Lithium ion battery pole piece coating defect online detection method based on machine vision

Abstract

The application discloses a kind of lithium ion battery pole piece coating defect on-line detection method based on machine vision, it is related to lithium ion battery manufacturing quality detection technical field.For the problem of missing detection due to insufficient contrast of conventional dry film optical detection of small metal foreign matter in high-speed coating production line, piezoelectric acoustic emission sensor array is arranged in pairs on the coating die lip plate, background vibration is suppressed by differential processing, foreign matter acoustic emission pulse is captured and acoustic characteristic is extracted to predict foreign matter parameters;Based on the predicted parameters, the expected template of the interference fringes in the downstream wet film area is calculated using the thin film interference physical model;Acoustic emission event triggers downstream camera to collect interference image at predicted position and extract features;Match the expected template with the actual observation, confirm and classify foreign matter through acoustic-optical matching score, and evaluate the embedded depth.The method increases the detection sensitivity of metal foreign matter to 40 μm level when the coating speed is above 80 m / min, and can identify fully embedded foreign matter that cannot be detected by pure optical solution.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery manufacturing quality inspection technology, specifically to an online detection method for lithium-ion battery electrode coating defects based on machine vision. Background Technology

[0002] Lithium-ion battery electrode coating is a core process in power battery manufacturing, and its quality directly affects the battery's electrochemical performance and safety. During the coating process, metal debris from equipment wear, metal particles from environmental dust, or residual metal impurities from raw materials may mix into the slurry or adhere to the electrode surface in the form of aluminum powder, copper powder, or iron powder. These metal foreign particles, typically ranging in size from twenty to several hundred micrometers, can pierce the separator during charging and discharging due to volume expansion or lithium dendrite growth, forming an internal short circuit. This can lead to thermal runaway or even fire and explosion, posing the most serious safety hazard to power batteries. Industry statistics show that internal short-circuit failures caused by metal foreign objects account for the majority of on-site safety accidents involving power batteries. Therefore, achieving highly sensitive online detection of metal foreign objects during the coating process is a crucial step in ensuring the safety of power batteries.

[0003] Currently, machine vision-based online inspection technology has been widely used in lithium-ion battery electrode coating production lines. A typical approach involves deploying a high-resolution line scan camera in the dry film section after coating and drying, along with a bright-field or dark-field illumination system, to continuously scan and image the electrode coating surface. Image processing algorithms or deep learning-based defect recognition models are then used to detect various defects on the coating surface. This approach has achieved good results in detecting macroscopic defects with obvious morphological or grayscale characteristics, such as missed coatings, uneven edges, and surface scratches, and has become a standard configuration for quality inspection in lithium-ion battery coating production lines. However, for the detection of tiny metallic foreign objects, existing dry film section machine vision inspection solutions face a fundamental technical obstacle: after the coating slurry has dried and cured in the oven, metallic foreign particles mixed in the coating are partially or completely covered by the dried active material coating. Their surface optical reflection characteristics tend to be consistent with the surrounding dried coating, and the grayscale contrast between the two under broadband visible light illumination drops to below five percent. Under high-speed production line conditions, the line exposure time of a line scan camera is typically only a few microseconds. The signal-to-noise ratio of a single line acquisition is limited, making such subtle grayscale differences easily masked by noise. Actual engineering data shows that existing visual inspection solutions for dry film sections typically only achieve a detection sensitivity of 80 to 100 micrometers for metallic foreign objects, which is insufficient to meet the detection requirements of power battery manufacturing for metallic foreign objects smaller than 50 micrometers. Especially for high-density foreign objects such as copper powder and iron powder, because their density is much higher than that of the coating slurry, they quickly settle into the coating during the coating process, generating almost no detectable optical signal on the dry film surface. This creates a systemic safety blind spot that pure visual inspection solutions cannot cover in principle. Summary of the Invention

[0004] Technical Objective: To address the technical problem in existing technologies where tiny metallic foreign objects are missed in high-speed coating production lines for lithium-ion battery electrodes due to insufficient optical contrast on the dry film surface and high-density foreign object deposition embedded within the coating, this invention discloses an online detection method for coating defects of lithium-ion battery electrodes based on machine vision, used for online detection of metallic foreign object defects during the high-speed coating process of lithium-ion battery electrodes.

[0005] Technical solution: To achieve the above technical objectives, the present invention adopts the following technical solution: A machine vision-based online detection method for coating defects in lithium-ion battery electrodes includes the following steps: S1. Acoustic emission signals during the coating process are synchronously collected by a pair of piezoelectric acoustic emission sensors on the outer side of the lip plate of the coating die head. The signals of the pair of sensors are differentially processed to suppress the common mode component of the background vibration of the production line and obtain differential acoustic emission signals. S2. Perform short-time energy detection on the differential acoustic emission signal. When the local energy exceeds the adaptive threshold, it is determined to be an acoustic emission event. Extract the acoustic features of the acoustic emission event and predict the physical parameters of the metal foreign object based on the acoustic features to obtain the foreign object parameter prediction vector. The foreign object parameter prediction vector includes material type, size estimate, lateral position and estimated time to reach the optical detection area. S3. Based on the predicted vector of the foreign object parameters, the expected characteristic parameters of the interference fringes that the metal foreign object should produce in the wet film region are calculated in the forward direction using the thin film interference physical model, and an expected interference fringe template is generated. The expected interference fringe template includes the expected fringe radial velocity, the expected fringe ring number, the expected fringe contrast and the spatial search window. S4. In the wet film exposure section downstream of the coating die head, an image stream of the wet film surface is continuously acquired using a high frame rate area array camera with narrow band light source illumination; when the pre-trigger signal of the acoustic emission event generated in step S2 is received, the frame difference sequence is calculated for the corresponding position of the continuous multiple frames of images within the spatial search window, and the radially symmetric dynamic response mode is searched in the frame difference sequence to extract the interference feature observation vector, which includes the radial velocity of the observed fringe, the number of observed fringe rings, and the contrast of the observed fringe; S5. Match and compare the expected template of the interference fringes with the observed vector of the interference features to calculate the acoustic-optic matching score; classify and determine the metallic foreign object according to the detection status of the acoustic emission event and the numerical range of the acoustic-optic matching score and output the detection result.

[0006] Beneficial Effects: The online detection method for coating defects of lithium-ion battery electrodes based on machine vision provided by this invention has the following beneficial effects: First, the present invention moves the detection time window forward to the exposed section when the coating is still in a wet film state. It uses the dynamic thin film interference fringes generated by the local accelerated evaporation of solvent due to the density and thermal conductivity differences of the metal foreign object in the wet film as a detection feature. The contrast of this dynamic optical feature is much higher than the static grayscale difference on the dry film surface, which fundamentally solves the problem of insufficient contrast in dry film detection.

[0007] Secondly, this invention captures the impact elastic wave signal of a foreign object passing through the lip at the coating die head using an acoustic emission sensor, achieving the earliest detection of metallic foreign objects. Based on acoustic characteristics, it predicts the parameters of the foreign object and uses a thin-film interference physical model to generate a matching template for downstream optical detection. This transforms optical detection from a blind search mode without prior information to a template matching mode guided by a physical model, resulting in a substantial improvement in detection sensitivity.

[0008] Third, this invention achieves online quantitative assessment of the embedding depth of metallic foreign objects in a wet film for the first time by exploiting the systematic deviation between the expected value of the interference characteristics predicted by acoustic emission and the actual optical observation value. It can distinguish between three foreign object states: surface type, semi-embedded type, and fully embedded type. Among them, the fully embedded type is the type of defect that cannot be detected by pure optical methods but has the highest safety risk.

[0009] Fourth, the acoustic emission pre-triggering mechanism of the present invention enables the optical system to perform high-cost interferometric analysis processing on a small area only when the acoustic emission event is triggered, which greatly reduces the real-time computing load and allows the method to meet the real-time requirements of high-speed production lines on industrial embedded computing platforms. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0011] Figure 1 This is a schematic diagram of the overall process of the method of the present invention; Figure 2 This is a schematic diagram showing the placement of the acoustic emission sensing module on the coating die head. Figure 3 A schematic diagram illustrating the principle of differential noise reduction for acoustic emission signals; Figure 4 Flowchart for acoustic emission event detection and foreign object parameter prediction; Figure 5 This is a flowchart of the forward calculation process for the thin-film interference physical model. Figure 6 This is a schematic diagram of the optical path layout of the wet film dynamic interferometry imaging module. Figure 7 This is a schematic diagram of the radial expansion dynamic mode of interference fringes from a metallic foreign object in a frame difference sequence. Figure 8 A schematic diagram illustrating the four scenarios for sound-optic matching fusion decision-making; Figure 9 This is a schematic diagram illustrating the principle of reverse estimation of the embedding depth of metallic foreign objects. Detailed Implementation

[0012] The present invention will now be described more clearly and completely by way of a preferred embodiment in conjunction with the accompanying drawings, but this does not limit the invention to the scope of the described embodiment.

[0013] See Figure 1The present invention provides an online detection method for coating defects of lithium-ion battery electrodes based on machine vision. The overall technical route is as follows: at the lip plate of the coating die, an acoustic emission sensor array captures the impact elastic wave signal generated when a metal foreign object passes through the lip gap, and extracts the physical parameters such as the material, size and position of the foreign object; using these parameters, a thin film interference physical model is used to forward calculate the dynamic interference fringe feature template that the foreign object should generate in the downstream wet film area; the acoustic emission event triggers a high frame rate area array camera in the downstream wet film area to perform directional interference image acquisition and feature extraction within a local window at the predicted position; the expected template predicted by the physical model is matched and compared with the actual optical observation, and the metal foreign object is identified, classified and quantitatively evaluated in terms of embedding depth by using the acoustic-optic matching degree score.

[0014] The reason this invention chooses the aforementioned acoustic-optical fusion technology route instead of relying solely on optical detection is that: when a metallic foreign object passes through the lip of the coating die, it undergoes a mechanical collision with the metal surface of the lip plate. This physical event leaves characteristic imprints in both the acoustic and optical domains. The characteristic imprint in the acoustic domain is the elastic wave pulse generated by the collision, whose spectral and temporal characteristics are determined by the material (density, elastic modulus) and size of the foreign object. The characteristic imprint in the optical domain is the dynamic thin-film interference fringes generated by the foreign object in the wet film due to the enhanced solvent evaporation driven by the difference in thermal conductivity; its spatiotemporal characteristics are also determined by the material and size of the foreign object. The characteristics of the two domains establish a deterministic mapping relationship through the physical properties of the same physical entity. Utilizing this cross-domain physical coupling relationship, acoustic information can provide precise prior constraints for optical detection, thereby transforming the detection problem from statistically significant anomaly detection into a deterministic template matching problem. According to the matched filter theory in signal processing, under additive white Gaussian noise conditions, a matched filter with a known signal waveform is the optimal detector in terms of signal-to-noise ratio (SNR). Its detection SNR is equal to the input SNR multiplied by the square root of the number of signal samples, which is far superior to an energy detector without prior information. This theory guarantees the essential advantage of the acoustic-optical fusion scheme in detection sensitivity compared to the pure optical scheme.

[0015] Furthermore, acousto-optic fusion produces an emergent capability that a single-modal approach cannot achieve—foreign object embedding depth assessment: the acoustic emission signal is generated at the moment of impact when the foreign object passes through the lip, before it has settled; the acoustic characteristics reflect the complete initial physical properties of the foreign object. Optical observation occurs in the wet film exposure section, with a time lag of approximately 0.4 to 1.0 seconds, during which time the foreign object may have partially or completely settled. The deviation between the acoustically predicted expected interference characteristics and the actual optical observation precisely corresponds to the reduction in the foreign object's protrusion height during the settling process. Pure acoustic approaches can only know that a foreign object has passed through the die but not its final position within the coating; pure optical approaches are completely ineffective for settled foreign objects. Only the acousto-optic fusion approach can infer the embedding state of the foreign object by leveraging the information discrepancy between the two domains.

[0016] Acquisition and differential denoising of acoustic emission signals (step S1): See Figure 2 and Figure 3 The goal of this step is to establish acoustic sensing capabilities for metal foreign object collision events at the coating die head, while effectively suppressing background vibration noise on the production line.

[0017] A row of piezoelectric emission sensors is installed on the outer sides of both the upper and lower lip plates of the coating die, forming a paired array. The piezoelectric emission sensors are made of PZT-5A piezoelectric ceramic, characterized by high sensitivity, wide bandwidth, and good temperature resistance. The frequency response range of the piezoelectric emission sensors is selected from 100kHz to 1MHz. The physical basis for selecting this frequency range is as follows: When a metallic foreign particle collides with the surface of a metal lip plate, the collision duration is determined by Hertzian contact theory. According to Hertzian theory, the contact duration of a collision between two elastic spheres is proportional to the square of the equivalent mass of the colliding particles, the negative fifth power of the collision velocity, and the negative fifth power of the equivalent elastic modulus. For a typical lithium battery coating scenario, with a coating speed of 60 to 120 m / min (i.e., 1.0 to 2.0 m / s) and a metallic foreign particle size of 20 to 200 micrometers, the theoretically calculated collision duration is 0.5 to 10 microseconds, corresponding to a dominant frequency range of 100 kHz to 2 MHz for the elastic waves excited by the collision. Considering the acoustic wave propagation attenuation characteristics of the lip plate material and the severe background noise pollution of the production line below 100 kHz, 100 kHz to 1 MHz is the optimal frequency band for balancing signal capture and noise suppression.

[0018] Piezoelectric acoustic emission sensors are uniformly arranged along the coating transversely with element spacing of 10mm to 20mm, covering the full width of the module. The selection of element spacing considers two factors: excessive spacing reduces the lateral positioning accuracy of the beamforming algorithm and results in an excessively large search window for downstream optical detection; insufficient spacing leads to an excessive number of sensors, increased wiring complexity and cost, and potential acoustic crosstalk between adjacent sensors. Within the 10mm to 20mm spacing range, based on a typical acoustic emission frequency of 400kHz, the sound wave propagation time difference between adjacent elements is 2 to 4 microseconds (assuming a sound velocity of 5000m / s in the lip plate). At a 5MHz sampling rate, a time difference resolution of 10 to 20 sampling points can be obtained, which is sufficient to support the beamforming algorithm to achieve a lateral positioning accuracy on the order of 5mm.

[0019] Each piezoelectric acoustic emission sensor on the upper lip plate and its opposite piezoelectric acoustic emission sensor on the lower lip plate form a paired sensor pair. This paired arrangement is based on common-mode and differential-mode analysis: background vibration sources during production line operation include roller rotation, tension regulating roller oscillation, fluid pulsation from the slurry supply pump, and mechanical vibrations transmitted from upstream and downstream equipment via the foundation. When these vibrations are transmitted to the upper and lower lip plates through the die head housing, due to the structural symmetry of the housing, the vibration signals generated on the upper and lower lip plates are highly similar in amplitude and phase, belonging to the common-mode component. However, the collision of a metallic foreign object through the lip gap occurs within the extremely narrow gap between the upper and lower lip plates. The propagation paths of the collision sound source to the upper and lower lip plates are different, resulting in significant differences in amplitude and phase of the signals reaching the sensors on both sides, belonging to the differential-mode component.

[0020] The specific method of differential processing is as follows: For each pair of upper and lower piezoelectric emission sensors, the signals from the upper and lower piezoelectric emission sensors are obtained respectively. The differential signal is equal to the upper piezoelectric emission sensor signal minus the product of the common-mode rejection coefficient k and the lower piezoelectric emission sensor signal, as expressed in the following formula: in It is a differential signal. The raw acoustic emission signal collected by the piezoelectric acoustic emission sensor on the upper lip plate. The signal is the raw acoustic emission signal collected by the piezoelectric acoustic emission sensor on the lower lip plate, where k is the common-mode rejection coefficient and t is the time variable.

[0021] The common-mode suppression coefficient k is determined as follows: Under normal coating production conditions with no foreign objects passing through, the signals from the upper and lower sensors are continuously collected for several seconds each (10 seconds in this embodiment). The ratio of the root mean square amplitude of the upper sensor signal to the root mean square amplitude of the lower sensor signal is calculated, which is the common-mode suppression coefficient k. The reason for introducing coefficient k instead of directly performing equal amplitude differential is that the geometry of the upper and lower lip plates and the sensor installation conditions may not be completely symmetrical, resulting in a systematic deviation in the signal amplitude generated by the same vibration source on both sides. The role of k is to compensate for this amplitude asymmetry, maximizing the common-mode suppression effect after differential. This calibration process can be automatically executed each time the production line starts or changes models, and the calibration takes about 10 seconds, without affecting the production line cycle time.

[0022] After differential processing, common-mode noise caused by background vibration of the production line is suppressed. In a typical coating die head structure, the common-mode suppression ratio can reach 20 to 30 dB.

[0023] Acoustic emission event detection and foreign object physical parameter prediction (step S2): See Figure 4The goal of this step is to detect foreign object collision events in real time from differential acoustic emission signals and extract physical parameter information such as the material, size, and location of the foreign object from the waveform characteristics of the acoustic emission pulse.

[0024] Short-time energy detection is performed on the differential acoustic emission signal to identify foreign object collision events. A short-time detection sliding window is set (window length is set to 10 microseconds to 100 microseconds, 50 microseconds is used in this embodiment), and the instantaneous energy value of the differential signal within the window (i.e., the integral value of the square of the signal within the window) is calculated. At the same time, a long-term background energy statistics window is maintained (window length is set to 0.5 seconds to 2 seconds, 1 second is used in this embodiment), and the mean and standard deviation of the background energy are continuously calculated. When the instantaneous energy exceeds the mean plus a certain number of times the standard deviation, an acoustic emission event is determined to have occurred, and the timestamp of the event is recorded.

[0025] The reason for choosing an adaptive threshold instead of a fixed threshold is that the background vibration level in the coating production line is not constant, but drifts slowly with factors such as coating speed adjustment, slurry flow rate changes, and roller wear. The adaptive threshold automatically adjusts the detection threshold by continuously tracking changes in background statistics to maintain a constant false alarm probability. This strategy is known as constant false alarm rate (CFRR) detection in signal processing and is a classic method for detecting weak signals against non-stationary noise backgrounds.

[0026] The threshold factor is selected in the range of 4 to 8 standard deviations: 4 standard deviations are suitable for scenarios with extremely high sensitivity requirements, while 8 standard deviations are suitable for scenarios with strict requirements on false alarm rate. In this embodiment, 6 standard deviations is preferred.

[0027] For the detected acoustic emission events, the differential signal segment centered on the event timestamp is used as the analysis window to extract the following three acoustic features: (a) Centroid Frequency of the Spectrum: After applying a Hanning window to the truncated acoustic emission pulse signal, perform a Fast Fourier Transform to calculate its power spectral density. Then, calculate the first moment of the power spectral density as the centroid frequency of the spectrum. The formula is as follows: in The centroid frequency of the spectrum, For the spectrum variable, Let be the power spectral density function of the acoustic emission pulse. This is the lower limit of the effective frequency band. This represents the upper limit of the effective frequency band. There is a definite physical correspondence between the centroid frequency of the spectrum and the material of the foreign object. The fundamental reason is that different metals have different densities and elastic moduli, leading to different collision durations. The density of aluminum is 2.7 g / cm³. 3The elastic modulus is 70 GPa, the collision duration is relatively long, and the spectral centroid is low, typically ranging from 200 to 350 kHz; the density of copper is 8.9 g / cm³. 3 The elastic modulus is 110 GPa, the collision duration is moderate, and the spectral centroid is in the range of 450 to 700 kHz; the density of iron is 7.9 g / cm³. 3 The elastic modulus is 200 GPa. The highest elastic modulus results in the shortest collision duration and the highest spectral centroid, which is in the range of 600 to 850 kHz.

[0028] (ii) Pulse Duration: The envelope of the acoustic emission pulse signal is calculated (the magnitude of the analytic signal is obtained through Hilbert transform). The pulse duration is defined as the duration during which the envelope amplitude exceeds 20% of its peak value. Pulse duration is positively correlated with foreign object size: larger foreign objects have a larger contact area and longer contact time with the lip surface. The equivalent spherical diameter of the foreign object can be approximately estimated as the product of the coating speed and the pulse duration, multiplied by a correction factor related to the foreign object shape. The correction factor for spherical particles is approximately 1.2, and for sheet-like particles, it is approximately 0.6. The formula is expressed as follows: , This is an estimate of the equivalent sphere diameter of the metallic foreign object. For coating speed, The duration of the pulse. This is the foreign object shape correction factor. The correction factor is determined through prior calibration experiments using standard metal particles of known size.

[0029] (iii) Pulse Energy: Calculate the square integral of the acoustic emission pulse within the analysis window. The pulse energy is proportional to the collision kinetic energy of the foreign object and can be used as an auxiliary verification parameter for size estimation.

[0030] The lateral position of a metallic foreign object passing through the lip gap is determined using a beamforming algorithm based on an acoustic emission sensor array. Specifically, a delay-summation beamforming method is employed: assuming the impact sound source is located at a certain lateral position in the lip gap, the sound wave propagates to each sensor at the longitudinal wave velocity in the lip material. For candidate lateral positions, the signals from each sensor are summed after delay compensation according to the theoretical arrival time, resulting in coherent superposition of the signals from that position. The range of lateral position values ​​is scanned, and the position with the largest superimposed signal amplitude is the estimated lateral position of the foreign object. In this embodiment, the standard deviation of the lateral positioning is approximately 3 to 7 mm.

[0031] Based on the above extraction and calculation results, the acoustic emission module outputs a foreign object parameter prediction vector, which includes four components: material type, size estimate, lateral position, and estimated arrival time to the optical detection area. The estimated arrival time is equal to the acoustic emission event timestamp plus the distance between the acoustic emission sensing area and the optical detection area divided by the coating speed.

[0032] Forward calculation of the thin-film interference physical model (step S3): See Figure 5 This step is the core innovative link in achieving deep fusion of sound and light. Based on the material and size information in the foreign object parameter prediction vector output in step S2, the characteristic parameters of the interference fringes that the specific foreign object should produce in the wet film region are calculated in a forward manner using the physical laws of thin film interference.

[0033] There is a wet film exposure section, typically 0.3 to 1.5 meters long, between the coating die lip and the oven inlet. In this section, the coating surface is covered with a thin film of solvent that has not yet fully evaporated. When metallic foreign particles are present in the wet film, the following two physical effects occur: The first effect is the geometric protrusion effect. The density of the metallic foreign matter is much higher than that of the slurry. Under the equilibrium of buoyancy, gravity, and surface tension, the foreign matter particles form local protrusions on the wet film surface. The protruding areas cause a change in curvature of the solvent surface around the foreign matter, producing thin film interference fringes of equal thickness under incident light illumination.

[0034] The second effect is the thermal conductivity-enhanced evaporation effect. The thermal conductivity of metallic materials is much higher than that of the active material particles in the coating (copper has a thermal conductivity of approximately 400 W / (m·K), aluminum approximately 237 W / (m·K), iron approximately 80 W / (m·K), and lithium iron phosphate approximately 1.5 W / (m·K)). The metallic foreign particles, acting as locally high thermal conductivity bodies, accelerate heat conduction in their surrounding area, resulting in a higher solvent evaporation rate near the foreign particles compared to areas farther away. This causes the solvent film thickness around the foreign particles to decrease faster than in the surrounding area, thus exhibiting a dynamic outward radial expansion of the equal-thickness interference fringes.

[0035] This dynamic interference fringe pattern is a characteristic optical signal unique to metallic foreign objects. Although the agglomeration of active material particles in normal coatings may also produce weak surface protrusions and interference fringes, because the difference in thermal conductivity between the active material and the surrounding coating material is minimal, there is no significant evaporation enhancement effect. The dynamic change rate of the interference fringes is only one-fifth to one-half that of metallic foreign objects, and the fringe contrast is low.

[0036] The metallic foreign object is approximated as a spherical particle, partially submerged in the wet film slurry, reaching equilibrium under the influence of gravity, buoyancy, and surface tension. The protrusion height Δh is determined by the following equilibrium relationship: the protrusion height is approximately equal to the foreign object diameter d multiplied by (1 minus the slurry density divided by the foreign object density) multiplied by (1 minus the slurry surface tension divided by the foreign object density multiplied by the gravitational acceleration multiplied by the square of d), expressed by the following formula: ,in For the density of the coating slurry, Density of metallic foreign matter The surface tension coefficient of the coating slurry. This represents gravitational acceleration. The physical meaning of this formula is: the first term in parentheses reflects the degree of buoyancy imbalance driven by density difference, and the second term in parentheses reflects the restraining effect of surface tension. Taking copper foreign matter in lithium iron phosphate cathode slurry as an example: the theoretical protrusion height of a 50-micrometer diameter copper powder foreign matter is approximately 12 micrometers, and the protrusion height of a 100-micrometer diameter copper powder foreign matter is approximately 35 micrometers.

[0037] Given the protrusion height, illumination source wavelength, and wet film solvent refractive index, the radial propagation velocity of the interference fringes is calculated using the principle of thin film equal-thickness interference.

[0038] The basic condition for thin-film equal-thickness interference is that two beams of light reflected from the upper and lower surfaces of the solvent film interfere, and bright fringes appear when the optical path difference satisfies the condition of an integer multiple of the wavelength. Each equal-thickness interference fringe corresponds to a specific solvent film thickness value. When the solvent around the metallic foreign object evaporates more rapidly due to the enhanced thermal conductivity effect, the local film thickness continues to decrease, and the fringes are equivalent to being pushed outward from the center.

[0039] There is a deterministic physical relationship between the radial expansion rate of the stripes and the solvent evaporation rate. This relationship can be calculated analytically based on the evaporation rate, wavelength, refractive index, and geometry of the convex region. The formula is expressed as follows: in The desired velocity for the outward radial expansion of the interference fringes. The evaporation rate of the solvent surrounding the metallic foreign object. The wavelength of the lighting source, The solvent refractive index, The gradient of the solvent film thickness along the radial direction is determined by the geometry of the protruding region.

[0040] The solvent evaporation enhancement effect around the metallic foreign object is very significant: taking copper foreign objects in lithium iron phosphate coatings as an example, the ratio of the thermal conductivity of copper to that of lithium iron phosphate is about 267. According to the thermal resistance model of heat transfer, the solvent evaporation rate around the copper foreign object can reach 25 to 80 times that of the normal area. This huge difference is the fundamental reason for the high-speed radial expansion interference fringes generated by the metallic foreign object.

[0041] The number of interference fringe rings is determined by the range of film thickness variation in the protruding region and the illumination wavelength, as expressed by the following formula: ,in Theoretical ring number of observable equal-thickness interference fringes around the foreign object. To achieve a uniform solvent film thickness away from the foreign matter area, The thickness is the thinnest solvent film at the top of the foreign object. The fringe contrast is determined by the relationship between the reflectivity of the foreign object surface and the reflectivity of the wet film surface according to the two-beam interference visibility formula.

[0042] The center position of the search window is determined by the lateral position in the foreign object parameter prediction vector and the estimated arrival time. The size of the search window is determined according to the principle of 3 times the standard deviation of the positioning uncertainty, ensuring that the probability of the actual position of the foreign object falling within the search window is greater than 99.7%. The half-width in the lateral direction is approximately 15mm, and the half-width in the coating direction is approximately 6mm, as expressed by the following formula: , ,in This is half the width of the search window in the horizontal direction of the coating. The standard deviation of the lateral positioning in the beamforming algorithm. The search window is half the width in the coating direction. The standard deviation of longitudinal position uncertainty is determined by both the accuracy and speed fluctuations of the encoder on the production line.

[0043] Based on the above calculations, step S3 outputs the desired interference fringe template, which includes the desired fringe radial velocity, the desired fringe ring number, the desired fringe contrast, and the spatial search window parameters.

[0044] Directional interferometric image acquisition and feature extraction (step S4): See Figure 6 and Figure 7 In this step, an optical imaging module is deployed in the middle of the wet film exposure section downstream of the coating die head to perform directional interferometric image acquisition and feature extraction pre-triggered by acoustic emission events.

[0045] The center wavelength of the narrowband LED light source is selected to be 540 to 560 nm, with a bandwidth of ±10 nm. This wavelength range was chosen based on three factors: First, NMP solvent films exhibit weak refractive index dispersion in the visible light range, and narrowband illumination near the center wavelength reduces blurring of higher-order interference fringes caused by dispersion; second, 540 to 560 nm falls within the high-sensitivity band of silicon-based CMOS sensors, and commercially available narrowband LEDs and narrowband filters are readily available in this band; third, for typical wet film thicknesses, the interference order corresponding to this wavelength provides sufficient fringe density for spatial resolution.

[0046] The light source illuminates the wet film surface at a grazing incidence angle of 60 to 75 degrees. Grazing incidence increases the effective optical path length of the light in the thin film, resulting in a larger change in optical path difference for the same film thickness, thus improving the sensitivity to slight differences in film thickness. The incidence angle should not exceed 75 degrees; otherwise, the surface Fresnel reflectivity will increase sharply, leading to insufficient transmitted light.

[0047] The high frame rate area array CMOS camera has a frame rate of no less than 500 frames per second. The reason for choosing an area array camera instead of a line scan camera is that this method requires acquiring multiple consecutive frames of images in the same spatial region to observe the temporal evolution of interference fringes; an area array camera can acquire the instantaneous grayscale distribution of all pixels within a search window in a single frame. The camera is equipped with a narrowband filter to suppress stray light from the production line environment.

[0048] The optical module is positioned 0.3 to 0.8 meters downstream of the coating die lip. If the position is too close, the interference fringes have not yet fully developed; if the position is too far, the interference fringes may have already faded. The 0.3 to 0.8 meter position corresponds to the middle stage of wet film solvent evaporation.

[0049] When the acoustic emission module generates an acoustic emission event, the pre-trigger signal and spatial search window parameters are transmitted to the optical processing module. Based on the expected arrival time, the optical processing module locates the corresponding frame range in the circular buffer and crops the image region corresponding to the search window.

[0050] A frame difference sequence calculation is performed on 20 to 50 consecutive cropped images: each frame of the frame difference sequence is the pixel-by-pixel difference between two adjacent grayscale images. The frame difference operation eliminates static background components while preserving inter-frame variation information.

[0051] Searching for characteristic dynamic interference patterns generated by metallic foreign objects in frame difference sequences: concentric ring-shaped gray-level extreme value distribution centered on the location of the foreign object (spatial feature), and the radial position of the extreme value ring moving outward frame by frame over time (temporal feature).

[0052] The specific search method is as follows: within the search window, candidate center positions are searched by sliding the search with the predicted horizontal position of acoustic emission as the center. For each candidate position, the radial gray-level distribution of the frame difference image is calculated to check whether periodic alternating positive and negative extreme values ​​are present. Furthermore, it is checked whether the radius of the extreme value ring in the continuous frame difference image monotonically increases with the frame number. For candidate positions that satisfy the above characteristics, the interference feature observation vectors are extracted: the radial velocity of the observed fringes (radial displacement velocity of the extreme value ring in the continuous frame difference image), the number of observed fringe rings (the number of resolvable extreme value rings), and the contrast of the observed fringe (the ratio of the gray-level amplitude of the extreme value ring to the average gray-level of the background).

[0053] When no acoustic emission event pre-trigger signal is received, the optical module operates in basic detection mode, calculating the global frame difference energy between adjacent frames only for the full-frame image. When the global frame difference energy exceeds a preset global threshold, a complete interference feature extraction process is performed on the abnormal region. The basic detection mode ensures that the optical system maintains baseline detection capability even if the acoustic emission module misses a detection.

[0054] High-cost interference analysis in pre-triggered mode is performed only within the search window (approximately 9% of the total area), and the acoustic emission event triggering frequency is typically 0.1 to 1 times per second. Therefore, the average computational load of high-cost analysis is extremely low, which can meet real-time constraints on industrial embedded GPU platforms.

[0055] Acousto-optic matching and fusion decision (step S5): See Figure 8 and Figure 9 This step involves quantitatively matching and comparing the expected template of interference fringes generated in the acoustic domain with the interference features observed in the optical domain.

[0056] The acoustic-optic matching score (with a value greater than 0 and not exceeding 1) is calculated as follows: the first component is the square of the difference between the radial velocity of the desired fringe and the radial velocity of the observed fringe, divided by the square of the radial velocity uncertainty; the second component is the square of the difference between the number of rings of the desired fringe and the number of rings of the observed fringe, divided by the square of the ring number uncertainty; and the third component is the square of the difference between the contrast of the desired fringe and the contrast of the observed fringe, divided by the square of the contrast uncertainty. These three components are multiplied by weighting coefficients and then summed. The negative exponential function value of this weighted sum is then taken. The specific formula is expressed as follows: in The sound-light matching score. The radial velocity of the fringes as observed in actual optical measurements. The desired radial velocity of the fringes is calculated in the forward direction of the physical model. The uncertainty of the radial velocity of the fringes was determined through calibration experiments. This refers to the number of fringe rings observed in actual optical observations. The expected number of fringe rings calculated for the physical model. The uncertainty is the number of fringe rings. For the fringe contrast observed in actual optical observation, Stripe contrast calculated for the physical model The uncertainty of the stripe contrast. , , These are the weighting coefficients for the three feature components.

[0057] The score equals 1 when the prediction and observation are perfectly aligned; as the difference increases, the score decays exponentially towards 0. The choice of the exponential function form is based on the optimality principle of the Gaussian likelihood ratio test. All uncertainty parameters are determined through calibration experiments. The radial velocity component is given the highest weight among the weighting coefficients because it has the strongest ability to distinguish between metallic foreign objects and normal textures.

[0058] Classification and determination of four situations: Scenario 1: When an acoustic emission event has been detected and the acoustic-optic matching score is greater than or equal to the first threshold, it is determined to be a surface-type metallic foreign object, i.e. ,in For acoustic emission events, The threshold value is set at 1. Acoustic confirmation of the presence of the foreign object and the high consistency between the optical interference characteristics and the physical model prediction indicate that the foreign object protrudes from the wet film surface.

[0059] Scenario 2: When an acoustic emission event has been detected and the acoustic-optic matching score is greater than or equal to the second threshold and less than the first threshold, it is determined to be a semi-embedded metallic foreign object, i.e. ,in This is the second threshold. Acoustic confirmation of the foreign object's presence, coupled with weaker-than-expected optical interference characteristics, indicates that the foreign object has partially settled into the wet film. This type of foreign object poses a higher safety risk than surface-type foreign objects.

[0060] Scenario 3: When an acoustic emission event has been detected and the acoustic-optic matching score is less than the second threshold, it is determined to be a fully embedded metallic foreign object, i.e. Acoustic detection clearly showed the foreign object passing through the die lip, but optical detection showed almost no interference characteristics, indicating that the foreign object had completely settled into the wet film. This is a unique detection capability of the acousto-optic fusion solution—pure acoustic solutions cannot confirm the final state of the foreign object in the coating, and pure optical solutions are theoretically unable to detect completely settled foreign objects. Only the acousto-optic fusion solution can infer that the foreign object is completely embedded through a systematically low matching score. Fully embedded foreign objects pose the highest safety risk.

[0061] Scenario 4: When an acoustic emission event is not detected, the optical system operates independently according to the basic detection mode.

[0062] The first threshold is greater than the second threshold. In this embodiment, the first threshold is set to 0.7 and the second threshold is set to 0.3.

[0063] When the foreign object is identified as semi-embedded or fully embedded, the settling depth of the foreign object in the wet film is inferred from the ratio of the expected radial velocity to the observed radial velocity. The settling depth is equal to the theoretical protrusion height multiplied by 1 minus the absolute value of the difference between the observed radial velocity and the expected radial velocity, expressed by the following formula: in This represents the settling depth of the metallic foreign object in the wet film. The theoretical protrusion height calculated in step S3 is the protrusion height that the foreign object should have when it does not sink at all.

[0064] The physical basis of this assessment method is as follows: the acoustic emission signal is generated at the moment of impact when the foreign object passes through the lip of the mold head, at which point the foreign object has not yet settled. The theoretical protrusion height calculated in step S3 based on the complete initial physical properties represents the protrusion height that the foreign object should have if it did not settle at all. The optical observation occurs in the downstream wet film exposure section (time lag of approximately 0.4 to 1.0 seconds), during which Stokes settling driven by the density difference of the foreign object has already occurred. The deviation between the two precisely corresponds to the amount of reduction in protrusion height during the settling process.

[0065] Embedding depth information can be used as a quantitative basis for classifying the risk level of foreign objects: fully embedded foreign objects are marked as the highest risk level and are recommended to be discarded; the risk level of semi-embedded foreign objects is determined according to the ratio of settlement depth to total coating thickness.

[0066] During long-term use of the coating die, the acoustic emission signal characteristics of the lip surface slowly drift due to wear. This invention provides an acousto-optic closed-loop self-calibration mechanism: when step S5 determines that a surface-type metallic foreign object exists and the acousto-optic matching score is higher than a preset calibration threshold, this detection event is considered a high-confidence acousto-optic paired sample. The actual parameters of the foreign object obtained by optical interferometry can be used as calibration benchmark data for the acoustic model. The system combines the acoustic features with the optically derived actual parameters of the foreign object to form paired data, and updates the mapping model parameters from acoustic features to the physical parameters of the foreign object online.

[0067] The reason for using only highly matched surface-type foreign object data for calibration is that only the optical interference characteristics of surface-type foreign objects best match the assumptions of the physical model, and the foreign object parameters derived from their optical inversion have the highest accuracy, which can serve as a reliable calibration benchmark for the acoustic model.

[0068] Example The following is an example of the application of the method of the present invention on an actual lithium-ion battery cathode coating production line.

[0069] Basic production line parameters: coating speed 80m / min, positive electrode active material is lithium iron phosphate, slurry solid content 55%, slurry viscosity 3500mPa·s, slurry density 1.72g / cm³ 3 The surface tension of the slurry is 0.03 N / m, the refractive index of the NMP solvent is 1.47, the thickness of the wet film coated on one side is 120 micrometers, the width of the aluminum foil current collector is 620 mm, the length of the exposed section of the wet film is 0.6 m, and the ambient temperature of the production line is 28 degrees Celsius.

[0070] Acoustic emission module configuration: Sensor model: PZT-5A wideband piezoelectric acoustic emission sensor, sensitivity greater than or equal to 60dB, frequency response range 100 kHz to 1 MHz. 32 sensors are mounted on each of the upper and lower lip plates, with an element spacing of 18mm. Preamplifier gain 40dB, data acquisition card sampling rate 5MHz per channel, quantization accuracy 12 bits. Common-mode rejection ratio k is calibrated to 0.93. Adaptive threshold multiplier set to 6.

[0071] Optical module configuration: Narrowband LED light source: center wavelength 550nm, bandwidth ±10nm, power 5W, grazing incidence angle 65 degrees. Area array CMOS camera: resolution 2048 x 256 pixels, frame rate 1000 frames per second, pixel size 5.5 micrometers, lateral spatial resolution approximately 30 micrometers per pixel. Narrowband filter: 550nm ±10nm. Center distance of optical module from the module lip: 500mm. Annular buffer capacity: 300 frames.

[0072] Acousto-optic matching decision parameters: weighting coefficients: It equals 0.5. It equals 0.3. It equals 0.2. The uncertainty parameter is determined through calibration experiments. The first threshold is 0.7, the second threshold is 0.3, and the calibration threshold is 0.85.

[0073] Test methods and test results: Detection rate tests were conducted using standard spherical metal particle samples of known materials and sizes. The test particles included aluminum powder, copper powder, and iron powder, with sizes of 30 micrometers, 50 micrometers, 80 micrometers, 100 micrometers, and 150 micrometers. Twenty particles of each material and size were used. A conventional visual inspection system using a dry-film segmental scan camera was used as a comparison.

[0074] The test results are as follows: For aluminum powder foreign matter: the detection rate of this method is 100% for 150-micron and 100-micron sizes, while the conventional method is 100% and 95% respectively; for 80-micron sizes, the detection rate is 100% for this method, while the conventional method is 85%; for 50-micron sizes, the detection rate is 95% for this method, while the conventional method is 65%; for 30-micron sizes, the detection rate is 80% for this method, while the conventional method is 15%.

[0075] For copper powder foreign matter: 150 micrometers, this method achieves 100% accuracy, while conventional methods achieve 95%; 100 micrometers, this method achieves 100% accuracy, while conventional methods achieve 70%; 80 micrometers, this method achieves 100% accuracy, while conventional methods achieve 45%; 50 micrometers, this method achieves 95% accuracy, while conventional methods achieve 20%; 30 micrometers, this method achieves 85% accuracy, while conventional methods achieve 5%.

[0076] For iron powder foreign matter: 150 micrometers, this method achieves 100% accuracy, while conventional methods achieve 95%; 100 micrometers, this method achieves 100% accuracy, while conventional methods achieve 65%; 80 micrometers, this method achieves 100% accuracy, while conventional methods achieve 40%; 50 micrometers, this method achieves 95% accuracy, while conventional methods achieve 15%; 30 micrometers, this method achieves 85% accuracy, while conventional methods achieve 0%.

[0077] Test results show that this method achieves a 100% detection rate for foreign objects of various materials larger than 80 micrometers; for foreign objects in the 50-micrometer range, the detection rate is over 95%, while conventional methods only achieve 15% to 65%; for foreign objects in the 30-micrometer range, this method still achieves 80% to 85%, while conventional methods reduce the detection rate of iron powder to zero. The detection rates of copper powder and iron powder in conventional methods are significantly lower than those of aluminum powder, verifying the technical analysis that high-density foreign objects cause extremely low optical contrast in the dry film due to sedimentation and embedding.

[0078] Regarding the identification of embedding type, among 20 copper powder foreign objects of 50 micrometers, this method identified 8 as semi-embedded and 12 as fully embedded; among 20 iron powder foreign objects of 50 micrometers, 11 were identified as semi-embedded and 9 as fully embedded. None of the foreign objects identified as fully embedded were detected in the conventional method.

[0079] Regarding the false alarm rate, this method achieved a false alarm rate of 0.3% during a continuous 48-hour test, compared to 1.2% for the conventional approach. The low false alarm rate is attributed to the dual acoustic-optical verification mechanism—acoustically triggered false alarm signals are filtered out during the optical matching step.

[0080] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A machine vision-based online detection method for coating defects in lithium-ion battery electrodes, characterized in that, Specifically, the following steps are included: S1. Acoustic emission signals during the coating process are synchronously collected by a pair of piezoelectric acoustic emission sensors on the outer side of the lip plate of the coating die head. The signals of the pair of sensors are differentially processed to suppress the common mode component of the background vibration of the production line and obtain differential acoustic emission signals. S2. Perform short-time energy detection on the differential acoustic emission signal. When the local energy exceeds the adaptive threshold, it is determined to be an acoustic emission event. Extract the acoustic features of the acoustic emission event and predict the physical parameters of the metallic foreign object based on the acoustic features to obtain the foreign object parameter prediction vector. The foreign object parameter prediction vector includes the material type, size estimate, lateral position, and estimated time to reach the optical detection area. S3. Based on the foreign object parameter prediction vector, the expected characteristic parameters of the interference fringes that the metal foreign object should produce in the wet film region are calculated in the forward direction using the thin film interference physical model, and an expected interference fringe template is generated. The expected interference fringe template includes the expected fringe radial velocity, the expected fringe ring number, the expected fringe contrast and the spatial search window. S4. In the wet film exposure section downstream of the coating die head, an image stream of the wet film surface is continuously acquired using a high frame rate area array camera with narrow band light source illumination; when the pre-trigger signal of the acoustic emission event generated in step S2 is received, the frame difference sequence is calculated for the consecutive multiple frames of images at the corresponding position within the spatial search window, and the radially symmetric dynamic response mode is searched in the frame difference sequence to extract the interference feature observation vector, which includes the radial velocity of the observed fringe, the number of observed fringe rings, and the contrast of the observed fringe; S5. Match and compare the expected template of the interference fringes with the observed vector of the interference features, and calculate the acousto-optic matching score; classify and determine the metallic foreign objects according to the detection status of the acoustic emission event and the numerical range of the acousto-optic matching score, and output the detection results. In step S1, a row of piezoelectric emission sensors is installed on the outer side of the upper lip plate and the outer side of the lower lip plate of the coating die head to form an upper and lower paired array. The array element positions of the upper row of piezoelectric emission sensors and the lower row of piezoelectric emission sensors correspond one-to-one along the coating transverse direction. The frequency response range of the piezoelectric emission sensors is 100kHz to 1MHz, and the array element spacing of the sensors along the coating transverse direction is 10mm to 20mm. The specific method of differential processing is as follows: a differential signal is calculated for the signals of each pair of upper and lower paired piezoelectric emission sensors. The differential signal is equal to the upper piezoelectric emission sensor signal minus the common mode rejection coefficient multiplied by the lower piezoelectric emission sensor signal. The common mode rejection coefficient is determined by statistically analyzing the ratio of the root mean square amplitude of the upper and lower piezoelectric emission sensor signals under normal production conditions without foreign matter. In step S3, the specific method for forward calculation of the desired fringe radial velocity using the thin-film interference physical model includes: Based on the size estimate and density value corresponding to the material type in the foreign object parameter prediction vector, the protrusion height of the metal foreign object on the wet film surface is calculated through the buoyancy-surface tension balance relationship. Based on the protrusion height, illumination source wavelength, and refractive index of the wet film solvent, the radial velocity of the interference fringes due to local film thinning caused by solvent evaporation is calculated according to the thin film equal thickness interference condition, and the desired fringe radial velocity is obtained. The center position of the spatial search window is determined based on the lateral position in the foreign object parameter prediction vector and the estimated time to reach the optical detection area. The half-width of the spatial search window in the lateral direction is three times the lateral positioning standard deviation of the beamforming algorithm, and the half-width in the coating direction is determined based on three times the position uncertainty corresponding to the production line encoder accuracy and speed fluctuation.

2. The online detection method for coating defects of lithium-ion battery electrodes based on machine vision according to claim 1, characterized in that, The adaptive threshold is set to 4 to 8 times the standard deviation of the mean background energy of the differential acoustic emission signal within the sliding time window; the acoustic features include the spectral centroid frequency, pulse duration, and pulse energy; wherein, the spectral centroid frequency is obtained by calculating the first moment of the power spectral density of the acoustic emission pulse, and the pulse duration is defined as the duration during which the acoustic emission envelope amplitude exceeds 20% of the peak value.

3. The online detection method for coating defects of lithium-ion battery electrodes based on machine vision according to claim 2, characterized in that, In step S2, the specific methods for predicting the physical parameters of metallic foreign objects based on acoustic features include: The material type of a metallic foreign object is determined based on its spectral centroid frequency. Foreign objects with a spectral centroid frequency in the range of 200kHz to 350kHz are identified as aluminum, those in the range of 450kHz to 700kHz are identified as copper, and those in the range of 600kHz to 850kHz are identified as iron. The size of the metallic foreign object is estimated based on the pulse duration and coating speed. The foreign object size is equal to the product of the coating speed and the pulse duration, multiplied by a correction factor related to the shape of the foreign object. The lateral position of a metallic foreign object is determined using a beamforming algorithm for an acoustic emission sensor array.

4. The online detection method for coating defects of lithium-ion battery electrodes based on machine vision according to claim 1, characterized in that, In step S4, the narrowband light source is a narrowband LED light source with a center wavelength of 540nm to 560nm, which illuminates the wet film surface at a grazing incidence angle of 60 degrees to 75 degrees; the frame rate of the high frame rate area scan camera is not less than 500 frames per second, and the high frame rate area scan camera is equipped with a narrowband filter that matches the center wavelength of the narrowband light source; the number of frames in the continuous multi-frame image is 20 to 50 frames.

5. The online detection method for coating defects of lithium-ion battery electrodes based on machine vision according to claim 1, characterized in that, In step S5, the acoustic-optic matching score is calculated as follows: the first component is the square of the difference between the radial velocity of the expected fringe and the radial velocity of the observed fringe divided by the square of the radial velocity uncertainty; the second component is the square of the difference between the number of rings of the expected fringe and the number of rings of the observed fringe divided by the square of the ring number uncertainty; and the third component is the square of the difference between the contrast of the expected fringe and the contrast of the observed fringe divided by the square of the contrast uncertainty. The three components are then weighted and summed, and the negative exponential function value is taken to obtain the acoustic-optic matching score.

6. The online detection method for coating defects of lithium-ion battery electrodes based on machine vision according to claim 5, characterized in that, In step S5, based on the detection status of the acoustic emission event and the numerical range of the acoustic-optic matching score, the classification and determination are performed according to the following logic: When an acoustic emission event has been detected and the acoustic-optic matching score is greater than or equal to the first threshold, it is determined to be a surface-type metallic foreign object. When an acoustic emission event has been detected and the acoustic-optic matching score is greater than or equal to the second threshold and less than the first threshold, it is determined to be a semi-embedded metallic foreign object. When an acoustic emission event has been detected and the acoustic-optic matching score is less than the second threshold, it is determined to be a fully embedded metallic foreign object; When an acoustic emission event is not detected, the high frame rate area array camera performs an independent optical detection mode to make the determination. Wherein, the first threshold is greater than the second threshold; When a metal foreign object is determined to be semi-embedded or fully embedded, the settling depth of the metal foreign object in the wet film is inferred from the ratio of the expected stripe radial velocity to the observed stripe radial velocity. The settling depth is equal to the theoretical protrusion height multiplied by the absolute value of the difference between the ratio of the observed stripe radial velocity to the expected stripe radial velocity and 1.

7. The online detection method for coating defects of lithium-ion battery electrodes based on machine vision according to claim 1, characterized in that, It also includes an acousto-optic self-calibration step: when the determination result of step S5 is a surface-type metallic foreign object and the acousto-optic matching score is higher than the preset calibration threshold, the acoustic features in this detected event and the actual parameters of the foreign object obtained by optical interferometry are combined to form paired data, and the mapping model from acoustic features to the physical parameters of the foreign object in step S2 is updated online using the paired data.

8. The online detection method for coating defects of lithium-ion battery electrodes based on machine vision according to claim 1, characterized in that, In step S4, when no pre-trigger signal for an acoustic emission event is received, the high frame rate area array camera operates in basic detection mode. In basic detection mode, frame difference energy detection is performed on the full-frame image. When the full-frame frame difference energy exceeds the global threshold, complete interference feature extraction and matching decision processing are performed on the over-threshold area.

Images