Battery gluing joint overflow anti-reflection optical sensing detection method

Abstract

The application discloses a battery gluing joint overflow anti-reflection optical sensing detection method, adopts a double-mode adaptive light path to adapt to different battery surface detection conditions, sets a three-stage time-sharing multi-spectrum excitation structure to avoid colloid pre-solidification, and combines a double-wavelength differential compensation mode to eliminate substrate errors. The application realizes nanometer to millimeter full-range quantitative detection through a sectional optical model, is matched with a multi-parameter closed-loop compensation and an automatic calibration mechanism, and introduces a solidification kinetics model to realize gluing quality pre-judgment. The application effectively reduces detection missing and false detection rates, significantly improves detection precision and mass production stability, has strong working condition adaptability, and has high industrial application and quality control values.

Description

Technical Field

[0001] This invention belongs to the field of optical sensing and detection technology, specifically a method for detecting anti-reflective optical sensing of glue overflow at battery assembly seams. Background Technology

[0002] In the industrial production of new energy lithium batteries, sealing cell seams and applying adhesive to the sides are core processes for ensuring battery sealing and insulation, preventing electrolyte leakage, and preventing structural failure. The quality of the adhesive application directly determines the cycle life, safety performance, and mass production yield of the lithium battery. Defects such as excess adhesive, insufficient adhesive, uneven adhesive layer thickness, and poor adhesive curing can easily lead to battery encapsulation failure, localized overheating, leakage and bulging, or even thermal runaway safety accidents. Therefore, high-precision, high-stability, and fully automated online inspection of the battery bonding seam area is an indispensable core link in the quality control of intelligent lithium battery manufacturing.

[0003] Currently, mainstream industry testing methods have gradually replaced manual sampling, and machine vision optical imaging detection technology is widely used to achieve automated identification of adhesive coating defects. Among them, the existing authorized patent most similar to the technical solution of this invention is CN115661040B, entitled "A Method and Device for Detecting Adhesive Coating of Lithium Battery Cells, Electronic Equipment". This patent discloses a conventional visual inspection solution for lithium battery adhesive coating. Its core technical idea is: by acquiring images of the adhesive coating area of ​​the cell, combining image grayscale variance, region segmentation and circumscribed contour operators to identify the adhesive coating area, thereby realizing the determination of conventional appearance defects such as excess adhesive and missing adhesive. It is a representative basic technical solution in the current field of visual inspection of lithium battery adhesive coating, and is widely used as a general industry testing benchmark solution for application and reference.

[0004] However, existing conventional visual inspection technologies, represented by CN115661040B, still suffer from many inherent technical defects and insurmountable industry technical bottlenecks in actual mass production applications. Specific defects are as follows: First, existing technologies have limited optical path adaptability and cannot handle complex surface conditions with varying degrees of reflectivity. CN115661040B only uses a single conventional visible light imaging optical path and a general image thresholding algorithm, without any specific anti-interference design for the characteristics of the two main types of lithium battery casings. Hard-shell batteries have strong specular reflection characteristics in their metal casings, which easily leads to overexposure and signal saturation; soft-pack batteries have messy aluminum-plastic film surface textures and severe diffuse reflection, resulting in strong stray light interference. This leads to a very high rate of missed detections and false detections in mixed-use testing scenarios, making it impossible to simultaneously meet the mass production testing needs of all types of lithium batteries.

[0005] Second, existing technologies lack a substrate error compensation mechanism, resulting in poor consistency in batch testing. The surface of the battery's metal casing naturally generates an aluminum oxide layer of randomly discrete thickness. Fluctuations in oxide layer thickness directly alter the optical reflection phase and reflectivity, introducing systematic testing bias. CN115661040B completely ignores substrate errors caused by oxide layer discreteness, assuming all battery substrates have uniform optical characteristics. It cannot correct for testing deviations caused by substrate differences, resulting in a batch judgment deviation as high as 18.6% in mass production, failing to meet high-precision consistency quality control requirements.

[0006] Third, existing technologies can only identify qualitative defects and cannot complete full-range quantitative detection. The comparative patent can only determine the presence of appearance defects such as excess glue or missing glue by the difference in image morphology. It has not built an optical physical detection model and does not have the ability to quantitatively solve the glue layer thickness. At the same time, the detection range is narrow and cannot be adapted to the full gradient of working conditions from nanoscale ultrathin insulating film to millimeter-level accumulated glue overflow. It has serious range gaps and accuracy shortcomings.

[0007] Fourth, existing technologies use a synchronous supplemental lighting mode, resulting in poor data accuracy. Traditional visual inspection solutions, including this comparative patent, all employ a multi-source synchronous superposition imaging method. During the inspection process, the continuous accumulation of light and heat can cause incompletely cured sealant to flow prematurely or undergo local pre-curing. This results in the collected images and thickness data failing to reflect the original coating state of the sealant, leading to a long-standing quality control blind spot of "passing inspection but failing batches after curing."

[0008] Fifth, existing technologies only have instantaneous static inspection capabilities and lack proactive risk prediction functions. Conventional visual inspection technologies, represented by CN115661040B, can only perform static screening of the adhesive layer's appearance at the current moment. They cannot capture the dynamic changes during the adhesive curing process, nor can they predict potential quality risks such as adhesive curing shrinkage, internal stress concentration, and subsequent adhesive bursting or delamination. They can only achieve post-event defect detection and cannot achieve pre-event quality risk control, resulting in a low level of quality control.

[0009] Sixth, existing technologies have weak anti-interference capabilities and insufficient long-term mass production stability. Existing technologies lack systematic compensation and automatic calibration mechanisms for temperature, vibration, light source drift, and mechanical deformation. Under complex mass production conditions in lithium battery workshops, including wide temperature fluctuations, continuous production line vibration, dust and oil contamination, and light source aging, detection accuracy drifts severely. Accuracy degradation can reach 28.3% after 24 hours of continuous operation, requiring frequent manual shutdowns for calibration, and exhibiting extremely poor adaptability to large-scale continuous mass production.

[0010] In summary, current lithium battery coating inspection technologies, represented by CN115661040B, generally suffer from a series of technical pain points, such as poor adaptability to operating conditions, inability to eliminate substrate errors, low inspection accuracy, data distortion, lack of curing prediction capability, and poor mass production stability. These technologies are insufficient to meet the current high-precision, high-consistency, all-condition, predictable, and highly stable mass production quality control requirements of high-end lithium batteries. Summary of the Invention

[0011] The purpose of this invention is to provide an optical sensing detection method for anti-reflective properties of glue overflow at battery bonding seams. This invention effectively overcomes the shortcomings of traditional bonding detection methods, such as poor adaptability, large errors, data distortion, lack of predictive ability, and weak mass production stability, through integrated innovation of dual-optical-path adaptive adaptation, time-division spectral fidelity acquisition, dynamic compensation for substrate errors, full-range segmented detection and pre-curing prediction, and multi-dimensional working condition self-calibration. It significantly improves the accuracy, anti-interference ability, and consistency of battery glue overflow detection in mass production.

[0012] The technical solution adopted in this invention is as follows: An optical sensing method for detecting adhesive overflow at battery assembly seams to reduce glare is proposed. This method employs a dual-optical-path adaptive switching sensing architecture, combined with multi-spectral time-division acquisition, pre-emptive impurity shielding, segmented optical modeling, and multi-condition dynamic compensation strategies. It achieves anti-glare, full-range, and high-precision online detection of adhesive overflow at battery assembly seams for different types of batteries. The method includes the following steps: Identify the type of battery to be tested, adaptively match the corresponding anti-reflective optical detection path, and complete the pre-positioning lock; A time-sharing, staggered optical excitation method is adopted to complete impurity fluorescence detection, metal substrate oxide layer detection, and main optical signal acquisition of the adhesive layer in stages. Real-time detection of oxide layer parameters on metal surfaces, and dynamic phase compensation of optical detection signals based on a preset calibration database; Collect multi-temporal optical sensing data of the adhesive layer, and combine it with working condition sensing information to complete signal correction and noise suppression; The real-time thickness of the adhesive layer is obtained by calling the corresponding optical physical model based on the thickness range of the adhesive layer and solving it piecewise. By combining multi-time series detection data with a colloid curing kinetic model, the final overflow state after curing is extrapolated and predicted, thus completing defect identification and production line tracing and diversion.

[0013] Preferably, two types of anti-reflective detection optical paths are configured to adapt to different battery surface characteristics. For the smooth metal surface of the hard-shell battery, a 42° oblique incidence suppressed total internal reflection optical path is adopted. The 42° incident angle is slightly larger than the critical angle of total internal reflection at the air interface of the BK7 optical glass, and s-polarized light is used as the unified incident reference to stably excite the evanescent wave and thin film interference signal. For the rough aluminum-plastic film surface of the soft-pack battery, an 85° grazing incidence polarization differential optical path is adopted. The polarization and detection combination structure suppresses diffuse reflection stray light and specular reflection interference to extract the effective adhesive layer sensing signal.

[0014] Preferably, the time-division staggered optical excitation includes three independent timing modes, with no overlapping light sources in each mode, to avoid detection deviations caused by premature curing of the colloid due to heat. In the first time sequence, the fluorescent light source is turned on, and oil stains and dust impurities are identified based on the three-dimensional features of fluorescence intensity, spectral shift and decay rate. Impurity mask areas are generated and shielded at the pixel level. In the second time sequence, the fluorescent light source is turned off, the dual-path ultraviolet light source is turned on, and the reflectivity data of the metal substrate is collected for the calculation of the oxide layer thickness. In the third time sequence, the ultraviolet light source is turned off, and multiple visible and near-infrared main detection light sources are turned on to collect optical signals of adhesive layer interference and reflection.

[0015] Preferably, the thickness of the alumina oxide layer on the metal surface is calculated in real time using the dual-wavelength ultraviolet differential method, and the calculation formula is as follows: ; In the formula: The thickness of the aluminum oxide layer on the battery metal surface; The extinction coefficient of the alumina medium at a UV wavelength of 275 nm; The extinction coefficient of the alumina medium at a UV wavelength of 255 nm; The reflectance at a wavelength of 255nm is collected in real time for the area to be tested; The standard reflectance of a pure metal reference surface without an oxide layer at a wavelength of 255 nm; The reflectance at a wavelength of 275nm is collected in real time for the area to be tested; The standard reflectance of a pure metal reference surface without an oxide layer is measured at a wavelength of 275 nm. Based on the measured oxide layer thickness, the wavelength-oxide layer thickness-phase shift correspondence is retrieved, and independent phase compensation is performed on each main detection light source to eliminate the optical detection system error caused by the discreteness of the oxide layer on the metal substrate.

[0016] Preferably, for the 0.5nm~100nm ultrathin adhesive layer range, a modified FTIR suppressed total internal reflection model is used to solve for the thickness; a metal region of the same material, without adhesive or oxide layer differences in the vicinity of the joint to be tested is selected as the reference area, and the thickness is calculated using a normalized ratio formula. ; In the formula: The normalized effective signal strength of suppressed total internal reflection; The real-time FTIR reflected signal intensity of the adhesive layer area to be tested at the battery assembly seam; The FTIR reference signal intensity is obtained for the glue-free and same material reference metal region within the same field of view; through ratio normalization calculation, common mode cancels air gap fluctuations, light intensity jitter, and stray light interference, realizing stable detection of nanoscale ultrathin adhesive layers.

[0017] Preferably, for medium-thickness adhesive layers ranging from 100nm to 5mm, a four-layer dielectric thin film transmission matrix model is used to achieve full-range detection; the refraction angle of the colloidal medium is solved using Snell's law of refraction.

[0018] ; In the formula: The refractive index of air is fixed at 1.0. The angle of incidence of light; The equivalent refractive index of the battery encapsulating colloid; Let be the angle of refraction of light after it is incident on the colloidal medium. After solving for the angle of refraction of the colloidal medium, a four-layer medium transmission matrix of air-colloid-oxide-metal is constructed by combining the oblique incident optical path difference correction and oxide layer phase compensation. An overdetermined set of equations is constructed using the multi-wavelength reflectivity response, and the optimal thickness value is solved by the least squares iterative algorithm.

[0019] Preferably, it is equipped with a multi-dimensional composite vibration suppression mechanism; it distinguishes between horizontal and vertical vibrations through multi-directional MEMS sensors, uses image pixel offset correction for horizontal vibrations, and uses laser ranging closed-loop real-time compensation for optical path air gaps for vertical vibrations; it uses phase-locked synchronous sampling to suppress steady-state periodic vibrations of the production line, and uses multi-frame optimization and bad frame rejection strategies for instantaneous impact vibrations to ensure that there are no dropped frames and no missed detections in high-speed production line detection.

[0020] Preferably, it has a full-temperature adaptive temperature compensation mechanism; based on multi-point calibration at 0℃, 15℃, 30℃, and 45℃, it obtains segmented temperature compensation curves, collects real-time temperature parameters of optical cavity, prism, light source, and environment, and dynamically corrects optical sensing errors caused by colloidal refractive index, prism thermal deformation, and laser wavelength drift, adapting to complex temperature change conditions in industrial applications from 0 to 45℃.

[0021] Preferably, it is equipped with a full-range automatic calibration mechanism; the equipment automatically completes multi-gradient standard sample calibration at 0nm, 1nm, 10nm, 100nm, 1μm and 1mm every day when it is powered on, covering the full detection range of nanometer, micrometer and millimeter, and automatically iteratively updates the optical reference and algorithm coefficients to ensure the consistency and stability of the equipment's detection accuracy in the long term.

[0022] A high-precision detection device for excess glue at the joint of battery assembly includes a frame support module, a dual-mode independent optical path switching module, a multi-band time-division light source excitation module, an anti-cross-light zone optical imaging module, a multi-parameter closed-loop working condition compensation module, a main control computing module, and an explosion-proof and corrosion-proof safety protection module. The rack-mount module is internally physically isolated to form an optical cavity, an electrical cavity, a main control cavity, and a cleaning cavity; The dual-mode independent optical path switching module integrates the 42° suppressed total internal reflection optical path of the hard-shell battery and the 85° polarized grazing incidence anti-reflection optical path of the soft-pack battery, which can realize battery type prediction-based pre-switching positioning. The multi-band time-sharing light source excitation module is equipped with multiple independent light sources including fluorescence, ultraviolet, and visible and near-infrared, and has independent temperature control and time-sharing peak emission control logic. The anti-crosstalk partitioned optical imaging module uses an integrated BK7 partitioned prism with light-shielding and light-blocking ribs to achieve nano-FTIR detection and interference angle detection in partitioned sections. The multi-parameter closed-loop operating condition compensation module integrates vibration sensing, temperature sensing, and laser ranging closed-loop adjustment structure to achieve vibration correction, temperature compensation, and dynamic stability control of air gap. The explosion-proof and corrosion-resistant safety protection module integrates cavity explosion-proof sealing, laser opening interlock, and external interface corrosion-resistant sealing structure, meeting the explosion-proof safety regulations for lithium battery workshops.

[0023] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: Traditional technologies generally treat the discrete deviations in the thickness of the natural oxide layer of the metal casing as an inherent process error that cannot be eliminated, making batch consistency testing impossible. This invention innovatively introduces a dual-wavelength deep ultraviolet differential mechanism to accurately calculate the actual thickness of the oxide layer on each battery substrate in real time, and performs dynamic optical phase compensation for each of the multi-band detection light sources, significantly eliminating system detection offsets caused by random fluctuations in the oxide layer. This reduces the substrate judgment deviation of traditional technologies from 18.6% to less than 1.2%, breaking the long-standing industry prejudice that "substrate errors cannot be eradicated," achieving absolute uniformity in the testing benchmark for all battery samples, and providing a fundamental guarantee for ultra-high precision batch testing that conventional technologies cannot achieve.

[0024] Conventional optical inspection methods in the industry typically employ multi-source synchronous supplemental lighting. The accumulation of photothermal energy can cause uncured colloids to flow prematurely and undergo localized pre-curing, resulting in distorted test data and a long-standing batch quality hazard of "passing inspection but poor curing." This invention features a unique three-level time-division non-overlapping spectral excitation architecture that completely separates the timing of impurity detection, oxide layer detection, and main thickness detection. With no light overlap and no heat accumulation, it eliminates the pre-curing defect of colloids from the source, accurately reproducing the initial coating state of the colloid. This significantly addresses the core pain points of traditional optical inspection—inaccurate and unreliable measurements—and its state fidelity far surpasses conventional industry optimization practices.

[0025] In the prevailing understanding of the field, the saturation suppression optical path for hard-shell high-reflectivity mirror surfaces and the impurity suppression optical path for soft-pack diffuse reflection rough surfaces are technically incompatible, making it impossible to achieve high-precision detection with integrated equipment. This invention creatively integrates a 42° suppressed total internal reflection optical path and an 85° grazing incidence polarization differential optical path. Through predictive automatic switching and a dual-signal locking mechanism, it achieves universal high-precision detection for both types of extremely complex surfaces, increasing the defect detection rate of both soft and hard-shell batteries to over 99%. Simultaneously, it breaks through the limitations of traditional optics, which can only perform instantaneous static detection. It features a unique multi-time-series dynamic data fitting curing kinetics prediction model, which can predict later defects such as colloid curing shrinkage, internal stress concentration, and adhesive delamination in advance. This represents a disruptive technological upgrade from "post-defect screening" to "pre-risk prediction," with dual technological gains that are beyond the predictive capabilities of conventional industry technologies.

[0026] Traditional equipment cannot distinguish between the effective area of ​​colloids and impurities such as oil, dust, and electrolyte residues, resulting in a false detection rate as high as 9.5% in mass production. Furthermore, it is affected by temperature changes, vibration, device aging, and mechanical micro-deformation, leading to a 28.3% accuracy decay during 24-hour continuous operation. Frequent manual shutdowns for calibration are required, resulting in poor mass production adaptability. This invention utilizes the difference in ultraviolet fluorescence characteristics to achieve pixel-level impurity masking, reducing the false detection rate to 0.2% without increasing hardware costs. Simultaneously, it constructs a full-temperature-range dynamic compensation + daily full-range automatic calibration system to adaptively offset various operating condition interferences. The accuracy decay during 24-hour continuous mass production is controlled within 3.1%, maintaining high-precision and stable operation for extended periods without manual intervention. This significantly solves the industry-wide problems of mass production drift and frequent false detections in optical inspection equipment, achieving unmanned, highly consistent mass production inspection results that are unattainable by conventional technologies. Attached Figure Description

[0027] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the device of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0029] See Figure 1 and 2 This invention is divided into two main parts: a battery assembly seam overflow anti-reflective optical sensing detection method and a battery assembly seam overflow high-precision detection device. The device provides the hardware platform for the method, while the method provides the detection logic, optical algorithm and control process for the device. The two work together to complete the entire process of battery assembly seam overflow detection.

[0030] A method for detecting excess adhesive at battery bonding seams using an optical sensor to reduce reflectivity includes the following steps: Step 1. Battery type identification and optical path pre-positioning lock This step is the starting point of the testing process. Its core purpose is to adaptively match the corresponding anti-reflective optical path based on the battery material and surface characteristics, complete the optical path switching in advance, and adapt to the high-speed production line cycle.

[0031] The product category identification device simultaneously receives two identification signals: the first is the battery category code signal issued by the production line PLC, which is marked by the upstream process in advance; the second is the battery shape and surface morphology data collected by the shape contour sensor at the front of the frame. The two signals are cross-validated to accurately determine whether the workpiece to be inspected is a hard-shell battery or a soft-pack battery.

[0032] The optical path switching and locking mechanism determines that if the battery is a hard-shell battery, the main control unit sends a command to drive the electric translation slide to move the 42° oblique incidence suppressed total internal reflection optical path to the working position; if the battery is a soft-pack battery, it switches to the 85° grazing incidence polarization differential optical path.

[0033] The switching action is completed 0.5 seconds before the battery enters the testing station. After the slide is in place, the mechanical limit sensor and the photoelectric position sensor simultaneously send out the position signal. After the main control unit confirms that there is no deviation in the optical path attitude and position, it enters the next testing stage.

[0034] The high reflectivity of the hard-shell metal surface is addressed by using an oblique incident light path with an angle slightly greater than the critical angle to excite suppressed total internal reflection and thin film interference signals, thus weakening reflective interference. The surface of the soft-pack aluminum-plastic film has severe diffuse reflection, which is addressed by using a large-angle grazing incident light + polarization difference structure to filter stray light. These two types of light paths specifically solve different reflective problems and form the basis of anti-reflective detection.

[0035] Step 2. Time-sharing staggered optical excitation and pre-excitation impurity shielding This step employs a three-stage independent timing sequence to complete optical excitation, sequentially achieving oil and impurity detection, metal oxide layer detection, and main signal acquisition of the adhesive layer. The light sources at each timing sequence are completely staggered, with no overlapping illumination. Simultaneously, impurity interference is shielded at the detection front end, improving detection accuracy from the source.

[0036] First sequence: Detection of fluorescent oil stains and impurities (impurity shielding step) The main control unit only activates the 365nm fluorescent light source. The beam illuminates the area to be tested at the battery bonding seam through an independent optical path. The matching fluorescent acquisition sensor collects surface fluorescence images and fluorescence dynamic characteristic data over the entire area.

[0037] Impurities such as oil, organic dust, and electrolyte residues adhering to the battery surface will produce characteristic fluorescence under 365nm ultraviolet light excitation, while colloidal and metal substrates do not have this type of fluorescence response. The effective colloidal area and the impurity area can be distinguished by the fluorescence characteristics.

[0038] Three core fluorescence features are extracted: absolute fluorescence intensity, fluorescence spectral peak shift, and fluorescence dynamic decay rate. Based on preset multi-dimensional joint judgment thresholds according to workshop conditions, the feature data are compared with the thresholds to mark oil, dust, and impurity areas in the image at the pixel level, and an electronic mask is generated.

[0039] In all subsequent adhesive layer thickness calculations and defect judgment processes, the system automatically masks the pixel areas corresponding to the mask, performing calculations only on the effective adhesive areas, significantly eliminating false detections caused by impurities. Simultaneously, the system supports adaptive fine-tuning of shift thresholds to adapt to dynamically changing oil concentrations in the workshop.

[0040] After completing the full-domain image acquisition and mask generation, the 365nm fluorescent light source is turned off, and the stray light in the optical path is allowed to completely dissipate before entering the second time sequence.

[0041] Second time series: Metal oxide layer parameter detection After the fluorescent light source is turned off, two deep ultraviolet light sources, 255nm and 275nm, are simultaneously activated. The light beams are irradiated onto the surface of the battery metal substrate through independent isolated optical paths, and the ultraviolet reflection sensor collects the surface reflectivity data corresponding to the two wavelengths in real time.

[0042] An aluminum oxide layer naturally forms on the surface of the metal casing of lithium batteries. The thickness of this oxide layer exhibits random dispersion, which alters the phase of the light beam and is a major source of error in optical detection systems. Aluminum oxide shows significant differences in its absorption and reflection characteristics for ultraviolet light in the 255nm and 275nm wavelength bands. By utilizing the difference in reflectivity between these two wavelengths, the actual thickness of the oxide layer can be accurately calculated.

[0043] After the reflectivity data acquisition is completed, the two ultraviolet light sources are turned off, and the optical path stabilizes before entering the third timing sequence.

[0044] Third timing sequence: Acquisition of main optical signals of the adhesive layer After turning off the ultraviolet light source, the three visible and near-infrared main detection light sources of 450nm, 650nm and 850nm are turned on in sequence. The light source adopts a fixed frequency narrowband phase-locked modulation mode with a modulation frequency of 100kHz. The light source modulation signal and the imaging device sampling signal are synchronized at the microsecond level to improve the signal-to-noise ratio.

[0045] The three wavelengths of visible and near-infrared light have different interference response sensitivities to adhesive layers of different thicknesses. The combination of multiple wavelengths can construct an overdetermined set of equations, solving the problem of multiple solutions or no solutions in single-wavelength calculations and ensuring the accuracy of calculations across the entire range.

[0046] The three-stage time sequence operates independently without superimposed light source illumination, avoiding premature flow and solidification of the colloid due to absorption of light and heat, thus ensuring that the collected optical signals can truly reflect the initial state of the colloid.

[0047] Step 3. Calculate oxide layer thickness and perform wavelength-by-wavelength phase compensation using the dual-wavelength ultraviolet differential method. This step utilizes the dual-wavelength reflectivity data acquired in the second time series to calculate the thickness of the alumina oxide layer using a fixed formula, and combines it with a calibration database to perform dynamic phase compensation, thereby eliminating the optical phase shift error caused by the oxide layer.

[0048] The thickness of the alumina oxide layer on the metal surface is calculated using a dual-wavelength ultraviolet differential algorithm. The calculation formula is as follows: ; in, : The actual thickness of the aluminum oxide layer on the metal surface of the battery under test, in units of length, is the physical quantity to be solved by this formula; Pi (π) is a mathematical constant with a fixed value of 3.1415926. The intrinsic extinction coefficient of alumina medium at a UV wavelength of 275nm is dimensionless and belongs to the inherent physical parameters of the material. The equipment is calibrated and cured using standard alumina samples at the factory. : The intrinsic extinction coefficient of alumina medium at 255nm ultraviolet wavelength, dimensionless, an inherent parameter of the material, factory calibrated and cured; The reflectance of the metal surface under test at a wavelength of 255nm, acquired in real time by the device, is dimensionless and is sampled in real time. Standard reflectance of a pure metal reference surface without an oxide layer at a wavelength of 255nm, dimensionless, factory-calibrated and locked reference parameter; The reflectance of the metal surface under test at a wavelength of 275nm, acquired in real time by the device, is dimensionless and is sampled in real time. Standard reflectance of a pure metal reference surface without oxide layer at a wavelength of 275nm, dimensionless, factory-calibrated and locked reference parameter; : Natural logarithm operation function, standard mathematical operation rules.

[0049] Before the equipment leaves the factory, multiple sets of standard alumina samples with different thicknesses are used to conduct full-band testing to establish a three-dimensional calibration database of wavelength-oxide layer thickness-phase shift. The database records the beam phase shift of the three main detection light sources at 450nm, 650nm, and 850nm under different oxide layer thicknesses.

[0050] The actual oxide layer thickness calculated by the formula is input into the main control unit. The main control unit searches the three-dimensional calibration database and independently matches the corresponding phase offset compensation value for each main detection light source.

[0051] By dynamically correcting the phase of the wavelength, the beam phase reference is unified under different oxide layer thicknesses, which greatly eliminates the optical detection system error caused by the discreteness of the oxide layer on the metal surface, and ensures the uniqueness and accuracy of subsequent thin film interferometry calculation and suppressed total internal reflection calculation.

[0052] Step 4. Multi-time sequence signal acquisition and multi-dimensional composite vibration suppression This step is performed after the main detection light source in the third time sequence has completed optical signal acquisition and before the segmented adhesive layer thickness calculation. It is used to correct for vibration interference and filter out noise from the acquired raw optical sensing data to ensure that the input data for subsequent thickness calculation is true and valid.

[0053] During the operation of the main detection light source, the system continuously acquires optical image and signal data at three time points: 0s, 5s, and 10s. 0s timing: corresponds to the initial state of the colloid just after it has been applied, reflecting the original thickness and distribution characteristics of the colloid. 5s timing: corresponds to the transition state of the initial flow of the colloid; 10s timing: corresponds to the transitional state of the initial curing of the colloid surface layer.

[0054] The three sets of time-series data fully cover the dynamic changes in the early stage of colloid curing, providing sufficient data support for subsequent curing kinetic model fitting and prediction of the finished product's overflow state.

[0055] Based on vibration data collected by multi-directional MEMS vibration sensors, vibration type and vibration direction are distinguished, and a combined suppression strategy based on type and direction is adopted. The specific implementation method is as follows: Horizontal vibration correction: Horizontal vibration can cause lateral shift of the light beam spot and misalignment of imaging pixels. The main control unit calculates the image pixel offset in real time based on the horizontal vibration amplitude and direction fed back by the sensor, and performs pixel position correction on the acquired optical image to compensate for the imaging deviation caused by horizontal vibration.

[0056] Vertical vibration compensation: Vertical vibration can alter the air gap between the prism and the battery surface. The system, in conjunction with a laser ranging closed-loop compensation mechanism, fine-tunes the electric slide in real time to stabilize the air gap within a standard range, thus offsetting the impact of vertical vibration on the optical path reference.

[0057] Steady-state periodic vibration suppression: The operation of the production line transmission mechanism generates steady-state periodic vibration. The system adopts a phase-locked synchronous sampling method to synchronize the image sampling time sequence with the vibration period in reverse, and uses the timing to cancel the interference of periodic vibration.

[0058] Instantaneous impact vibration suppression: Instantaneous random impact vibrations generated during workpiece transfer and mechanical positioning can easily cause distortion in single-frame images. The system adopts a multi-frame image optimization and distorted frame rejection mechanism: It continuously acquires multiple frames of images and automatically rejects distorted frames based on image clarity and signal stability criteria, selecting valid clear frames to participate in subsequent calculations, thus preventing data loss and missed detection problems on high-speed production lines.

[0059] Step 5. Segmented optical mechanism modeling and adhesive layer thickness calculation This step strictly divides the optical physical model according to the thickness range of the adhesive layer, abandoning the defect of using a single model for the entire range. Based on the different optical response mechanisms, it is divided into two parts: the nano-thin adhesive layer range and the medium-thick adhesive layer range, which are calculated separately.

[0060] 5.1 nm ultrathin adhesive layer range (0.5 nm ~ 100 nm): Modified FTIR suppressed total internal reflection model When the thickness of the adhesive layer is in the range of 0.5nm to 100nm, there is no obvious macroscopic height difference in the adhesive layer, the sensitivity of the thin film interference signal is extremely low, and the beam mainly transmits the adhesive layer thickness information in the form of suppressed total internal reflection evanescent wave. Therefore, the thin film transmission matrix model is not used in this range, and the modified FTIR model is used alone to calculate the thickness.

[0061] A metal region adjacent to the adhesive joint to be tested, of the same material, without differences in adhesive or oxide layer, is selected as the reference area. This region is completely identical to the region to be tested in terms of dielectric structure, optical path conditions, and environmental interference, ensuring that the physical premise for differential calculation is valid.

[0062] The formula for calculating the normalized effective total reflection suppression signal is as follows: ; in, : The normalized effective suppressed total internal reflection signal intensity, dimensionless, is the effective calculated signal output by this formula; The intensity of the FTIR reflected signal in real time at the adhesive layer area to be tested at the joint of the battery assembly is dimensionless and is collected in real time. : FTIR reference signal intensity acquired in the reference reference area (glue-free metal area) within the same field of view, dimensionless, reference data collected synchronously.

[0063] By performing ratio normalization calculations, common-mode interference such as air gap fluctuations, light source intensity jitter, and ambient stray light is canceled out, enabling stable and high-precision measurement of nanoscale ultrathin adhesive layers ranging from 0.5nm to 100nm.

[0064] 5.2 Medium-thickness adhesive layer range (100nm~5mm): Transport matrix model of four-layer dielectric thin film When the thickness of the adhesive layer is greater than 100nm, the optical path exits the suppressed total internal reflection working range, and the thin film interference effect becomes the main response mechanism of the beam. In this range, the thickness is solved by combining the multilayer thin film transmission matrix (TMM) model with the oblique incidence optical path difference formula. The s-polarized light is used as the calculation reference throughout the entire range.

[0065] First, we use Snell's law of refraction to solve for the angle of refraction of light after it enters the colloidal medium. The formula is as follows: ; in, : The refractive index of air, dimensionless, a physical constant, with a fixed value of 1.0; Sine trigonometric functions, standard mathematical operation rules; The incident angle of the light beam in air is expressed in degrees. The optical path of the hard-shell battery is fixed at 42°, and the optical path of the soft-pack battery is fixed at 85°. These are known fixed parameters. The equivalent refractive index of the battery encapsulating colloid is dimensionless and is retrieved from the system's preset parameter library based on the colloid material; it is a known parameter. : The angle of refraction produced when light is incident on a colloidal medium, expressed in degrees, and is an intermediate parameter to be solved in this formula.

[0066] Substituting the obtained refraction angle into the oblique incidence optical path difference correction formula, and combining it with the previously calculated oxide phase compensation, the total phase optical path difference is obtained; then, based on the four-layer real medium structure of air-colloid-alumina oxide-metal substrate, the overall optical transmission matrix is ​​constructed: ; In the formula: It is a four-layer dielectric overall optical transmission matrix; For the transport matrix at the air-colloid interface; This represents the transport matrix at the interface between the colloid and the oxide layer. This is the transport matrix at the interface between the oxide layer and the metal substrate. The matrix elements are determined by the refractive index, extinction coefficient, thickness, and incident angle of the medium.

[0067] The system constructs an overdetermined set of equations based on real-time reflectivity responses at three wavelengths: 450nm, 650nm, and 850nm. A least-squares iterative algorithm is then used to solve these equations, yielding a unique optimal solution for the colloidal thickness. This approach significantly addresses the issues of signal aliasing, multiple solutions, and no solution in the transition region around 100nm, enabling seamless detection across the entire measurement range from 100nm to 5mm.

[0068] Step 6. Matching of colloid curing kinetic model, state prediction and defect determination The system pre-stores curing kinetic models for four types of commonly used lithium-ion battery colloids: a double-exponential curing flow model for thermally conductive silicone, an S-shaped curing flow model for epoxy resin, an exponential curing flow model for UV adhesives, and a delayed response model for slow-curing structural adhesives. All models are based on classical theories of polymer rheology and curing kinetics, modified to account for the time-dependent changes in lithium-ion battery encapsulation thickness, and possess clear physical meaning and material mechanism support. The specific function forms and construction methods are as follows: The variation of adhesive overflow thickness over time is as follows: after application, the adhesive initially flows and spreads, gradually decreasing in thickness. As the curing reaction proceeds, the viscosity increases, flow ceases, and the thickness eventually stabilizes. All four models use h(t) as the measured adhesive overflow thickness at time t, h∞ as the final stable thickness after complete curing, and h0 as the initial adhesive thickness at time 0s as the basic variables, as shown in the following forms: The bi-exponential curing flow model for thermally conductive silicone is suitable for sealing materials with good flowability, such as thermally conductive silicone. The curing process consists of two stages: rapid spreading and slow curing shrinkage. The functional expression is as follows: ; In the formula: The percentage of amplitude during the rapid flow phase (range of values) ); This represents the rate constant during the rapid spreading phase. This is the rate constant for the slow curing shrinkage stage; the other parameters are defined as above.

[0069] The S-curing flow model for epoxy resins is applicable to self-catalytic epoxy resin sealing materials. The flow-curing process exhibits an S-shaped change characteristic of "slow-fast-slow," and the functional expression is as follows: ; In the formula: The inflection point time of the flow rate; This is the flow-to-solidification transition rate constant; the other parameters are defined as above.

[0070] The UV adhesive exponential curing flow model is suitable for UV adhesives that cure rapidly. It exhibits fast curing rates and a monotonically convergent thickness distribution with a single exponential curve. The function expression is: ; In the formula: This is the overall flow rate constant for solidification; the other parameters are defined as above.

[0071] The delayed response model for slow-curing structural adhesives is applicable to materials with a distinct curing induction period. During this period, the thickness change is minimal, followed by rapid and stable curing. The functional expression is: when... hour, ;when hour, In the formula: This refers to the curing induction period (delay time). This is the curing flow rate constant after the induction period; the other parameters are defined as above.

[0072] The above model is not an unfounded empirical fit, but a mechanism model solidified after calibration through materials science experiments. The construction process is as follows: Based on the curing reaction mechanism and rheological properties of various sealing materials, the corresponding kinetic model is matched to ensure that the curve shape is consistent with the actual change law of the material. For each type of colloid, an in-situ optical thickness measurement platform was used in conjunction with a constant temperature chamber to conduct 24-hour full curing cycle thickness tracking tests at four typical operating temperature points: 0℃, 15℃, 30℃, and 45℃, collecting hundreds of sets of thickness-time continuous data. The model characteristic parameters (rate constant, inflection point time, induction period, etc.) at each temperature were obtained by fitting with a nonlinear least squares algorithm, and a three-dimensional calibration database of "colloid type-ambient temperature-model parameters" was established and pre-stored in the main control system. During actual testing, the system interpolates and retrieves the baseline parameters at the corresponding temperature based on the real-time collected workshop ambient temperature, and then fine-tunes them in combination with the measured data to eliminate the influence of temperature change on the curing rate.

[0073] The system automatically identifies the colloid type and matches the corresponding curing baseline model based on the spectral characteristics, reflectance characteristics, and thickness change rate over a 0-10s time series of the colloid under test, according to a preset model matching threshold, without manual intervention. Simultaneously, the system collects real-time temperature and humidity data from the workshop, dynamically correcting the internal rate coefficient and steady-state coefficient of the model to compensate for the effects of temperature and humidity on colloid flow and curing rate.

[0074] By substituting the measured thickness data of three time series (0s, 5s, and 10s) into the corrected curing kinetics model, and through curve fitting and data extrapolation calculation, the final overflow thickness after the colloid is fully cured and the dimensions are stable can be accurately predicted, thus allowing for early prediction of the finished product quality status.

[0075] This invention has been verified through multiple sets of comparative experiments. The experimental logic is as follows: the form of the model function is predetermined by the material curing mechanism (the curve shape is known), and it is not an unconstrained free fitting; the initial boundary is determined at 0s, and the core rate parameters and the final stable value of the model can be solved at the two dynamic points of 5s and 10s. For the monotonically convergent curing flow curve, the three-point fitting has sufficient engineering accuracy.

[0076] Four typical sealing materials were selected for the experiment. Under two typical mass production conditions of 25℃ (room temperature) and 35℃ (high temperature), the final cured thickness was extrapolated using three-point data at 0s, 5s, and 10s. The predicted thickness was then compared with the actual thickness after 24 hours of complete curing to verify the accuracy of the prediction. The experimental data are shown in the table below: The table below is a comparison and verification table of the predicted final cured thickness based on three-point extrapolation and the measured value after 24 hours.

[0077] With the mechanism model pre-calibrated, the final cured thickness was extrapolated by fitting three time points: 0s, 5s, and 10s. The relative prediction error under all working conditions was less than 3%, which fully met the industrial precision requirements for overflow detection in lithium battery production lines and demonstrated sufficient engineering authenticity and performance reliability.

[0078] The system integrates three major judgment indicators: uniformity of nano-adhesive film, real-time dynamic thickness of colloid, and predicted final thickness after curing. Combined with the process thresholds of the lithium battery industry, it classifies the results into four categories: qualified product, nano-adhesive film distribution defects, localized thin adhesive defects, and thick adhesive accumulation defects.

[0079] The device is equipped with a barcode recognition unit that scans the battery's unique traceability barcode and binds the barcode number, inspection time, battery type, adhesive layer thickness in each area, defect type, defect coordinates, and curing prediction data. This data is then uploaded in real-time to the production line's PLC and MES production management system via industrial Ethernet, enabling end-to-end quality traceability. Simultaneously, based on the judgment results, it outputs switch signals to drive the production line's sorting mechanism to automatically divert defective products and automatically release qualified products.

[0080] Step 7. End-to-end temperature adaptive compensation mechanism This step is performed simultaneously with the vibration suppression mentioned above, together completing the full-condition correction and noise reduction of the original optical signal, providing a unified and stable detection benchmark for subsequent segmented thickness calculation.

[0081] The ambient temperature in lithium battery production workshops fluctuates between 0℃ and 45℃. Temperature changes can cause drift in four types of parameters: colloidal refractive index, optical refractive index of partitioned prisms, lens surface deformation, and laser source output wavelength, thus leading to detection errors. This invention establishes a full-link temperature adaptive compensation mechanism, the specific implementation of which is as follows: Multi-point calibration was performed using a high and low temperature test chamber. The calibration temperature points were set to 0℃, 15℃, 30℃, and 45℃. Parameter change data of various optical components and colloids were collected at different temperatures to generate a segmented temperature compensation curve. A smooth transition algorithm was used between adjacent temperature nodes to avoid data step deviation.

[0082] During operation, the multi-point infrared temperature sensor collects temperature data from four types of components in real time: the optical cavity, the prism, the light source, and the environment. The main control unit retrieves the compensation coefficients for the corresponding temperature range and synchronously corrects the four types of drift parameters to achieve adaptive compensation across the entire temperature range of 0~45℃, ensuring stable detection accuracy under high and low temperature conditions.

[0083] Step 8. Full-range automatic calibration mechanism To address issues such as mechanical micro-deformation, component aging, and parameter drift that occur during long-term equipment operation, the device is equipped with a full-range automatic calibration mechanism. The specific implementation process is as follows: After the device is powered on and initialized daily, it automatically starts the full-range self-calibration process without the need for manual triggering.

[0084] A gradient standard thickness sample set was adopted, with sample thicknesses of 0nm, 1nm, 10nm, 100nm, 1μm, and 1mm, respectively, fully covering the entire detection range of nanometers, micrometers, and millimeters.

[0085] The system sequentially tests each standard sample, compares the measured data with the nominal standard data of the sample, and automatically iteratively corrects the optical path parameters, algorithm coefficients, and signal references to eliminate accuracy deviations caused by overnight temperature changes, mechanical deformation, and device aging, ensuring daily start-up testing accuracy and batch consistency.

[0086] The complete continuous workflow of this invention is as follows: Device standby → Category identification + optical path pre-switching lock → First time sequence: Fluorescent light source start-up → Impurity identification + mask generation → Fluorescent light source off → Second time sequence: Dual ultraviolet light source start-up → Reflectivity acquisition + oxide layer thickness calculation → Wavelength-by-wavelength phase compensation → Ultraviolet light source off → Third time sequence: Three main light sources start-up → 0s / 5s / 10s multi-time sequence signal acquisition → Vibration correction, temperature compensation → Call the corresponding optical model and calculate the adhesive layer thickness according to the thickness range → Adaptive matching curing model + extrapolation prediction of final adhesive overflow thickness → Defect classification judgment → Barcode binding + data upload to MES + automatic production line diversion → Daily power-on automatic full-range calibration → Cyclic execution.

[0087] The entire process is a continuous closed loop, with all steps operating automatically, making it suitable for continuous mass production on automated lithium battery production lines.

[0088] A high-precision detection device for excess adhesive at battery bonding seams is disclosed. This device, used to execute the aforementioned detection method, serves as a dedicated hardware carrier for the optical sensing detection method. It adopts a modular integrated design and mainly comprises seven components: a frame support module, a dual-mode independent optical path switching module, a multi-band time-division light source excitation module, an anti-crosstalk zoned optical imaging module, a multi-parameter closed-loop operating condition compensation module, a main control and computing module, and an explosion-proof and corrosion-resistant safety protection module. The modules are physically isolated, electrically partitioned, and allow for independent signal transmission.

[0089] The frame support module is the basic support structure of the entire device. It is assembled from industrial aluminum alloy profiles with hard anodizing treatment on the surface. It features high structural strength, small deformation, corrosion resistance, and excellent shock absorption performance, and can withstand the continuous vibration of the production line and the corrosive gas erosion of the workshop for a long time.

[0090] The interior of the rack is divided into four completely independent, sealed cavities: an optical cavity, an electrical cavity, a main control cavity, and a cleaning cavity, by metal sheet partitions. There are no gaps between the cavities, achieving complete physical isolation.

[0091] Optical cavity: This is the core precision area of ​​the device, where all optical components and optical path structures are arranged. All joints and gaps in the cavity are filled with corrosion-resistant sealant to achieve the effects of dustproofing, moistureproofing, and blocking the penetration of electrolyte mist. A slightly positive pressure dry inert gas is continuously introduced into the cavity, and the gas pressure is maintained at 0.02MPa~0.03MPa. The pressure difference is used to prevent external dust and corrosive gases from entering the optical path area.

[0092] Electrical cavity: All light source drive circuits, light source temperature control circuits, and sensor signal conditioning circuits are centrally located. The interior of the cavity is treated with a three-proof coating, which has anti-static, anti-corrosion, and anti-oxidation capabilities, and avoids circuit overheating and electromagnetic radiation interference with precision optical signals.

[0093] Main control cavity: used to house the main control computing hardware, data storage unit, and communication unit. The independent space isolates external vibration and electromagnetic interference, ensuring stable computing and data transmission.

[0094] Cleaning chamber: A separate non-destructive cleaning mechanism is arranged, and an openable and closable pneumatic isolation baffle is set between it and the optical chamber. Under normal conditions, the baffle is closed, completely separating the cleaning operation area from the optical area to prevent airflow, charged particles, and ozone from interfering with the optical path during the cleaning process.

[0095] The bottom of the frame is equipped with passive shock-absorbing pads to attenuate low-frequency ground vibrations; the top of the frame has a pre-installed standardized mounting flange, which can be adapted to the two mainstream production line installation methods of vertical hoisting and horizontal fixing, and is compatible with lithium battery automated assembly lines of different structures.

[0096] The dual-mode independent optical path switching module is the core structure for achieving anti-reflective detection. It is designed to address the high reflectivity of the smooth metal surface of hard-shell batteries and the high diffuse reflection of the aluminum-plastic film surface of soft-pack batteries. It is equipped with two completely independent optical detection optical paths with fixed parameters. The two optical paths are installed as a whole on a high-precision electric translation slide. The optical path switching is completed by the movement of the slide. The switching logic relies on the production line signal to achieve predictive pre-switching, matching the high-speed production line operation cycle.

[0097] The two sets of optical path structures and optical parameters are calibrated and permanently cured at the factory. During assembly, metal pressure rings are used to lock all optical lenses and lens groups, ensuring no loose gaps and guaranteeing that the optical path angle, optical path, and spot position will not shift under long-term vibration conditions.

[0098] The 42° oblique incidence suppressed total internal reflection optical path for hard-shell batteries such as those with aluminum or steel shells is designed for use with these batteries. The incident angle is set at 42°. The critical angle for total internal reflection at the interface between the BK7 optical glass and the air medium is approximately 41.8°. The 42° incident angle is slightly larger than the critical angle, which can stably excite the suppressed total internal reflection effect and generate evanescent waves carrying nanoscale adhesive layer thickness information. At the same time, this angle can simultaneously excite the thin film interference effect, covering the detection range of adhesive layers from micrometers to millimeters.

[0099] This optical path uniformly uses s-polarized light as the incident reference, artificially defining a unique polarization state to avoid the problems of phase abrupt changes and optical path difference calculation errors caused by different polarized light. The optical path integrates a collimating lens, a focusing lens, and a narrowband filter. After collimation, the beam is incident on the surface of the battery under test at a fixed angle, and the reflected light enters the back-end optical imaging module according to the predetermined optical path.

[0100] The 85° grazing incidence polarization grazing optical path for pouch batteries is designed for aluminum-plastic film packaged pouch batteries. It adopts an 85° ultra-large angle grazing incidence structure, which greatly shortens the propagation path of stray light and suppresses diffuse reflection stray light and specular reflection caused by the rough structure of the aluminum-plastic film surface from a physical level.

[0101] The optical path integrates a pre-linear polarizer, a post-high-precision analyzer, and a polarization angle calibration structure. These three components work together to form a polarization difference detection system: the polarizer converts the incident beam into unidirectional polarized light, the analyzer filters the effective polarization signal in the reflected beam, filters out unpolarized stray light randomly scattered from the rough surface, and retains only the regular mirror signal reflected by the adhesive layer, which greatly solves the signal distortion problem caused by reflection and stray light interference from the soft-pack battery.

[0102] The electric translation slide has a repeatability of ≤0.005mm. Mechanical limit sensors and photoelectric position sensors are installed on both sides of the slide for dual verification of the optical path position. The device receives battery type signals from the production line PLC in real time, and at the same time, it uses the outline sensor at the front of the frame to identify the shape characteristics of the battery for dual determination of battery type. 0.5 seconds before the battery body enters the inspection station, the slide completes the optical path switching and locks, eliminating the problem of missed detection and waste frames caused by "battery in position, optical path not switched" in high-speed production lines.

[0103] The multi-band time-division light source excitation module provides excitation light sources of different bands for optical detection. According to function, it is divided into three categories: fluorescent oil stain detection light source, ultraviolet oxide layer detection light source, and visible and near-infrared main detection light source. All light sources are independent units, and adopt an independent optical path, independent constant temperature control, and time-division staggered emission mode. From both hardware and control logic aspects, problems such as light source crosstalk, light intensity drift, and heat curing of colloids are eliminated.

[0104] The light source composition and hardware structure include: Fluorescent oil stain detection light source: A 365nm wavelength light source is used to excite oil stains and organic impurities on the battery surface to generate fluorescent signals; Ultraviolet oxide layer detection light source: includes two deep ultraviolet light sources of 255nm and 275nm, used to collect metal surface reflectivity data and calculate oxide layer thickness; Visible and near-infrared main detection light source: includes three light sources of 450nm, 650nm and 850nm, which serve as the core excitation light source for adhesive layer thickness detection.

[0105] Each light source is equipped with an independent temperature control component, with a temperature control accuracy of ±0.1℃. This compensates for the operating temperature rise of the light source chip and fluctuations in ambient temperature in real time, ensuring long-term stability of the light source's output wavelength, spot shape, and intensity amplitude. Each light source's output end is equipped with a narrow-band deep-cutoff filter corresponding to the wavelength band, with a cutoff depth of no less than OD6, completely filtering stray light from non-target wavelength bands and ensuring the purity of the optical path for each light source. All light source drive circuits and temperature control circuits are housed in independent electrical cavities, physically isolated from the optical cavity.

[0106] This module strictly adheres to a three-level time-sequential independent emission rule, ensuring no overlap in irradiation periods between different functional light sources. This prevents premature flow and solidification of the colloid due to the combined heating of multiple light sources, guaranteeing that the optical data collected at different time sequences accurately reflect the original state of the colloid. The time sequence division and working content will be explained in detail in the subsequent detection methods.

[0107] The anti-crosstalk partitioned optical imaging module is responsible for receiving the optical signals reflected from the surface under test and completing the signal partitioning acquisition according to the thickness range of the adhesive layer. The core component is an integrated BK7 optical glass partitioned prism, which, together with the light-shielding structure and independent light-shielding channel, achieves physical decoupling of the two signals, greatly eliminating crosstalk and spatial stray light interference problems inside the prism medium.

[0108] The working surface of the integrated BK7 partitioned prism is equipped with hard black light-blocking ribs. These ribs strictly divide the prism's working surface into two physically independent areas: the left FTIR nano-detection area and the right angular interference detection area. Light cannot propagate across these areas.

[0109] The left-side FTIR nano-detection area is a fixed-point signal acquisition area with no angled beam splitting structure. It is specifically designed to acquire suppressed total internal reflection evanescent wave signals of 0.5nm~100nm nano- and submicron-scale ultrathin films, and to acquire signals for minute defects such as nano-adhesive overflow, missing ultrathin insulating films, and uneven film distribution. Right-side angular interference detection area: Equipped with an astigmatism-reducing cylindrical lens group, it can continuously split the reflected beam at an angle, specifically for collecting thin film interference signals of adhesive layers with thicknesses ranging from 100nm to 5mm (micrometers to millimeters), and for collecting signals of macroscopic defects such as conventional adhesive overflow, adhesive accumulation, and localized adhesive deficiency.

[0110] The reflected light paths corresponding to the two regions are respectively connected to independent black light-shielding channels within the cavity. The inner walls of the channels are coated with a high-absorption light-absorbing coating, which can absorb more than 99.9% of stray light, achieving complete decoupling of the two signals. The two imaging signals are ultimately projected onto independent pixel areas of the industrial imaging device, with physical isolation at pixel boundaries and no edge diffraction crosstalk.

[0111] The surface of the partitioned prism adopts a double-layer composite coating process. The bottom layer is a sapphire hard protective layer with a thickness of 100nm~150nm, which improves the surface hardness and resists friction and scratches from workshop dust particles. The top layer is an inert fluoride self-healing coating with a thickness of 8nm~12nm, which has the ability to resist electrolyte corrosion, moisture, and minor scratches, thus extending the service life of core optical components.

[0112] The multi-parameter closed-loop operating condition compensation module integrates various sensors and electric actuators to collect on-site operating parameters in real time and form a closed-loop adjustment circuit. It dynamically compensates for three major interference sources: air gap fluctuations, mechanical vibrations, and ambient temperature. It is a key structure to ensure the long-term detection accuracy of the equipment. The module mainly includes a multi-directional MEMS vibration sensor, a multi-point infrared temperature sensor, a laser ranging closed-loop sensor, and a high-precision electric fine-tuning slide.

[0113] The laser ranging closed-loop sensor monitors the vertical air gap between the optical working surface of the partition prism and the battery surface to be measured in real time. The standard working air gap of the equipment is set to 0.15mm, and the allowable fluctuation range is ±0.02mm.

[0114] The sensor transmits real-time gap data to the main control unit, which then drives a high-precision electric fine-tuning slide to correct the gap size in real time based on the deviation. This closed-loop circuit can dynamically compensate for gap drift caused by battery casing dimensional tolerances, long-term minor deformation of the equipment, and vertical vibration, solving problems such as FTIR signal attenuation, interference fringe shift, and large thickness measurement dispersion caused by air gap fluctuations.

[0115] Multi-directional MEMS vibration sensors synchronously collect vibration data in both horizontal and vertical dimensions, distinguishing between steady-state periodic vibrations generated by production line transmission mechanisms and instantaneous random impact vibrations generated by workpiece transfer and mechanical positioning. The vibration data is uploaded to the main control unit in real time, providing the original basis for vibration suppression algorithms.

[0116] Multi-point infrared temperature sensors collect four types of temperature data: the temperature inside the optical cavity, the surface temperature of the partitioned prism, the temperature of the light source chip, and the ambient temperature of the workshop. This comprehensively covers the temperature changes of the optical path, the light source, and the environment, providing raw parameters for the end-to-end temperature adaptive compensation algorithm.

[0117] The main control and computing module adopts an FPGA combined with an industrial main control architecture. It is the logic and computing center of the entire device and undertakes five core tasks: First, it receives signals from various sensors, production line PLCs, and appearance sensors, and completes logical judgments and command issuance; second, it controls the actions of actuators such as the time-division timing of the light source, optical path switching, cleaning mechanism, and isolation baffle; third, it runs optical calculation models and various algorithm formulas to complete adhesive layer thickness calculation, phase compensation, vibration correction, and temperature compensation; fourth, it runs the adhesive curing dynamics model to complete the prediction of adhesive overflow state and defect judgment; and fifth, it realizes functions such as data storage, barcode binding, MES system communication, and fault early warning.

[0118] The module has built-in full-range standard sample calibration data, temperature compensation curve database, wavelength-oxide layer thickness-phase shift calibration database, and parameter library for various colloidal curing models. All data is calibrated at the factory and supports on-site iterative updates. The calculation process is fully automated and requires no manual intervention.

[0119] The explosion-proof and corrosion-resistant safety protection module is designed in strict accordance with the relevant specifications for explosion-proof, electrical safety, laser safety, and corrosion prevention in lithium battery production workshops. It completes the safety protection structure required for industrial mass production and ensures that the equipment passes the workshop safety regulations acceptance.

[0120] The entire unit has an explosion-proof rating of ExdIIBT4. The optical and electrical cavities are all sealed explosion-proof structures, with no exposed spark-generating components inside. The circuit is partitioned: the low-voltage signal acquisition circuit uses an intrinsically safe design, while the high-power drive circuit uses explosion-proof packaging, significantly reducing the risk of ignition in flammable gas environments.

[0121] All laser light sources have their optical paths completely enclosed within a sealed optical cavity, with no exposed beams, achieving Class 1 laser safety standards. The optical cavity maintenance hatch is equipped with a dual interlocking structure of mechanical lock and electrical power cut-off. Whenever the hatch is open, the system automatically cuts off power to all laser light sources, preventing maintenance personnel from being exposed to scattered laser light.

[0122] All external aviation plugs, signal terminals, and cable interfaces of the equipment use anti-corrosion sealed connectors and are equipped with waterproof and anti-corrosion protective sleeves to prevent electrolyte mist and humid air from penetrating and to prevent terminal oxidation and signal breakage.

[0123] To objectively and intuitively verify the actual detection performance and anti-interference capability of the battery adhesive seam overflow anti-reflective optical sensing detection method and device of the present invention, the present invention simulates production line conditions in the laboratory and selects a traditional single optical detection device as a comparison prototype.

[0124] To ensure the fairness and validity of the comparative experiments, all comparative tests adopted completely uniform experimental preconditions to eliminate the interference of differences in environment, samples, and equipment parameters on the experimental results. The specific uniform test parameters are as follows: 1. Test environment parameters: ambient temperature range 0℃~45℃, ambient humidity 40%~85%RH, simulating the normal temperature and humidity fluctuations in the workshop; equipped with normal transmission vibration of the production line (vibration amplitude 0.02mm~0.15mm); there are trace amounts of electrolyte mist and floating dust impurities to reproduce the real complex working conditions of mass production.

[0125] 2. Test Sample Configuration: Two types of samples were selected simultaneously: mainstream aluminum-cased hard-shell power batteries and aluminum-plastic film soft-pack energy storage batteries. For each type of battery, a gradient standard adhesive layer sample was prepared, covering nanoscale ultrathin adhesive layers (5nm, 20nm, 80nm), micron-scale conventional adhesive layers (1μm, 50μm, 200μm), and millimeter-scale overflow adhesive layers (0.5mm, 2mm, 5mm). At the same time, typical defect samples such as missing adhesive, thin adhesive, local overflow adhesive, and surface oil and impurities were also prepared. The total number of test samples was 1200pcs, of which 600pcs were qualified samples and 600pcs were defective samples.

[0126] 3. Comparison equipment: The control group was a traditional single-path optical vision adhesive overflow detection device (without anti-reflective optical path, without oxide layer compensation, without segmented model, and without working condition compensation mechanism); the experimental group was the optimized anti-reflective optical sensing detection device of this invention.

[0127] The quantitative comparative experimental data obtained are shown in the table below: The table below shows the quantitative performance comparison test data between the present invention and traditional single-optical-path detection technology.

[0128] Based on the above quantitative comparative test data, the performance advantages, technical mechanism gains, and adaptability of this invention compared with traditional technologies are objectively analyzed item by item: 1. Traditional single-optical-path detection technology is only suitable for a single thickness range and cannot be compatible with the detection of nano-scale ultrathin adhesive layers and millimeter-scale thick adhesive layers. Under conditions of high reflectivity and diffuse reflection, the optical signal is severely distorted, resulting in extremely large measurement errors. This invention adopts a segmented FTIR suppressed total internal reflection model + thin film transmission matrix dual model, combined with a dual-optical-path anti-reflection structure, to achieve seamless detection across the entire range of 0.5nm to 5mm. The detection accuracy of each range (nanometer, micrometer, and millimeter) is improved by more than 75%, solving the core defects of traditional technology such as poor range adaptability and imbalance of detection accuracy at high and low thicknesses.

[0129] 2. Traditional equipment lacks specific anti-reflection and anti-diffuse reflection structures. The specular reflection from hard-shell batteries easily causes signal saturation, while the diffuse reflection from soft-pack batteries easily causes signal disturbance, leading to frequent missed and false detections. This invention adapts a 42° oblique incidence suppressed total internal reflection optical path to hard-shell batteries and an 85° grazing incidence polarization differential optical path to soft-pack batteries. Combined with pre-positioned fluorescent impurity masking technology, this increases the defect detection rate of both hard and soft-shell batteries to over 99%, fundamentally solving the problems of false and missed detections.

[0130] 3. Traditional technologies lack vibration compensation, temperature compensation, and oxide layer phase compensation mechanisms. Under mass production conditions characterized by fluctuating workshop temperature and humidity, continuous production line vibration, oxide layer dispersion on the metal substrate, and light source aging and drift, detection accuracy rapidly declines and data dispersion is extremely high, failing to meet long-term mass production requirements. This invention effectively counteracts various operating condition interferences through multi-parameter closed-loop operating condition compensation, dual-wavelength oxide layer phase dynamic compensation, multi-dimensional vibration suppression, and a daily automatic calibration mechanism, improving data stability by over 85% across a wide temperature range and vibration conditions.

[0131] 4. Traditional techniques cannot distinguish between effective colloidal signals and surface impurity signals, and cannot compensate for systematic errors caused by the dispersion of the metal oxide layer, making batch misjudgments highly likely. This invention achieves pixel-level impurity shielding through time-division fluorescence impurity detection, combined with real-time calculation of oxide layer thickness and wavelength-by-wavelength phase compensation, significantly eliminating detection deviations caused by substrate differences and surface impurities.

[0132] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting excess adhesive at battery bonding seams using an optical sensor to reduce reflectivity, characterized in that, Employing a dual-optical-path adaptive switching sensing architecture, combined with multispectral time-sharing acquisition, pre-positioned impurity shielding, segmented optical modeling, and multi-condition dynamic compensation strategies, this system achieves anti-reflective, full-range, and high-precision online detection of adhesive overflow at bonding seams in different types of batteries. The process includes the following steps: Identify the type of battery to be tested, adaptively match the corresponding anti-reflective optical detection path, and complete the pre-positioning lock; A time-sharing, staggered optical excitation method is adopted to complete impurity fluorescence detection, metal substrate oxide layer detection, and main optical signal acquisition of the adhesive layer in stages. Real-time detection of oxide layer parameters on metal surfaces, and dynamic phase compensation of optical detection signals based on a preset calibration database; Collect multi-temporal optical sensing data of the adhesive layer, and combine it with working condition sensing information to complete signal correction and noise suppression; The real-time thickness of the adhesive layer is obtained by solving the adhesive layer thickness range in segments: For the 0.5nm~100nm ultrathin adhesive layer range, a modified FTIR suppressed total internal reflection model is used to solve for the thickness. A metal region of the same material, without adhesive or oxide layer differences in the vicinity of the joint to be tested is selected as the reference area, and the normalized ratio formula is used: In the formula: The normalized effective signal strength of suppressed total internal reflection; The real-time FTIR reflected signal intensity of the adhesive layer area to be tested at the battery assembly seam; The FTIR reference signal intensity of the reference metal region with no adhesive and the same material within the same field of view is used; through ratio normalization calculation, common mode is used to cancel air gap fluctuations, light intensity jitter and stray light interference, so as to achieve stable detection of nanoscale ultrathin adhesive layer; For medium-thickness adhesive layers ranging from 100nm to 5mm, a four-layer dielectric thin film transport matrix model is used to achieve full-range detection; the refraction angle of the colloidal medium is solved using Snell's law of refraction. In the formula: The refractive index of air is fixed at 1.

0. The angle of incidence of light; The equivalent refractive index of the battery encapsulating colloid; Let be the angle of refraction of light after it is incident on the colloidal medium. After solving for the angle of refraction of the colloidal medium, we combine the oblique incident optical path difference correction and oxide layer phase compensation to construct a four-layer medium transmission matrix of air-colloid-oxide-metal. We use the multi-wavelength reflectivity response to construct an overdetermined set of equations and solve for the optimal thickness value through the least squares iterative algorithm. By combining multi-time series detection data with the colloid curing kinetic model, the final overflow state after curing is extrapolated and predicted to complete defect judgment and production line traceability and diversion. Among them, the system has four types of curing kinetic models corresponding to commonly used lithium battery colloids: thermally conductive silicone double-exponential curing flow model, epoxy resin S-type curing flow model, UV adhesive exponential hardening flow model, and slow-curing structural adhesive delayed response model. All four models use h(t) as the measured thickness of the overflow adhesive at time t. To achieve the final stable thickness after complete curing, The initial adhesive coating thickness at time 0s is the basic variable, and its specific form is as follows: The bi-exponential curing flow model for thermally conductive silicone is suitable for sealant materials with good flowability, such as thermally conductive silicone. The curing process is divided into two stages: rapid spreading and slow curing shrinkage. The function expression is: In the formula: The range of values ​​represents the percentage of amplitude during the rapid flow phase. ; This represents the rate constant during the rapid spreading phase. This represents the rate constant for the slow curing shrinkage stage. The S-curing flow model for epoxy resins is applicable to self-catalytic epoxy resin sealing materials. The flow-curing process exhibits an S-shaped change characteristic of "slow-fast-slow," and the functional expression is as follows: In the formula: The inflection point time of the flow rate; It is the flow-to-solidification transition rate constant; The UV adhesive exponential curing flow model is suitable for UV adhesives that are fast-curing sealing materials. The curing rate is fast, and the thickness exhibits monotonically convergent exponential curing. The function expression is: In the formula: This represents the overall flow rate constant for solidification. The delayed response model for slow-curing structural adhesives is applicable to materials with a distinct curing induction period. During this period, the thickness change is minimal, followed by rapid and stable curing. The functional expression is: when... hour, ;when hour, In the formula: This is the curing induction period; is the curing flow rate constant after the induction period.

2. The battery bonding seam overflow anti-reflective optical sensing detection method according to claim 1, characterized in that, Two types of anti-reflective detection optical paths are configured to adapt to different battery surface characteristics. For the smooth metal surface of hard-shell batteries, a 42° oblique incidence suppressed total internal reflection optical path is adopted. The 42° oblique incidence angle is slightly larger than the critical angle of total internal reflection at the air interface of BK7 optical glass, and s-polarized light is used as the unified incident reference to stably excite the evanescent wave and thin film interference signal. For the rough aluminum-plastic film surface of soft-pack batteries, an 85° grazing incidence polarization differential optical path is adopted. The polarization and detection combination structure suppresses diffuse reflection stray light and specular reflection interference to extract the effective adhesive layer sensing signal.

3. The battery assembly seam overflow anti-reflective optical sensing detection method according to claim 1, characterized in that, The time-sharing and staggered optical excitation method includes three independent time-series working modes, with no overlapping light sources in each time sequence, to avoid detection deviations caused by premature curing of the colloid due to heat. In the first time sequence, the fluorescent light source is turned on, and oil stains and dust impurities are identified based on the three-dimensional features of fluorescence intensity, spectral shift and decay rate. Impurity mask areas are generated and shielded at the pixel level. In the second time sequence, the fluorescent light source is turned off, the dual-path ultraviolet light source is turned on, and the reflectivity data of the metal substrate is collected for the calculation of the oxide layer thickness. In the third time sequence, the ultraviolet light source is turned off, and multiple visible and near-infrared main detection light sources are turned on to collect optical signals of adhesive layer interference and reflection.

4. The battery assembly seam overflow anti-reflective optical sensing detection method according to claim 3, characterized in that, The thickness of the alumina oxide layer on the metal surface is calculated in real time using the dual-wavelength ultraviolet differential method. The calculation formula is as follows: ； In the formula: The thickness of the aluminum oxide layer on the battery metal surface; The extinction coefficient of the alumina medium at a UV wavelength of 275 nm; The extinction coefficient of the alumina medium at a UV wavelength of 255 nm; The reflectance at a wavelength of 255nm is collected in real time for the area to be tested; The standard reflectance of a pure metal reference surface without an oxide layer at a wavelength of 255 nm; The reflectance at a wavelength of 275nm is collected in real time for the area to be tested; The standard reflectance of a pure metal reference surface without an oxide layer is measured at a wavelength of 275 nm. Based on the measured oxide layer thickness, the wavelength-oxide layer thickness-phase shift correspondence is retrieved, and independent phase compensation is performed on each main detection light source to eliminate the optical detection system error caused by the discreteness of the oxide layer on the metal substrate.

5. The method for detecting excess adhesive at battery bonding seams and its anti-reflective optical sensing properties according to claim 1, characterized in that, In the step of combining working condition sensing information to complete signal correction and noise suppression, multi-dimensional composite vibration suppression is performed: horizontal and vertical dual-dimensional vibration data are collected synchronously by multi-directional vibration sensors; real-time pixel position correction is performed on the imaging pixel shift caused by horizontal vibration; the air gap is finely adjusted by the laser ranging closed-loop mechanism linked to vertical vibration; phase-locked synchronous sampling is used to cancel interference for steady-state periodic vibration of the production line; and a multi-frame image optimization and distortion frame rejection mechanism is used to select effective and clear frames for subsequent calculations for instantaneous impact vibration of workpiece transfer.

6. The battery assembly seam overflow anti-reflective optical sensing detection method according to claim 1, characterized in that, In the step of combining working condition sensing information to complete signal correction and noise suppression, full-link temperature adaptive compensation is performed: four types of temperature data are collected in real time by multi-point temperature sensors: the internal temperature of the optical cavity, the surface temperature of the partitioned prism, the temperature of the light source chip, and the ambient temperature of the workshop. The segmented compensation coefficients of the corresponding temperature range are retrieved, and the detection errors caused by the colloidal refractive index, prism optical parameters, and light source wavelength drift are corrected simultaneously to achieve full-temperature adaptive compensation from 0℃ to 45℃.

7. The battery assembly seam overflow anti-reflective optical sensing detection method according to claim 1, characterized in that, Equipped with a full-range automatic calibration mechanism, the equipment automatically completes multi-gradient standard sample calibration at 0nm, 1nm, 10nm, 100nm, 1μm, and 1mm every day upon power-up, covering the full detection range of nanometers, micrometers, and millimeters. It automatically iterates and updates the optical reference and algorithm coefficients, ensuring the consistency and stability of the equipment's detection accuracy over the long term.

8. A high-precision detection device for excess adhesive at battery bonding seams, characterized in that, The device for performing the detection method according to any one of claims 1-7 includes a rack support module, a dual-mode independent optical path switching module, a multi-band time-division light source excitation module, an anti-cross-light partitioned optical imaging module, a multi-parameter closed-loop working condition compensation module, a main control computing module, and an explosion-proof and corrosion-proof safety protection module. The rack-mount module is internally physically isolated to form an optical cavity, an electrical cavity, a main control cavity, and a cleaning cavity; The dual-mode independent optical path switching module integrates the 42° suppressed total internal reflection optical path of the hard-shell battery and the 85° polarized grazing incidence anti-reflection optical path of the soft-pack battery, realizing battery category-based pre-switching positioning. The multi-band time-sharing light source excitation module is equipped with multiple independent light sources including fluorescence, ultraviolet, and visible and near-infrared, and has independent temperature control and time-sharing peak emission control logic. The anti-crosstalk partitioned optical imaging module uses an integrated BK7 partitioned prism with light-shielding and light-blocking ribs to achieve nano-FTIR detection and interference angle detection in partitioned sections. The multi-parameter closed-loop operating condition compensation module integrates vibration sensing, temperature sensing, and laser ranging closed-loop adjustment structure to achieve vibration correction, temperature compensation, and dynamic stability control of air gap. The explosion-proof and corrosion-resistant safety protection module integrates cavity explosion-proof sealing, laser opening interlock, and external interface corrosion-resistant sealing structure, meeting the explosion-proof safety regulations for lithium battery workshops.

Images