-level values ​​of each pixel in the corresponding frame, mapping the original gray-level values ​​to a new gray-level range to enhance the gray-level contrast of the image and obtain the corresponding gray-level enhanced frame. The grayscale enhancement frames are rearranged and combined according to the acquisition order of the material conveying direction to form an enhanced image sequence; The enhanced image sequence is output as the first preprocessed sequence.

[0018] In this embodiment, the current viscosity of the adhesive is the flow resistance characteristic value of the adhesive obtained by continuous sampling under the flow state of the adhesive using an online rotational viscometer. Specifically, when the adhesive passes through the measurement area in the coating section during the production process, the viscometer records its shear force and flow velocity response. By analyzing the stability of the adhesive flow state over a certain period of time and extracting the mean value, the viscosity parameter representing the current physical state of the adhesive is obtained.

[0019] In this embodiment, the coating speed is a production speed manually set within the allowable coating speed range obtained by matching the current adhesive viscosity in the preset coating process parameter table. Specifically, it is a speed that can ensure uniform spread of the adhesive while avoiding the aggravation of flow marks, and serves as the target for conveying and controlling the roller speed of the roll-to-roll coating system in the production line.

[0020] In this embodiment, the original image sequence is a set of continuous grayscale frames acquired by scanning the surface of the continuously conveyed adhesive material line by line with a linear scan camera located downstream of the coating section. Specifically, each scan line acquired in the order of material conveying forms an image frame, and the frames are arranged in chronological order to form a continuous sequence for subsequent image preprocessing and texture analysis.

[0021] In this embodiment, the continuous grayscale sequence is a data sequence formed by mapping each frame of the original image sequence into a single-channel grayscale value through grayscale conversion and arranging them according to the acquisition order. Specifically, the brightness information of the pixels in each frame image is mapped into continuous grayscale values ​​and arranged along the material conveying direction so as to reflect the grayscale change pattern in subsequent spatial analysis.

[0022] In this embodiment, the grayscale histogram is a distribution map that statistically analyzes the frequency of pixel grayscale values ​​in each frame of grayscale image and arranges them according to grayscale level. Specifically, it counts the number of pixels at each grayscale level from the darkest to the brightest to characterize the grayscale distribution characteristics of the image and is used for subsequent equalization processing.

[0023] In this embodiment, the cumulative gray-level distribution function is a function that performs cumulative frequency mapping on pixel gray-level values ​​based on the gray-level histogram. Specifically, it accumulates the number of pixels at each gray level in order of gray level and normalizes it to obtain the cumulative proportion corresponding to each gray level, which is used to guide the gray-level mapping transformation.

[0024] In this embodiment, the grayscale enhancement frame is an image frame generated by mapping the original grayscale value according to the grayscale cumulative distribution function. Specifically, each pixel finds a corresponding new grayscale level in the cumulative distribution function based on its original grayscale value, thereby enhancing the brightness and contrast of the image and improving the distinguishability of subsequent defects and texture features.

[0025] In this embodiment, the first preprocessing sequence is an image sequence formed by rearranging and combining each grayscale enhancement frame according to the acquisition order in the material conveying direction. Specifically, each enhanced frame image is arranged in order to maintain spatial continuity, providing a basic data sequence for subsequent mechanical spatial period establishment and flow mark texture extraction.

[0026] It should be noted that the frame-by-frame histogram equalization of the continuous grayscale sequence in the above steps not only improves the image contrast, but also, by maintaining the relative order of grayscale changes, enables normal flow mark textures and potential defects to be accurately distinguished in subsequent mechanical mapping and order analysis, thus laying a precise data foundation for order suppression and defect detection.

[0027] It should be noted that the process of collecting adhesive viscosity and setting coating speed not only involves the measurement of physical parameters, but also relates to the spatial periodic stability of the subsequent flow mark texture. By matching the combination of adhesive viscosity and coating speed, the formation characteristics of flow mark texture can be controlled in the pretreatment stage, providing a controllable process benchmark for order spectrum analysis.

[0028] It should be noted that the significance of this step in the whole method is that by accurately acquiring the physical state of the adhesive and the coating speed, and generating a high-contrast first preprocessing sequence, a reliable data foundation is provided for subsequent establishment of mechanical spatial period, extraction of periodic flow mark texture, texture order analysis, and suppression of interference of normal flow mark texture on defect detection, thereby effectively reducing the probability of normal flow mark texture being misjudged as a defect.

[0029] In this embodiment, the online rotational viscometer is a real-time measuring device installed in the adhesive delivery pipeline. Its output data includes shear stress response value, rotor speed and corresponding timestamp. By continuously sampling, a viscosity time series is formed to reflect the rheological state change process of the adhesive under the current working conditions.

[0030] In this embodiment, the stable interval is a period of time in the viscosity time series obtained by continuous sampling where the viscosity fluctuation amplitude is continuously within the preset fluctuation threshold range. Specifically, by comparing the change amplitude between adjacent sampling points point by point, when the change amplitude is continuously less than the set threshold and the duration exceeds the set time length, the time period is marked as the stable interval, and the average viscosity value is extracted as the representative value within the interval.

[0031] In this embodiment, the preset coating process parameter table is a multi-parameter mapping table established based on historical production data. It includes at least the viscosity range of the adhesive, the corresponding recommended coating speed range, and the corresponding coating quality grade label. The data is formed by classifying and organizing the viscosity, speed, and defect detection results recorded during historical batch production.

[0032] In this embodiment, the coating speed range is the speed interval obtained by matching the current adhesive viscosity parameter in the preset coating process parameter table. Its upper and lower limits are determined by the speed boundary under the viscosity condition in historical production where there is no obvious increase in flow marks and defects, and can be obtained by screening historical defect rate data.

[0033] In this embodiment, the acquisition parameters of the line scan camera include exposure time, sampling frequency, and resolution. The exposure time is set according to the reflectivity of the material surface to avoid overexposure or underexposure. The sampling frequency is matched according to the material transport speed to ensure that there is no overlap or gap between adjacent scan lines. The resolution is selected according to the minimum identifiable size of the defect.

[0034] In this embodiment, the grayscale conversion relationship is a rule that maps the RGB three-channel pixel values ​​to a single-channel grayscale value according to a preset weight combination. The weights are calibrated based on the camera's spectral response characteristics and the material's surface reflectivity to ensure that the grayscale value can truly reflect changes in surface brightness.

[0035] In this embodiment, the gray level range is the gray value interval determined according to the camera output bit depth. For example, when using 8-bit gray level representation, the gray level range is a discrete set of levels from the minimum gray value to the maximum gray value, which is used to statistically analyze the gray level distribution.

[0036] In this embodiment, the normalization process in the gray-scale cumulative distribution function is to map the cumulative number of pixels to the total number of pixels in the current frame, so that each gray level corresponds to a monotonically increasing ratio value from low to high, which is used to guide the subsequent gray-scale mapping process.

[0037] In this embodiment, the grayscale mapping transformation is to redistribute the original grayscale values ​​to a new grayscale level range according to the correspondence of the grayscale cumulative distribution function. Specifically, it is implemented by finding the cumulative proportion corresponding to the original grayscale value of each pixel and mapping the proportion to the target grayscale level, thereby achieving contrast stretching.

[0038] In this embodiment, the material conveying direction is the actual movement direction of the material during roll-to-roll production, which corresponds to the arrangement direction of the scanning lines of the line scan camera in the image sequence, in order to ensure the consistency of subsequent sequence rearrangement and spatial analysis.

[0039] It should be noted that, in addition to the viscosity signal, ambient temperature, adhesive temperature and production line tension signals can also be collected simultaneously in the above steps, and these can be used as auxiliary data to determine the reliability of the viscosity stability range. When the temperature or tension changes abruptly, the stability range can be re-screened to improve the representativeness of the viscosity parameters.

[0040] It should be noted that when setting the coating speed range, the historical false detection rate data of the defect detection system can be combined to classify and screen different speed ranges, and the speed range with a lower false detection rate in the historical data can be selected as the candidate range, thereby reducing subsequent detection interference at the source.

[0041] It should be noted that this step, by standardizing the viscosity of the adhesive, the coating speed, and the grayscale features of the image, provides stable input conditions for establishing the mechanical spatial cycle in the future. On the other hand, by enhancing the grayscale contrast of the image, the flow mark texture and defect features are fully revealed, thus providing a separable data basis for subsequent order analysis and texture suppression based on mechanical signals, effectively reducing the misjudgment of defects caused by the enhancement of process texture.

[0042] S2, combine the mechanical signals of the production line to establish the mechanical spatial cycle, and extract the grayscale change sequence of the conveying direction based on the first preprocessing sequence, and obtain the periodic flow mark texture sequence through spatial autocorrelation.

[0043] In this embodiment, the process of establishing a mechanical spatial period by combining the mechanical signals of the production line, extracting the grayscale change sequence of the conveying direction based on the first preprocessing sequence, and obtaining a periodic flow mark texture sequence through spatial autocorrelation is specifically as follows: After obtaining the first preprocessing sequence, the rotation angle signal of the coating roller and the material conveying speed signal are collected by the mechanical sensor installed in the coating section. The signals are synchronously sampled to obtain the mechanical motion parameter sequence corresponding to each acquisition frame. Based on the mechanical motion parameter sequence, a spatial mapping is established between the first preprocessing sequence and the mechanical motion according to the material conveying direction. The grayscale enhancement data of each frame is mapped to the corresponding mechanical motion position to form a mechanical reference grayscale sequence. The average gray value is extracted line by line from the mechanical reference gray value sequence along the material conveying direction to generate a gray value change sequence, where each element corresponds to the average gray value information of the material at the mechanical motion reference position. A spatial autocorrelation operation is performed on the grayscale change sequence, and the spatial periodic characteristic sequence of grayscale change is obtained by analyzing the autocorrelation coefficient of grayscale values ​​under different spatial lag distances. Identify the main peaks and corresponding hysteresis intervals in the spatial periodic feature sequence, and map the hysteresis intervals to a periodic flow mark texture sequence on the material surface.

[0044] In this embodiment, the mechanical sensor is a detection device installed in the coating section, used to collect the rotation angle signal of the coating roller and the material conveying speed signal in real time. Specifically, the sensor continuously monitors every minute angle change of the coating roller and the moving speed of the material in the coating section, and outputs these signals synchronously in time sequence to form a mechanical motion parameter sequence, providing a reference benchmark for the subsequent establishment of mechanical space cycle.

[0045] In this embodiment, the mechanical motion parameter sequence is a data sequence formed by arranging the coating roller rotation angle and material conveying speed signals collected by the mechanical sensor in the order of collection time. Specifically, the coating roller angle and material displacement information corresponding to each collection frame are combined to represent the mechanical motion state corresponding to each position on the material surface, so that the image grayscale information can be mapped to the mechanical motion position.

[0046] In this embodiment, the mechanical reference grayscale sequence is a sequence obtained by establishing a mapping relationship between the grayscale enhancement data of each frame in the first preprocessing sequence and the mechanical motion parameter sequence according to the material conveying direction. Specifically, each pixel or each row of grayscale value corresponds to the coating roller angle or material displacement position, so as to ensure that the spatial position of grayscale change can be precisely aligned with the mechanical cycle.

[0047] In this embodiment, the grayscale change sequence is a sequence formed by extracting the average grayscale value of the mechanical reference grayscale sequence line by line along the material conveying direction. Specifically, the grayscale value of each line or each small area is averaged to generate average brightness information representing the material at the corresponding mechanical position, which is used for subsequent periodic texture analysis.

[0048] In this embodiment, spatial autocorrelation is a method for analyzing the repeatability of gray values ​​at different lag distances in a gray-scale change sequence. Specifically, a lag distance is selected along the sequence, and the similarity of the sequence itself with the delayed sequence is compared to reflect the repeating pattern and spatial periodic characteristics of gray-scale changes.

[0049] In this embodiment, the periodic flow mark texture sequence is a sequence formed by identifying the main peaks and their corresponding hysteresis intervals based on the spatial autocorrelation results. Specifically, the spacing of the grayscale repetition pattern is mapped to the actual length of the material surface, thereby obtaining the flow mark texture position sequence that appears periodically along the conveying direction, providing input for subsequent mechanical periodic normalization and order analysis.

[0050] It should be noted that the construction of the periodic flow mark texture sequence is to overcome the obstacles in applying order tracking analysis of analog rotating machinery vibration signals to adhesive coating images. The mechanical cycle information of the coating roller cannot be directly obtained from purely visual images. Furthermore, the spatial spacing of the flow mark texture is affected by fluctuations in coating speed, adhesive viscosity, and tension, leading to periodic instability. By establishing a mechanical reference grayscale sequence and performing spatial autocorrelation analysis, the repetitive pattern of image grayscale changes can be transformed into a relatively stable periodic flow mark texture sequence. This allows subsequent mechanical cycle normalization and order analysis to reference real mechanical motion, thereby distinguishing between normal flow mark textures and abnormal stripe defects generated by the periodic motion of the coating roller.

[0051] It should be noted that this step not only maps the image grayscale sequence to the mechanical signal, but also extracts the flow mark texture into a quantifiable sequence through periodic analysis, providing a necessary foundation for subsequent construction of texture order sequences and suppression of normal process textures. This step ensures that subsequent defect detection methods can identify abnormal stripe defects without being interfered with by normal flow mark textures.

[0052] In this embodiment, the coating roller rotation angle signal is the angle change data output by the encoder installed at the end of the coating roller shaft. Specifically, the encoder outputs a pulse signal every time the coating roller rotates by a fixed angle. The continuous angle sequence is obtained by the relationship between the cumulative number of pulses and the angle corresponding to a single pulse, which is used to characterize the precise rotation position of the coating roller at any time.

[0053] In this embodiment, the material conveying speed signal is the linear velocity data of the material collected by a speed measuring device installed on the traction roller or conveying roller. Specifically, it is the material movement distance per unit time obtained by converting the pulse signal or voltage signal generated by the contact between the speed measuring wheel and the material, and forming a continuous speed sequence corresponding to the timestamp.

[0054] In this embodiment, synchronous sampling is the process of aligning the frame acquisition time of the line scan camera with the output time of the mechanical sensor. Specifically, it is achieved by using a trigger signal or timestamp matching method to ensure that each frame of image data corresponds to a unique coating roller angle value and material conveying speed value, thereby ensuring the consistency of image data and mechanical data in the time dimension.

[0055] In this embodiment, the spatial mapping relationship is to establish a one-to-one correspondence between the position of the image frame in the material conveying direction and the actual displacement of the material and the rotation angle of the coating roller. Specifically, the frame spacing is converted by the material conveying speed, the position of each image frame in the sequence is converted into the actual material length position, and a unified reference coordinate is formed by combining the coating roller angle information.

[0056] In this embodiment, the average gray value is extracted row by row by summing the pixel rows of each frame image along the direction perpendicular to the transport direction. Specifically, the gray values ​​of all pixels in the same row are accumulated and divided by the number of pixels to obtain a single gray value that represents the overall brightness level of the row, so as to reduce the impact of local noise on the overall trend.

[0057] In this embodiment, the hysteresis distance is the displacement interval selected in the gray-scale change sequence, specifically the spatial distance represented by the sequence index interval. By gradually increasing this interval and comparing the similarity between the original sequence and the delayed sequence, the position interval where the gray-scale change recurs is found.

[0058] In this embodiment, the autocorrelation coefficient is an index used to measure the similarity between the original gray-level change sequence and its lag sequence. Specifically, it is a similarity measure obtained by normalizing the difference in gray-level values ​​at corresponding positions. The closer the value is to the maximum value, the stronger the repeatability.

[0059] In this embodiment, the main peak is the autocorrelation response point that is significantly higher than the neighboring position in the spatial autocorrelation results. Specifically, by comparing the autocorrelation response intensity under adjacent lag distances, when the response value of a certain lag position reaches the maximum in the local range and exceeds the set proportion threshold, that position is determined as the main peak, which is used to characterize the texture repetition period.

[0060] In this embodiment, the process of mapping the hysteresis distance to the actual length of the material is to convert the hysteresis distance in the sequence into the corresponding actual physical distance based on the material conveying speed and the sampling time interval, thereby obtaining the spatial period length of the flow mark texture on the material surface.

[0061] In this embodiment, the spatial periodic feature sequence is a data set consisting of multiple hysteresis distances and their corresponding autocorrelation response intensities. Specifically, it is used to record the distribution of the strength of grayscale change repeatability under different spatial intervals, so as to support the subsequent screening and stability judgment of periodic textures.

[0062] It should be noted that a limitation on the hysteresis distance search range can be introduced during the spatial autocorrelation analysis process. This range is preset based on the coating roller diameter, theoretical rotation speed, and material conveying speed to avoid interference from irrelevant long or short periods, thereby improving the accuracy of period identification.

[0063] It should be noted that before extracting the grayscale variation sequence, the mechanical reference grayscale sequence can be detrended, that is, the part of the overall brightness that changes slowly is removed and only the local fluctuation component is retained, so that the spatial autocorrelation result can better highlight the periodic texture features.

[0064] It should be noted that this step introduces mechanical signals and establishes a spatial correspondence between images and mechanical motion, transforming the texture analysis, which originally relied solely on image grayscale changes, into an analysis process associated with the mechanical cycle. At the same time, by extracting stable repetitive patterns through spatial autocorrelation, unstable flow mark textures are transformed into describable periodic sequences, providing an aligned data foundation for subsequent mechanical cycle normalization and order analysis, thereby enabling effective differentiation between normal process textures and abnormal stripe defects.

[0065] S3. Based on the coating roller speed signal, the periodic flow mark texture sequence is mechanically normalized to construct a texture order sequence and perform order spectrum analysis to identify the normal flow mark texture order and the candidate stripe defect order.

[0066] In this embodiment, the mechanical periodic normalization of the periodic flow mark texture sequence based on the coating roller rotation speed signal, the construction of a texture order sequence, and the execution of order spectrum analysis to identify the normal flow mark texture order and the candidate stripe defect order are specifically as follows: The real-time rotation speed signal of the coating roller and the synchronous material conveying speed signal are collected. The signals are continuously sampled and a time correspondence is established with the periodic flow mark texture sequence to achieve mechanical cycle normalization, so that each flow mark texture point corresponds to the rotation angle position of the coating roller. The normalized flow mark texture sequence is segmented according to the coating roller rotation cycle, and each segment of texture data is mapped to the corresponding mechanical order position to form a texture order sequence, where each order represents the grayscale change pattern synchronized with the coating roller rotation cycle. Perform order spectrum analysis on the texture order sequence, and analyze the amplitude spectrum and phase spectrum of each order through Fourier transform to obtain the frequency distribution characteristics of the texture on the mechanical cycle. Peak identification is performed on the amplitude spectrum and phase spectrum. The order that is synchronized with the mechanical cycle of the coating roller and has a continuous and stable amplitude is identified as the normal flow mark texture order. The order that is not at the synchronous peak position or has an unstable amplitude is marked as the candidate stripe defect order. The identified normal flow mark texture order and candidate stripe defect order are output as input sequences for subsequent construction of preprocessing suppression templates and backing adhesive defect detection, which are used to distinguish between process textures and real stripe defects.

[0067] In this embodiment, the coating roller speed signal is the number of rotations per unit time output by a speed sensor installed at the end of the coating roller shaft. Specifically, the number of pulses per unit time is converted into a speed value, and a continuous speed sequence is formed by corresponding timestamps, which is used to characterize the real-time rotation state of the coating roller during the production process.

[0068] In this embodiment, mechanical cycle normalization is the process of uniformly mapping the spatial or temporal positions in the periodic flow mark texture sequence to the range of a single rotation cycle of the coating roller. Specifically, it involves determining the rotation angle corresponding to each moment based on the rotation speed signal, compressing or expanding the texture data within multiple cycles to a standard cycle interval, so that the data in different time periods have a unified cycle reference.

[0069] In this embodiment, the time correspondence is a one-to-one correspondence between the texture sequence sampling points and the rotation speed signal sampling points established through the synchronous sampling mechanism. Specifically, each texture sampling point can be traced back to the corresponding acquisition time and further matched to the coating roller rotation speed and rotation angle position at that moment.

[0070] In this embodiment, rotation cycle segmentation is the process of dividing the normalized flow mark texture sequence into multiple equally periodic segments based on the time or angle range corresponding to the coating roller completing one full rotation. Specifically, the continuous texture data is segmented with an angle from zero to a full circle as one interval.

[0071] In this embodiment, the mechanical order position is a discrete position identifier after dividing the different angular positions within the rotation cycle according to equal proportions. Specifically, one rotation cycle is divided into a fixed number of intervals, and each interval corresponds to an order position, which is used to represent the relative position of the texture in the mechanical cycle.

[0072] In this embodiment, the texture order sequence is a sequence formed by rearranging the texture data after each rotation cycle segment according to the mechanical order position. Specifically, it is to align the grayscale changes at the same order position in different cycles, thereby forming a data set that reflects the stable characteristics of each order.

[0073] In this embodiment, the amplitude spectrum is the intensity distribution of each order component obtained after frequency decomposition of the texture order sequence. Specifically, it represents the magnitude of grayscale change corresponding to each order and is used to determine the contribution of that order to the overall texture.

[0074] In this embodiment, the phase spectrum is a sequence that reflects the positional relationship of each order component within the mechanical cycle. Specifically, it represents the offset of each order grayscale change relative to the starting position of the cycle, and is used to analyze the consistency of texture distribution within the cycle.

[0075] In this embodiment, peak identification is the process of finding response points in the amplitude spectrum that are significantly higher than the background level. Specifically, it involves comparing the amplitude of each order with the overall amplitude distribution. When the amplitude of a certain order exceeds a preset ratio threshold and is a local maximum among adjacent orders, it is marked as a valid peak.

[0076] In this embodiment, continuous and stable amplitude means that the amplitude change of the same order remains within a preset fluctuation range in multiple consecutive rotation cycles. Specifically, the amplitude of the same order in multiple cycles is compared, and when its change is always lower than a set threshold, it is identified as a stable order.

[0077] In this embodiment, the candidate stripe defect order is a set of orders that do not meet the amplitude stability requirement or are not synchronized with the rotation cycle of the coating roller. Specifically, it includes orders with large amplitude fluctuations, positions that shift in different cycles, or orders that do not form stable peak values. These are the key analysis areas for subsequent defect detection.

[0078] It should be noted that when performing mechanical cycle normalization, the rotation speed signal can be smoothed to eliminate the influence of instantaneous fluctuations on angle mapping, thereby ensuring the continuity and consistency of the texture sequence during the cycle mapping process.

[0079] It should be noted that when constructing the texture order sequence, a minimum number of valid orders can be set. If a certain order does not form continuous valid data in multiple periods, it can be removed to avoid noise interference with the order spectrum analysis.

[0080] It should be noted that the amplitude threshold and stability threshold used in the peak recognition process can be set based on historical production data. Specifically, the amplitude distribution range of each order under normal production conditions is statistically analyzed, and the range that can cover most normal textures is selected as the judgment standard, thereby improving the accuracy of recognition.

[0081] It should be noted that this step, by introducing mechanical cycle normalization and order spectrum analysis, transforms the flow mark texture, which originally only appears in the spatial domain, into an order feature synchronized with the rotation cycle of the coating roller. This makes the normal process texture appear as a stable fixed order component, while the abnormal stripe defects appear as asynchronous or unstable orders, thereby achieving separation of the two at the feature level and providing a clear basis for subsequent preprocessing suppression and defect detection.

[0082] S4. A preprocessing suppression template is constructed based on the normal flow mark texture order. The preprocessing is subjected to order suppression to obtain the second preprocessing sequence, and machine vision adhesive defect detection is performed.

[0083] In this embodiment, the step of constructing a preprocessing suppression template based on the normal flow mark texture order, performing order suppression on the preprocessing to obtain a second preprocessing sequence, and performing machine vision adhesive defect detection specifically involves: After obtaining the normal flow mark texture order and the candidate stripe defect order, for each frame of the first preprocessing sequence, the grayscale data and texture order sequence are mapped along the material conveying direction, and the grayscale change pattern corresponding to each normal flow mark texture order is marked as the process texture region in the image sequence. Based on the marked process texture region, a preprocessing suppression template is established. By suppressing the gray-level enhancement amplitude of the corresponding region, the gray-level fluctuation of normal flow mark texture in the image is reduced, while the original gray-level features of the unmarked region are preserved, forming a second preprocessing sequence. The second preprocessed sequence is used for defect detection based on machine vision. Feature extraction is performed on each frame of the image, and stripe-like defects on the adhesive surface are identified according to the set defect judgment rules. For regions corresponding to candidate stripe defect orders, texture analysis is performed first. For each frame of detection results, continuous frame fusion is performed, and the defect regions that appear repeatedly in adjacent frames are merged to generate the final defect location sequence and defect level information, which is then output as the online detection result of adhesive defects.

[0084] In this embodiment, mechanical cycle normalization is the process of aligning the periodic flow mark texture sequence with the rotation angle signal of the coating roller and the material conveying speed signal. Specifically, based on the time position of each flow mark texture point in the image sequence, the corresponding coating roller rotation angle and material conveying displacement are found, and the flow mark texture point is mapped to the corresponding mechanical position, so that the flow mark texture data is presented with the coating roller rotation cycle as the reference scale, thereby eliminating the spatial cycle drift caused by changes in coating speed or adhesive viscosity.

[0085] In this embodiment, the texture order sequence is a sequence formed by segmenting the normalized periodic flow mark texture sequence according to the coating roller rotation cycle. Specifically, each segment corresponds to a complete mechanical cycle, and the grayscale change pattern within each mechanical cycle is mapped to an order position, so that each order represents the texture feature synchronized with the coating roller rotation, which makes it easy to distinguish between process textures and abnormal textures.

[0086] In this embodiment, order spectrum analysis is a method for frequency pattern analysis of texture order sequences. Specifically, the gray-level change patterns within each mechanical cycle are arranged along the order position, and the amplitude intensity and phase characteristics of each order are identified by the gray-level repetition patterns within the cycle, so as to quantify the periodicity and regularity of the texture and provide identifiable features for normal flow mark textures and abnormal stripe defects.

[0087] In this embodiment, the normal flow mark texture order is the order position in the order spectrum where the amplitude is continuous and stable and synchronized with the mechanical cycle of the coating roller. Specifically, by comparing the amplitude and phase peaks in multiple consecutive mechanical cycles, the order with continuous amplitude, repetitive cycle and strict synchronization with mechanical movement is selected to represent the process texture features generated by the periodic movement of the coating roller.

[0088] In this embodiment, the candidate stripe defect order is the order position in the order spectrum where the amplitude is unstable or not strictly synchronized with the mechanical cycle. Specifically, it is the order with large amplitude fluctuations, discontinuous phases, or deviations from the coating roller cycle. It is used to mark possible abnormal stripe defect areas and provide a reference for subsequent suppression template construction and defect detection.

[0089] It should be noted that the techniques of mechanical cycle normalization and texture order sequence construction are key steps in achieving analogous rotational machinery vibration signal order tracking analysis in adhesive coating images. Since the images themselves lack direct mechanical references, and the spatial spacing of flow mark textures is unstable with changes in process parameters, mechanical cycle normalization can align the flow mark textures with the coating roll cycle, enabling order analysis to correctly distinguish between process textures and abnormal stripes.

[0090] It should be noted that this step, by constructing a texture order sequence and performing order spectrum analysis, provides a foundation for the design of subsequent preprocessing suppression templates. This enables the normal flow mark texture to be suppressed while retaining abnormal stripe defects during image enhancement, thereby improving the accuracy and stability of defect identification. At the same time, it provides a repeatable and quantifiable mechanical reference standard for the entire defect detection method.

[0091] It should be noted that the significance of this step for the entire method lies in realizing the periodic quantification of flow mark texture, providing a mechanical periodic reference for the subsequent suppression of normal flow mark texture and accurate identification of candidate stripe defects. This solves the problem that the spatial periodic instability of flow mark texture when the coating speed or adhesive viscosity changes makes it difficult to distinguish between process texture and real defects, thereby ensuring the reliability and accuracy of subsequent defect detection.

[0092] In this embodiment, the mapping of grayscale data to the texture order sequence is a process of aligning the pixel position of each frame of the image in the first preprocessing sequence with the corresponding mechanical order position according to the material conveying direction. Specifically, according to the coating roller rotation angle corresponding to the acquisition time of each frame, each row of pixels in the image is mapped to the corresponding order interval, so that the grayscale changes in the image can be expressed in the order dimension as a change pattern consistent with the mechanical cycle, thereby realizing a unified coordinate expression of image space and mechanical cycle.

[0093] In this embodiment, the process texture area is the area marked in the image according to the spatial position corresponding to the normal flow mark texture order. Specifically, in each frame of the image, the grayscale change positions within the normal flow mark texture order range are marked as intervals and a continuous set of regions is formed. This set of regions repeats in multiple mechanical cycles and is used to represent the stable process texture distribution range generated by the periodic movement of the coating roller.

[0094] In this embodiment, the preprocessing suppression template is a set of grayscale modulation rules constructed for the process texture region. Specifically, it sets a suppression coefficient for the grayscale enhancement result in the marked region, compresses its grayscale change amplitude by a fixed ratio, and keeps the original enhancement result unchanged in the unmarked region, thereby forming a grayscale adjustment template with dual constraints of spatial position and order position. This template is executed according to the same rules in each frame of the image to ensure that the suppression process is consistent.

[0095] In this embodiment, the grayscale enhancement amplitude is the degree of change of the original grayscale after histogram equalization in the first preprocessing sequence. Specifically, it compares the grayscale difference of each pixel before and after enhancement, and uses the difference as the enhancement amplitude. The difference is proportionally reduced in the suppression template, thereby weakening the contrast performance of the normal flow mark texture.

[0096] In this embodiment, the second preprocessing sequence is an image sequence after applying a preprocessing suppression template to the first preprocessing sequence. Specifically, in each frame of the image, grayscale enhancement amplitude compression is performed on the process texture area while the non-process area remains unchanged, so that the grayscale fluctuation of normal flow mark texture in the output image is reduced, while the abnormal stripe area still retains its original contrast features, thereby improving the desirability of defects.

[0097] In this embodiment, feature extraction is the process of analyzing the gray-level changes in local regions of each frame of the second preprocessing sequence. Specifically, the image is divided into several regions according to a fixed window, and the mean gray-level, gray-level change amplitude, and gray-level change direction of each region are statistically analyzed to form a region feature vector for subsequent defect determination.

[0098] In this embodiment, the defect determination rule is a set of rules for identifying stripe defects based on regional feature vectors. Specifically, it sets a threshold for grayscale change amplitude, a threshold for continuous length, and a threshold for directional consistency. When a region satisfies the following conditions in multiple adjacent windows: grayscale change amplitude exceeds the threshold, change direction is consistent, and continuous length reaches the set range, the region is determined as a candidate region for stripe defects.

[0099] In this embodiment, the priority analysis of the region corresponding to the order of candidate stripe defects is a strategy of prioritizing feature extraction and judgment of the spatial region corresponding to the order of candidate stripe defects during the detection process. Specifically, when performing defect detection, these regions are preferentially divided into windows and feature calculations are performed, and the judgment threshold is reduced to improve the detection sensitivity of potential defects.

[0100] In this embodiment, continuous frame fusion is a process of spatial position superposition and temporal consistency screening of multi-frame detection results. Specifically, the defect areas detected in adjacent frames are aligned according to the material conveying direction, and the number of times the same spatial position appears in continuous frames is counted. When the number of occurrences reaches a set frame number threshold, the area is identified as a stable defect area; otherwise, it is removed, thereby eliminating false detections caused by random noise.

[0101] In this embodiment, the defect level information is a classification result obtained by comprehensively evaluating the grayscale variation range, spatial length and occurrence frequency of the defect area. Specifically, defects with a large grayscale variation range and a long continuous length are marked as high-level defects, and defects with a small variation range or a short length are marked as low-level defects, so as to facilitate subsequent quality assessment and processing decisions.

[0102] It should be noted that the inhibition coefficients in the above-mentioned inhibition template and the various thresholds in the defect judgment are all obtained through calibration using historical production data. Specifically, in the known normal process sample and known defect sample datasets, the grayscale variation range of normal flow mark texture and the grayscale variation range of defect texture are statistically analyzed, and the boundary interval between the two is selected as the basis for threshold setting. At the same time, the thresholds are fine-tuned through online feedback in actual production to ensure that they are adapted to different adhesive viscosities and coating conditions.

[0103] It should be noted that this step can also be extended to introduce position weight adjustment in the suppression template, that is, to set different suppression intensities for process textures near the coating edge or tension change area, so as to adapt to the influence of material edge effect on flow mark texture, thereby further improving the spatial adaptability of the suppression effect.

[0104] It should be noted that the beneficial effect of this step is that by constructing a suppression template consistent with the mechanical order, normal flow mark textures are weakened in a targeted manner, while the grayscale features of abnormal stripe defects are preserved. Combined with a continuous frame fusion mechanism to eliminate random noise interference, this significantly reduces false detections caused by process textures and improves the accuracy and stability of stripe defect detection.

[0105] S5. While keeping the viscosity of the adhesive constant, change the coating speed and repeat the texture extraction and order analysis. Enumerate and compare the periodic performance of the flow mark texture under different coating speeds, and select the coating speed with the most stable periodic performance as the optimal coating speed corresponding to the viscosity of the adhesive.

[0106] In this embodiment, while keeping the adhesive viscosity constant, the coating speed is varied, and texture extraction and order analysis are repeatedly performed. The periodic performance of flow mark textures at different coating speeds is enumerated and compared, and the coating speed with the most stable periodic performance is selected as the optimal coating speed corresponding to the adhesive viscosity. Specifically: After obtaining the second preprocessing sequence and the corresponding normal flow mark texture order and candidate stripe defect order, keep the current adhesive viscosity unchanged and set several pre-selected coating speeds in sequence. For each pre-selected coating speed, a linear array image sequence of the adhesive surface is acquired, and the first preprocessing and mechanical spatial periodic mapping are performed to extract the periodic flow mark texture sequence. Mechanical periodic normalization is performed on the periodic flow mark texture sequence based on the coating roller rotation speed signal to construct a texture order sequence; At each texture level, the deviation range of the peak position of the normal flow mark texture is statistically analyzed along the material conveying direction within each mechanical cycle, and the maximum and average values ​​of the peak position deviation within the mechanical cycle are calculated. For each pre-selected coating speed, the minimum mean peak position deviation and the maximum deviation not exceeding a set threshold are used as the evaluation criteria to determine whether the flow mark texture cycle performance is stable at that coating speed. The coating speed that meets the conditions and has the smallest deviation is recorded as the optimal coating speed corresponding to the viscosity of the adhesive, and stored in conjunction with the adhesive viscosity parameter.

[0107] In this embodiment, the pre-selected coating speed is a set of coating speed parameters set according to the process capability range while maintaining the current adhesive viscosity. Specifically, it selects the minimum speed, maximum speed and several intermediate speeds allowed by the production line to form a testable speed sequence for analyzing the periodic stability of flow mark textures under different coating speeds.

[0108] In this embodiment, the peak position deviation is the range of difference between the grayscale peak position of the normal flow mark texture in each mechanical cycle and its corresponding position in the reference mechanical cycle along the material conveying direction. Specifically, the main peak points are identified for the grayscale change pattern of each mechanical cycle, and their position offsets between different cycles are recorded to quantify the repeatability and stability of the texture in continuous cycles.

[0109] In this embodiment, a threshold is set as the standard range for judging the periodic stability of flow mark texture. Specifically, it is the maximum allowable range of peak position deviation manually selected based on production experience and process requirements. This is used to screen out texture conditions with good periodic repeatability and minimal impact from changes in coating speed, so as to ensure the reliability of subsequent defect detection.

[0110] In this embodiment, the optimal coating speed is the speed parameter with the smallest average peak position deviation and the largest deviation not exceeding a set threshold among all pre-selected coating speeds. Specifically, by enumerating the texture peak deviation at each speed, the speed with the smallest deviation and stability meeting the threshold is recorded as the optimal process speed corresponding to the adhesive viscosity. This speed is used to adjust the coating speed in actual production to optimize the periodicity of flow mark texture.

[0111] It should be noted that when performing this step, image acquisition, preprocessing, and mechanical cycle mapping need to be repeated to ensure that the flow mark texture cycle analysis results at each pre-selected speed are comparable, and a sufficient number of mechanical cycles should be covered when calculating the peak position deviation to obtain a stable and reliable periodic assessment.

[0112] It should be noted that this step enumerates the periodic behavior of flow mark textures under different coating speeds to establish the corresponding optimal coating speed parameters for each adhesive viscosity, thereby quantifying and optimizing the periodic stability of the process texture and making subsequent order-based flow mark texture suppression and defect identification more accurate.

[0113] It should be noted that the significance of this step for the entire method lies in solving the problem of periodic instability of flow mark texture caused by changes in process parameters. By determining the optimal coating speed for each adhesive viscosity, the flow mark texture can be made to exhibit regularity in mechanical cycles, thereby ensuring that subsequent order analysis and defect detection methods can reliably distinguish between normal process textures and real stripe defects, and improving detection accuracy and stability.

[0114] In this embodiment, the pre-selected coating speed sequence is generated by discretely taking values ​​at fixed step sizes within the upper and lower limits of the speed allowed by the equipment. Specifically, based on the minimum and maximum speeds given by the production line control system, the interval is divided into several speed points at equal intervals, and speed points that exceed the stable operating capacity are removed, thereby forming a discrete speed set for testing, ensuring that each speed is comparable.

[0115] In this embodiment, the reference mechanical cycle is a baseline cycle selected at each pre-selected coating speed. Specifically, the first complete cycle or the cycle with the most stable grayscale change is selected from multiple continuously collected mechanical cycles as the reference cycle, and its corresponding peak position is used as the baseline position for subsequent deviation calculation, so as to unify the comparison standard between different cycles.

[0116] In this embodiment, the extraction of grayscale peak positions is a process of searching for local extrema of the grayscale change curve in each mechanical cycle. Specifically, the grayscale change sequence is scanned along the material conveying direction, and the position where the grayscale changes from rising to falling is identified as the peak point. In each mechanical cycle, several peaks with the largest amplitude are selected as the main peak positions to characterize the main structural features of the flow mark texture.

[0117] In this embodiment, the peak position deviation is calculated by matching the main peak position in each mechanical cycle with the corresponding peak position in the reference mechanical cycle one by one, and calculating the distance difference in the conveying direction. Specifically, the position difference of each pair of matched peaks is recorded, and a deviation set is formed in one mechanical cycle to describe the consistency of texture repetition in that cycle.

[0118] In this embodiment, the peak matching relationship is established by matching the peaks in the order of the mechanical cycle. Specifically, the peaks in the reference cycle are sorted by position, and the peak closest to that position is found in the current cycle for matching. This avoids mismatch due to changes in the number of peaks, thereby ensuring the stability of the deviation calculation.

[0119] In this embodiment, the mean peak position deviation is a statistical measure obtained by averaging all peak deviations over multiple mechanical cycles. Specifically, the deviation values ​​in each cycle are summarized and the overall average value is calculated to reflect the overall repeatability level of flow mark texture at the coating speed.

[0120] In this embodiment, the maximum value of the peak position deviation is the maximum deviation value that occurs in all mechanical cycles and all peaks. Specifically, the maximum value is selected after statistically analyzing all deviation data to constrain local abnormal fluctuations and prevent situations where there is a severe local offset but the average value is small.

[0121] In this embodiment, the threshold is obtained based on historical normal production data. Specifically, multiple sets of data are collected under known stable process conditions, the distribution range of the maximum value of the corresponding peak position deviation is calculated, and the upper limit of this range is selected as the threshold. At the same time, the threshold is appropriately relaxed in combination with the equipment accuracy and detection resolution to form the final upper limit standard of deviation used for judgment.

[0122] In this embodiment, the data consistency control during the enumeration comparison process is a constraint process to ensure that the collected data are comparable under each pre-selected coating speed. Specifically, it maintains the same number of collection frames, the same number of mechanical cycles, and the same collection time window under each speed condition to avoid deviations in statistical results due to different sampling lengths.

[0123] In this embodiment, the selection rule for the optimal coating speed is to select the speed with the smallest average peak position deviation among all pre-selected speeds that meet the condition that the maximum deviation does not exceed the threshold. If multiple speeds meet the same minimum average, the speed with the most stable number of peaks in the corresponding mechanical cycle is selected as the final result to further improve process consistency.

[0124] It should be noted that in this step, the frequency of candidate stripe defect order occurrence at different coating speeds can also be statistically analyzed. When the number of candidate stripe defect order occurrences at a certain speed is significantly reduced, it can be used as an auxiliary criterion and together with the peak deviation index to screen the optimal coating speed, thereby improving the reliability of the screening results.

[0125] It should be noted that this step can also be extended to establish a database of mapping relationships between adhesive viscosity and optimal coating speed. Specifically, the above enumeration process is repeated under different viscosity conditions, and the optimal speed corresponding to each viscosity is recorded and stored. In subsequent production, this mapping relationship can be directly called for quick setting, reducing the need for repeated testing.

[0126] It should be noted that the beneficial effect of this step is that by quantitatively evaluating the peak position shift of flow mark texture within the mechanical cycle under different coating speed conditions, a clear and executable stability judgment standard is constructed, thereby avoiding the use of vague stability descriptions, making the selection of coating speed have a repeatable and verifiable basis, and controlling the periodic consistency of flow mark texture from the source, providing stable and reliable input conditions for subsequent order analysis and defect detection.

[0127] S6. When a candidate stripe defect is detected, the coating speed is controlled to the optimal coating speed, and the corresponding area is re-acquired and a secondary inspection is performed for confirmation.

[0128] In this embodiment, when a candidate stripe defect is detected, the coating speed is controlled to the optimal coating speed, and the corresponding area is re-acquired and a secondary detection is performed for confirmation. Specifically, this involves: Under normal production speed, the surface of the adhesive backing is continuously monitored. After a candidate stripe defect area is found, the start and end positions of the defect area in the material conveying direction and the corresponding image frame number are recorded. The coating roller drive system is controlled to adjust the coating speed to the optimal coating speed corresponding to the viscosity of the adhesive solution selected in the previous step, while keeping the adhesive solution viscosity constant. Under optimal coating speed conditions, a continuous image sequence was re-acquired for candidate defect areas using a line scan camera, while simultaneously acquiring coating roller speed signals and material conveying speed signals, and establishing the correspondence between the acquired frames and mechanical motion parameters. The re-acquired image sequence is subjected to grayscale normalization and histogram equalization to form a secondary preprocessing sequence; A mechanical space mapping was established between the secondary preprocessing sequence and the coating roller speed and material conveying signal. The grayscale change sequence was extracted along the conveying direction, and the periodic flow mark texture sequence was obtained. Mechanical periodic normalization is performed on the periodic flow mark texture sequence to construct a texture order sequence and identify the normal flow mark texture order and candidate stripe defect order. The defect areas that reappear in the secondary inspection are identified as real stripe defects, and their location and defect level information are updated in the online detection result output sequence of adhesive defects.

[0129] In this embodiment, the candidate stripe defect region is a potential abnormal texture region identified in the first defect detection process according to the second preprocessing sequence. Its grayscale change characteristics are different from normal flow mark texture, but it has not yet been fully confirmed by mechanical cycle and order analysis. Specifically, it is the start and end positions and corresponding image frame numbers recorded along the material conveying direction, which are used for subsequent secondary detection to confirm the real defect.

[0130] In this embodiment, the secondary preprocessing sequence is an image sequence obtained by re-acquiring images of candidate stripe defect areas under the optimal coating speed conditions and then performing grayscale normalization and histogram equalization. Specifically, the image grayscale values ​​are mapped to a uniform range and the contrast is enhanced to ensure that flow mark texture and potential defect information can be accurately extracted under repeated acquisition conditions.

[0131] In this embodiment, mechanical spatial mapping is the process of establishing a correspondence between each frame of the image in the secondary preprocessing sequence and the rotation angle of the coating roller and the material conveying speed along the material conveying direction. Specifically, it involves recording the acquisition time and mechanical motion parameters of each frame and mapping the image grayscale data to the corresponding mechanical position for subsequent order analysis and periodic texture extraction.

[0132] In this embodiment, secondary detection confirmation is the process of identifying abnormal textures that repeatedly appear in continuous mechanical cycles and marking them as real stripe defects by repeatedly acquiring images, processing grayscale, extracting periodic flow mark textures and performing order analysis on candidate defect areas under optimal coating speed conditions. Specifically, it involves comparing the consistency of abnormal textures in continuous frames and mechanical cycles in the secondary acquisition sequence and eliminating occasional noise or non-defect textures.

[0133] It should be noted that when performing secondary inspection, the adjustment of the coating speed and the re-acquisition of images should ensure that the viscosity of the adhesive remains constant and covers the full length of the candidate defect area, so that mechanical cycle normalization and order analysis can fully reflect the texture cycle characteristics of the area and ensure the accurate identification of real defects.

[0134] It should be noted that the technical extensions of this step include the ability to perform secondary acquisition and periodic analysis on multiple groups of candidate defect areas in parallel to improve detection efficiency. Furthermore, this process can be repeated across multiple batches of adhesive or under different coating conditions, forming a stable defect confirmation process and thereby improving the robustness of the online detection system.

[0135] It should be noted that this step is significant for the entire method in that it addresses the potential for false detections and missed detections during the initial inspection. By adjusting the coating speed to the optimal value, re-acquiring images, and performing periodic texture analysis, it is possible to effectively distinguish between normal process textures and real stripe defects, thereby ensuring the reliability and accuracy of the adhesive defect detection results and providing a reference for optimizing process parameters.

[0136] Example 2, Figure 2 The present invention provides an online detection system for adhesive defects based on the fusion of image and mechanical signals, including a material acquisition module, a mechanical spatial period establishment module, a texture order tracking module, a preprocessing suppression module, a period stability control module, and a defect confirmation module. The material acquisition module is used to obtain the current viscosity of the adhesive and limit the coating speed, acquire the linear array image sequence of the adhesive surface, and preprocess it through histogram equalization to obtain the first preprocessed sequence. The mechanical space cycle establishment module is used to establish the mechanical space cycle by combining the mechanical signals of the production line, and extract the grayscale change sequence of the conveying direction based on the first preprocessing sequence, and obtain the periodic flow mark texture sequence through spatial autocorrelation. The texture order tracking module is used to perform mechanical periodic normalization on the periodic flow mark texture sequence based on the coating roller speed signal, construct the texture order sequence and perform order spectrum analysis to identify the normal flow mark texture order and the candidate stripe defect order. The preprocessing suppression module is used to construct a preprocessing suppression template based on the normal flow mark texture order, perform order suppression on the preprocessing to obtain a second preprocessing sequence, and perform machine vision adhesive defect detection. The cycle stability control module is used to change the coating speed while keeping the viscosity of the adhesive constant, repeatedly perform texture extraction and order analysis, enumerate and compare the cycle performance of flow mark textures under different coating speeds, and select the coating speed with the most stable cycle performance as the optimal coating speed corresponding to the viscosity of the adhesive. The defect confirmation module is used to control the coating speed to the optimal coating speed when a candidate stripe defect is detected, and to re-acquire images of the corresponding area and perform secondary detection and confirmation.

[0137] An electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform an online detection method for adhesive backing defects based on image and mechanical signal fusion.

[0138] A computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for online detection of adhesive defects based on the fusion of image and mechanical signals.

[0139] In the embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0140] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0141] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0142] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0143] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0144] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0145] In the embodiments provided in this disclosure, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0146] It should be noted that, in this disclosure, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element limited by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0147] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An online detection method for adhesive backing defects based on image and mechanical signal fusion, characterized in that, Includes the following steps: The current viscosity of the adhesive solution is obtained and the coating speed is limited. A linear array image sequence of the adhesive backing surface is acquired and preprocessed by histogram equalization to obtain the first preprocessed sequence. A mechanical spatial cycle is established by combining the mechanical signals of the production line, and a grayscale change sequence in the conveying direction is extracted based on the first preprocessing sequence. A periodic flow mark texture sequence is obtained through spatial autocorrelation. Mechanical periodic normalization is performed on the periodic flow mark texture sequence based on the coating roller speed signal, a texture order sequence is constructed and order spectrum analysis is performed to identify the normal flow mark texture order and the candidate stripe defect order. A preprocessing suppression template is constructed based on the normal flow mark texture order. The preprocessing is subjected to order suppression to obtain a second preprocessing sequence, and machine vision adhesive defect detection is performed. While keeping the viscosity of the adhesive constant, the coating speed was changed, and texture extraction and order analysis were repeatedly performed. The periodic performance of flow mark textures under different coating speeds was enumerated and compared, and the coating speed with the most stable periodic performance was selected as the optimal coating speed corresponding to the viscosity of the adhesive. When a candidate stripe defect is detected, the coating speed is controlled to the optimal coating speed, and the corresponding area is re-acquired and a second inspection is performed for confirmation.

2. The online detection method for adhesive defects based on image and mechanical signal fusion according to claim 1, characterized in that, The process involves obtaining the current adhesive viscosity and limiting the coating speed, acquiring a linear array image sequence of the adhesive surface, and preprocessing it using histogram equalization to obtain the first preprocessed sequence, specifically: The viscosity signal of the current batch of adhesive solution is acquired by collecting the viscosity signal of the adhesive solution through an online rotational viscometer. The viscosity signal is continuously sampled and the average viscosity value of the stable range is extracted to obtain the viscosity parameter of the current adhesive solution. Based on the viscosity parameters of the adhesive, the corresponding coating speed range is matched in the preset coating process parameter table, and the current coating speed of the production line is set within the coating speed range; Under the current coating speed conditions, start the roll-to-roll adhesive coating production line, and use a line scan camera set downstream of the coating section to scan and collect data on the surface of the continuously conveyed adhesive material line by line. Record each scan line according to the material conveying order and construct the original image sequence. The original image sequence is processed frame by frame to perform grayscale conversion, and the RGB pixel values ​​of each frame are mapped to single-channel grayscale values ​​according to the grayscale conversion relationship, and arranged in the acquisition order to form a continuous grayscale sequence. For each frame of the continuous grayscale sequence, the grayscale distribution is statistically analyzed, a grayscale histogram is established according to the grayscale range, and the corresponding cumulative grayscale distribution function is calculated based on the grayscale histogram. The gray-level cumulative distribution function is used to map and transform the gray-level values ​​of each pixel in the corresponding frame, mapping the original gray-level values ​​to a new gray-level range to enhance the gray-level contrast of the image and obtain the corresponding gray-level enhanced frame. The grayscale enhancement frames are rearranged and combined according to the acquisition order of the material conveying direction to form an enhanced image sequence; The enhanced image sequence is output as the first preprocessed sequence.

3. The online detection method for adhesive defects based on image and mechanical signal fusion according to claim 2, characterized in that, The mechanical spatial cycle is established by combining the mechanical signals of the production line, and the grayscale change sequence of the conveying direction is extracted based on the first preprocessing sequence. The periodic flow mark texture sequence is obtained through spatial autocorrelation, specifically as follows: After obtaining the first preprocessing sequence, the rotation angle signal of the coating roller and the material conveying speed signal are collected by the mechanical sensor installed in the coating section. The signals are synchronously sampled to obtain the mechanical motion parameter sequence corresponding to each acquisition frame. Based on the mechanical motion parameter sequence, a spatial mapping is established between the first preprocessing sequence and the mechanical motion according to the material conveying direction. The grayscale enhancement data of each frame is mapped to the corresponding mechanical motion position to form a mechanical reference grayscale sequence. The average gray value is extracted line by line from the mechanical reference gray value sequence along the material conveying direction to generate a gray value change sequence, where each element corresponds to the average gray value information of the material at the mechanical motion reference position. A spatial autocorrelation operation is performed on the grayscale change sequence, and the spatial periodic characteristic sequence of grayscale change is obtained by analyzing the autocorrelation coefficient of grayscale values ​​under different spatial lag distances. Identify the main peaks and corresponding hysteresis intervals in the spatial periodic feature sequence, and map the hysteresis intervals to a periodic flow mark texture sequence on the material surface.

4. The online detection method for adhesive defects based on image and mechanical signal fusion according to claim 3, characterized in that, The process involves mechanically normalizing the periodic flow mark texture sequence based on the coating roller rotation speed signal, constructing a texture order sequence, and performing order spectrum analysis to identify the normal flow mark texture order and candidate stripe defect order. Specifically: The real-time rotation speed signal of the coating roller and the synchronous material conveying speed signal are collected. The signals are continuously sampled and a time correspondence is established with the periodic flow mark texture sequence to achieve mechanical cycle normalization, so that each flow mark texture point corresponds to the rotation angle position of the coating roller. The normalized flow mark texture sequence is segmented according to the coating roller rotation cycle, and each segment of texture data is mapped to the corresponding mechanical order position to form a texture order sequence, where each order represents the grayscale change pattern synchronized with the coating roller rotation cycle. Perform order spectrum analysis on the texture order sequence, and analyze the amplitude spectrum and phase spectrum of each order through Fourier transform to obtain the frequency distribution characteristics of the texture on the mechanical cycle. Peak identification is performed on the amplitude spectrum and phase spectrum. The order that is synchronized with the mechanical cycle of the coating roller and has a continuous and stable amplitude is identified as the normal flow mark texture order. The order that is not at the synchronous peak position or has an unstable amplitude is marked as the candidate stripe defect order. The identified normal flow mark texture order and candidate stripe defect order are output as input sequences for subsequent construction of preprocessing suppression templates and backing adhesive defect detection, which are used to distinguish between process textures and real stripe defects.

5. The online detection method for adhesive defects based on image and mechanical signal fusion according to claim 4, characterized in that, The preprocessing suppression template is constructed based on the normal flow mark texture order, and the preprocessing is subjected to order suppression to obtain a second preprocessing sequence, which is then used for machine vision adhesive defect detection. Specifically: After obtaining the normal flow mark texture order and the candidate stripe defect order, for each frame of the first preprocessing sequence, the grayscale data and texture order sequence are mapped along the material conveying direction, and the grayscale change pattern corresponding to each normal flow mark texture order is marked as the process texture region in the image sequence. Based on the marked process texture region, a preprocessing suppression template is established. By suppressing the gray-level enhancement amplitude of the corresponding region, the gray-level fluctuation of normal flow mark texture in the image is reduced, while the original gray-level features of the unmarked region are preserved, forming a second preprocessing sequence. The second preprocessed sequence is used for defect detection based on machine vision. Feature extraction is performed on each frame of the image, and stripe-like defects on the adhesive surface are identified according to the set defect judgment rules. For regions corresponding to candidate stripe defect orders, texture analysis is performed first. For each frame of detection results, continuous frame fusion is performed, and the defect regions that appear repeatedly in adjacent frames are merged to generate the final defect location sequence and defect level information, which is then output as the online detection result of adhesive defects.

6. The online detection method for adhesive defects based on image and mechanical signal fusion according to claim 5, characterized in that, The process involves changing the coating speed while keeping the adhesive viscosity constant, repeatedly performing texture extraction and order analysis, enumerating and comparing the periodic performance of flow mark textures at different coating speeds, and selecting the coating speed with the most stable periodic performance as the optimal coating speed corresponding to the adhesive viscosity. Specifically: After obtaining the second preprocessing sequence and the corresponding normal flow mark texture order and candidate stripe defect order, keep the current adhesive viscosity unchanged and set several pre-selected coating speeds in sequence. For each pre-selected coating speed, a linear array image sequence of the adhesive surface is acquired, and the first preprocessing and mechanical spatial periodic mapping are performed to extract the periodic flow mark texture sequence. Mechanical periodic normalization is performed on the periodic flow mark texture sequence based on the coating roller rotation speed signal to construct a texture order sequence; At each texture level, the deviation range of the peak position of the normal flow mark texture is statistically analyzed along the material conveying direction within each mechanical cycle, and the maximum and average values ​​of the peak position deviation within the mechanical cycle are calculated. For each pre-selected coating speed, the minimum mean peak position deviation and the maximum deviation not exceeding a set threshold are used as the evaluation criteria to determine whether the flow mark texture cycle performance is stable at that coating speed. The coating speed that meets the conditions and has the smallest deviation is recorded as the optimal coating speed corresponding to the viscosity of the adhesive, and stored in conjunction with the adhesive viscosity parameter.

7. The online detection method for adhesive defects based on image and mechanical signal fusion according to claim 6, characterized in that, When a candidate stripe defect is detected, the coating speed is controlled to the optimal coating speed, and the corresponding area is re-imaged and a secondary detection is performed for confirmation. Specifically: Under normal production speed, the surface of the adhesive backing is continuously monitored. After a candidate stripe defect area is found, the start and end positions of the defect area in the material conveying direction and the corresponding image frame number are recorded. The coating roller drive system is controlled to adjust the coating speed to the optimal coating speed corresponding to the viscosity of the adhesive solution selected in the previous step, while keeping the adhesive solution viscosity constant. Under optimal coating speed conditions, a continuous image sequence was re-acquired for candidate defect areas using a line scan camera, while simultaneously acquiring coating roller speed signals and material conveying speed signals, and establishing the correspondence between the acquired frames and mechanical motion parameters. The re-acquired image sequence is subjected to grayscale normalization and histogram equalization to form a secondary preprocessing sequence; A mechanical space mapping was established between the secondary preprocessing sequence and the coating roller speed and material conveying signal. The grayscale change sequence was extracted along the conveying direction, and the periodic flow mark texture sequence was obtained. Mechanical periodic normalization is performed on the periodic flow mark texture sequence to construct a texture order sequence and identify the normal flow mark texture order and candidate stripe defect order. The defect areas that reappear in the secondary inspection are identified as real stripe defects, and their location and defect level information are updated in the online detection result output sequence of adhesive defects.

8. A system for online detection of adhesive defects based on image and mechanical signal fusion as described in any one of claims 1-7, characterized in that, It includes a material acquisition module, a mechanical space period establishment module, a texture order tracking module, a preprocessing suppression module, a period stability control module, and a defect confirmation module; The material acquisition module is used to obtain the current viscosity of the adhesive and limit the coating speed, acquire the linear array image sequence of the adhesive surface, and preprocess it through histogram equalization to obtain the first preprocessed sequence. The mechanical space cycle establishment module is used to establish the mechanical space cycle by combining the mechanical signals of the production line, and extract the grayscale change sequence of the conveying direction based on the first preprocessing sequence, and obtain the periodic flow mark texture sequence through spatial autocorrelation. The texture order tracking module is used to perform mechanical periodic normalization on the periodic flow mark texture sequence based on the coating roller speed signal, construct the texture order sequence and perform order spectrum analysis to identify the normal flow mark texture order and the candidate stripe defect order. The preprocessing suppression module is used to construct a preprocessing suppression template based on the normal flow mark texture order, perform order suppression on the preprocessing to obtain a second preprocessing sequence, and perform machine vision adhesive defect detection. The cycle stability control module is used to change the coating speed while keeping the viscosity of the adhesive constant, repeatedly perform texture extraction and order analysis, enumerate and compare the cycle performance of flow mark textures under different coating speeds, and select the coating speed with the most stable cycle performance as the optimal coating speed corresponding to the viscosity of the adhesive. The defect confirmation module is used to control the coating speed to the optimal coating speed when a candidate stripe defect is detected, and to re-acquire images of the corresponding area and perform secondary detection and confirmation.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the online detection method for adhesive defects based on image and mechanical signal fusion as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the online detection method for adhesive defects based on the fusion of image and mechanical signals as described in any one of claims 1 to 7.

Images