A method and apparatus for precision assembly of a micro-collimating lens and an optical fiber

Abstract

The application discloses a precision assembly method and device for a micro-collimating lens and an optical fiber, and the method comprises the following steps: determining an optimal alignment axial direction and positioning an optimal coupling position by detecting an outgoing light field asymmetric vector zero-crossing locus; combining a glue point side surface morphology to extract a solidification shrinkage main direction and amplitude, adding a reverse pre-offset to the optimal alignment position to determine a glue application reference and establish a glue point distribution scheme; continuously monitoring a light spot centroid drift rate change in an ultraviolet solidification process, identifying a transition moment when glue changes from a flowing state to a solidified state, and outputting a transient compensation instruction according to a current deviation and a step margin in a soft glue adjustable window; tracking a stress accumulation rate of each axis servo drive in a solidification later stage, identifying a main stress axial direction, calculating a residual displacement correction amount, outputting a precision assembly instruction to complete final position correction, and improving assembly position consistency between batches.

Description

Technical Field

[0001] This invention relates to the field of precision assembly technology for optoelectronic devices, and in particular to a method and apparatus for the precision assembly of a miniature collimating lens and an optical fiber. Background Technology

[0002] The precision assembly of miniature collimating lenses with optical fibers is a critical process in the production of optical communication modules and laser devices. The directionality and coupling efficiency of the emitted beam are extremely sensitive to axial orientation deviations, and additional errors introduced during assembly can lead to irreversible degradation of the device's optical performance. Existing assembly methods involving active alignment followed by UV adhesive curing have two core problems: First, the volume shrinkage during adhesive curing can push the lens away from its aligned position. The direction and magnitude of the shrinkage are influenced by the geometry of the adhesive layer and the constraints of the surrounding structure, resulting in significant batch-to-batch variations that are difficult to compensate for in advance with a uniform offset. Second, the time window for intervention in the transition of the adhesive from a fluid state to a cured state is extremely short. Existing processes lack online identification methods for this transition, making it impossible to effectively utilize the position adjustment opportunities during the soft adhesive stage. After curing, residual deviations can only be compensated for through repeated fine-tuning based on manual experience, making it difficult to guarantee batch consistency.

[0003] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the Invention

[0004] This invention discloses a precision assembly method and apparatus for a miniature collimating lens and an optical fiber. The optimal alignment axis is established by quantifying the zero-crossing trajectory of the asymmetric vector of the outgoing light field. The side morphology of the adhesive dots is transformed into a shrinkage pre-compensation basis, and the adhesive application reference is determined by the reverse pre-bias amount. The adhesive dot distribution scheme is established. During the curing process, the adhesive transition time is identified and online compensation is implemented in the soft adhesive stage. In the later stage of curing, the residual correction amount is calculated based on the stress accumulation direction. This realizes a complete assembly closed loop from alignment search, shrinkage pre-compensation to adaptive control throughout the curing process.

[0005] The first aspect of this invention provides a method for the precise assembly of a miniature collimating lens and an optical fiber, comprising the following steps:

[0006] Collect far-field spot data and glue dot morphology data. Based on the far-field spot data, identify the moment when the asymmetric vector direction of the spot reverses and generate a misalignment crossing record. Based on the glue dot morphology data, analyze the side profile center offset direction to determine the glue dot shrinkage estimate.

[0007] Based on the misalignment crossing record, the boundary of the asymmetric vector reversal region is identified to form a zero crossing axis. The optimal alignment pose is obtained by fitting the position of maximum coupling efficiency in the zero-crossing scan using the zero crossing axis.

[0008] The optimal alignment pose is used to implement the glue dot shrinkage prediction amount reverse pre-bias compensation to determine the pre-bias compensation pose. The pre-bias compensation pose is used to verify the light spot index tolerance to obtain the pre-bias verification result. The pre-bias verification result is used to determine the multi-point symmetrical glue application position and establish the glue dot distribution scheme.

[0009] Based on the glue dot distribution scheme, the real-time offset of the centroid of the UV curing spot is recorded to generate a curing drift record. The curing drift record is used to identify the moment of sudden increase in offset rate to obtain the gelation initiation parameter set. The offset that can be compensated in the soft glue state is calculated from the gelation initiation parameter set, and a transient compensation command is output.

[0010] Based on the transient compensation command, the slope of the servo drive current increment of each axis in the later stage of curing is analyzed to form a curing stress record. The curing stress record is used to identify the axis of sudden increase in current increment slope to obtain the principal stress axis. The axial residual displacement correction amount is calculated from the principal stress axis and a precision assembly command is output.

[0011] A second aspect of the present invention provides a precision assembly device for a miniature collimating lens and an optical fiber, comprising:

[0012] The data acquisition module is used to acquire far-field spot data and glue dot morphology data. Based on the far-field spot data, it identifies the moment when the asymmetric vector direction of the spot reverses and generates a misalignment crossing record. Based on the glue dot morphology data, it analyzes the offset direction of the side contour center to determine the estimated amount of glue dot shrinkage.

[0013] The alignment search module is used to identify the boundary of the asymmetric vector reversal region based on the misalignment crossing record to form a zero crossing axis, and to obtain the optimal alignment pose by fitting the position of the maximum coupling efficiency in the zero-crossing scan using the zero crossing axis.

[0014] The pre-biased gluing module is used to perform reverse pre-biased compensation on the estimated amount of glue dot shrinkage in the optimal alignment pose to determine the pre-biased compensation pose, verify the compliance of the light spot index with the pre-biased compensation pose to obtain the pre-biased verification result, and determine the multi-point symmetrical gluing position and establish the glue dot distribution scheme through the pre-biased verification result.

[0015] The curing monitoring module is used to collect the real-time offset of the centroid of the UV curing spot according to the glue dot distribution scheme to generate a curing drift record. The curing drift record is used to identify the moment of sudden increase in offset rate to obtain the gelation initiation parameter set. The offset that can be compensated in the soft glue state is calculated from the gelation initiation parameter set and a transient compensation command is output.

[0016] The stress quality control module is used to analyze the slope of the servo drive current increment of each axis in the later stage of curing based on the transient compensation command to form a curing stress record. The curing stress record is used to identify the axis of sudden increase in current increment slope to obtain the principal stress axis. The axial residual displacement correction amount is calculated from the principal stress axis and a precision assembly command is output.

[0017] The beneficial effects of this invention are reflected in the following points: 1. The optimal alignment axis is determined by detecting the zero-crossing trajectory of the outgoing light field asymmetry vector during the scanning process, and each crossing coordinate is weighted by the light field crossing depth at the crossing point, so that the axis estimation is not affected by low-quality crossing points; on this basis, a coupling efficiency fine scan is performed along the optimal axis, and the peak position is fitted by a Gaussian curve instead of directly taking the sampling extreme value, and the positioning accuracy exceeds the sampling interval limit. The above two steps together make the direction of the alignment search path and the coordinates of the alignment position based on quantitative quality assessment, avoiding the experience dependence and insufficient repeatability problems of manual judgment of the optimal point by single-axis reciprocating scanning in traditional schemes. 2. The weighted center offset of the height profile of the glue dot side and the volume principal inertial axis deviation angle jointly describe the main direction of curing shrinkage. The correction factor of the constraint of adjacent glue dots and the asymmetry of the wetting boundary further adjusts the morphology analysis results to the true constrained direction. The shrinkage components of each degree of freedom are converted and superimposed on the optimal alignment coordinates by the reverse pre-bias. Pre-bias compensation is completed before adhesive application rather than relying on post-curing adjustments, thus preemptively offsetting the displacement effect of curing shrinkage during the adhesive layer forming stage. This avoids the problem in traditional processes where post-curing fine-tuning is a necessary step due to the unpredictable shrinkage direction. The verification of the feasible pre-bias range also ensures that the spot index still meets the process tolerance under bias conditions. 3. The UV curing process continuously tracks the acceleration of the spot centroid drift rate, identifying the first sudden increase moment when the adhesive enters the curing transition stage. Within the adjustable window of the soft adhesive, the compensable coefficient is calculated based on the instantaneous deviation and single-step margin, and the position correction amount is output, allowing online intervention for the drift caused by curing shrinkage while the adhesive layer is still fluid. The sudden increase direction of the servo current increment slope of each axis in the later stage of curing reveals the stress distribution weight in each degree of freedom. The principal stress axial direction and residual centroid deviation are jointly used to calculate the correction amount and execute the final position correction at the allowable rate. The two-stage compensation mechanism covers the complete displacement evolution process from soft adhesive to full curing, and the handling of residual alignment errors after curing no longer relies on repeated manual fine-tuning. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a precision assembly method for a miniature collimating lens and an optical fiber according to the present invention.

[0019] Figure 2 This is a structural block diagram of a precision assembly device for a miniature collimating lens and an optical fiber according to the present invention. Detailed Implementation

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.

[0021] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0022] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0023] The technical solutions of the embodiments of this application will be described below.

[0024] like Figure 1 As shown, this embodiment of the invention provides a method for the precise assembly of a miniature collimating lens and an optical fiber, including the following steps S101-S105:

[0025] Step S101: Collect far-field spot data and glue dot morphology data; based on the far-field spot data, identify the moment when the asymmetric vector direction of the spot reverses and generate a misalignment crossing record; and based on the glue dot morphology data, analyze the offset direction of the side contour center to determine the estimated amount of glue dot shrinkage.

[0026] Specifically, far-field spot data and adhesive dot morphology data are acquired. Far-field spot data is acquired frame-by-frame on the far-field plane of the collimating lens's output side using a CCD camera. The frame rate must be higher than the displacement update frequency of the motion platform during alignment scanning, ensuring that each scanning step corresponds to at least one complete frame of far-field spot data. The spot images are stored in 16-bit grayscale format, preserving complete intensity gradient information. Oversaturated areas are avoided during acquisition through automatic exposure adjustment; if the core area overflows, the intensity center of gravity will be distorted. The intensity distribution pattern of each frame of far-field spot data directly reflects the current alignment state between the collimating lens and the fiber axis: when the axis is perfectly aligned, the spot is circularly symmetrical; when the axis is deflected, the spot is stretched into an ellipse along the deflection direction. The major axis angle of the ellipse has a monotonic correspondence with the deflection direction. The spot asymmetry vector is the core sensing quantity for subsequent misalignment and crossing detection. Adhesive dot morphology data were acquired immediately after adhesive application using a side-mounted optical microscope. The side image covered the entire height of the adhesive dot and the area in contact with the bottom surface. The acquisition time of the adhesive dot morphology data must fall within the stable window after the adhesive flow has stopped. If the acquisition is too early, the adhesive dot is still leveling and the morphology is unstable; if it is too late, local curing has begun under UV pre-irradiation and the morphology has been deformed. The determination method is to determine that the change in the real-time centroid coordinates of the adhesive dot outline in the side image does not exceed 0.4 pixels over three consecutive frames. The acquisition sequence of far-field spot data and adhesive dot morphology data is independent: far-field spot data is continuously acquired throughout the alignment scan, while adhesive dot morphology data is acquired only once after adhesive application. The two are associated with the assembly batch number.

[0027] In some embodiments, the step of identifying the moment of reversal of the asymmetric vector direction of the light spot based on the far-field light spot data to generate an inaccurate crossing record includes: extracting the elliptical direction angle of the light spot frame by frame from the far-field light spot data to obtain a sequence of light spot direction angles; identifying the moment when the positive and negative signs of the direction angles alternate in the sequence of light spot direction angles to determine a set of reversal moments; establishing crossing position coordinates by recording the six-dimensional coordinates of the motion platform at each moment through the set of reversal moments; and generating an inaccurate crossing record by summarizing the direction angle reversal amplitude of each crossing point based on the crossing position coordinates.

[0028] The beam direction angle sequence is obtained by extracting the beam ellipse direction angle of each frame from the far-field beam data. The intensity matrix of each frame of the far-field spot data is used to calculate the ellipse fitting parameters of the spot using the second-order central moments. The direction angle θ is obtained by the following formula: θ = (1 / 2) × arctan2(2·M11,M20-M02), where M20 and M02 are the horizontal and vertical second-order central moments of the spot intensity distribution, respectively, M11 is the cross second-order central moment, and arctan2 is the two-parameter arctangent (quadrant determined by the signs of the numerator and denominator). The value range of θ is limited to (-90°, 90°). When θ is positive, it indicates that the major axis of the ellipse is biased in the positive direction; when θ is negative, it indicates that the major axis is biased in the negative direction. When the sign of θ crosses zero, the corresponding spot tends to be circularly symmetrical, that is, the collimating lens axis approaches the optimal alignment direction. The calculation of the second-order central moments requires high uniformity of background noise in the far-field spot data. When the background intensity distribution is uneven, the contribution of low-intensity areas to M11 is biased by the background offset, resulting in a systematic shift of θ in the spot direction angle sequence. Therefore, the far-field spot data... Before moment calculation, intensity threshold segmentation is performed. Pixels with intensity below the maximum intensity set for the entire frame are set to zero before participating in moment calculation. The threshold ratio is determined during the initialization phase based on the average background noise during standard component testing. Each frame's θ is arranged according to its frame number to form a spot orientation angle sequence. The difference between adjacent frames' θ in the spot orientation angle sequence reflects the rate of change of the spot asymmetry vector with platform movement. A stable rate of change corresponds to a uniform platform scanning condition, while a sudden change in rate of change usually corresponds to a jump in scanning step size or platform vibration interference. When the difference between M20 and M02 approaches zero, the denominator of the arctan function approaches zero. The θ calculation result is extremely sensitive to small perturbations in the numerator. Frames with a near-circular symmetry label are added to the spot orientation angle sequence for frames whose absolute denominator value is below the set minimum moment difference threshold. The θ value of the near-circular symmetry labeled frames does not participate in the main sequence for sign switching determination, but is replaced by interpolation from surrounding frames at the current platform position. This prevents unstable θ calculations when the spot ellipticity is extremely low from disrupting the reversal time identification.

[0029] The set of reversal moments is determined by identifying the alternating positive and negative signs of the θ sign in the beam direction angle sequence. The frame positions where the θ sign changes from positive to negative or vice versa correspond to the moments when the collimating lens axis crosses the optimal alignment axis during scanning. These crossing moments are extracted sequentially to form the basic candidate set of reversal moments. Simply judging crossings based on sign changes carries the risk of false triggering. When the θ value of the beam direction angle sequence oscillates frequently near zero due to noise, the signs of adjacent frames may switch repeatedly within a short period without the axis actually crossing. Therefore, the set of reversal moments requires that the average sign of θ in each of the three consecutive frames before and after the switch be consistent, and that the switching amplitude be no less than 5°. Only when both conditions are met is the set included, thus filtering out false crossings. On precision assembly lines, motion platforms typically advance unidirectionally with fixed step lengths. During normal unidirectional scanning, the beam direction angle sequence only undergoes one sign switch. Platform homing errors or operator adjustments to the scanning range can cause multiple switches. In such cases, the reversal time set uses the reversal amplitude of the direction angle at each crossing point (the absolute value of the difference between the average θ values ​​of three frames before and after the switch) for identification. Crossing points with larger reversal amplitudes correspond to axis depth crossings and are considered true crossings, while those with smaller reversal amplitudes are marked with low confidence. If a near-circular symmetrical frame falls within the switching decision window, the reversal time set extends the switching window boundary outwards to skip the near-circular symmetrical frame. The average of the effective frames on both sides of the near-circular symmetrical frame is then used to determine sign consistency. Otherwise, the near-circular symmetrical interval will block the switching signal of a true crossing, resulting in missed detections.

[0030] Crossing position coordinates are established by recording the six-dimensional coordinates of the motion platform at each moment using a reversed time set. After aligning the frame number of each crossing moment in the reversed time set with the timestamp in the motion platform log, the X, Y, and Z displacements and three-axis rotation angles of the platform at that time are read. The six-dimensional coordinates together constitute the spatial state description of the crossing point. The six-dimensional coordinates of all crossing points in the reversed time set are arranged in the order of crossing moments to form the crossing position coordinates. The timestamp accuracy of the motion platform log must match the frame rate of the far-field spot data acquisition. If the accuracy is insufficient, the platform coordinate reading points at each moment in the reversed time set may fall between two scan steps. If the alignment error exceeds half a scan step, a timing deviation label is added to the corresponding entry of the crossing position coordinates. The coordinates of the crossing points with timing deviation labels are downweighted when summarizing inaccurate crossing records. The rotation component in the crossing position coordinates reveals the residual deviation between the platform attitude and the ideal alignment attitude when the axis crosses the optimal position. Optimal alignment of the lens axis sometimes requires a combination of lateral displacement and pitch angle. In alignment motions coupled with deflection and displacement, attitude information cannot be replaced by pure displacement coordinates. If the crossing position coordinates only record the displacement component and discard the rotation component, the two types of crossing—displacement crossing and attitude crossing—cannot be separated. When changes in the assembly environment temperature cause thermal drift in the optical path, the crossing position coordinates of two consecutive crossings within the same batch exhibit systematic monotonic drift in the Z-axis direction. The monotonicity of the Z-axis coordinates of consecutive crossing points can serve as an early detection signal for thermal drift. When the drift exceeds the tolerance (e.g., 0.2 micrometers), a thermal drift warning label is added to the crossing position coordinates, prompting the operator to pause scanning. After the optical path reaches thermal equilibrium, far-field spot data is reacquired, and the crossing coordinates are valid only after resuming scanning.

[0031] Inaccurate crossing records are generated by summarizing the directional angle reversal amplitude of each crossing point based on the crossing position coordinates. For each crossing point, the average θ value of the spot directional angle sequence frame is taken for three frames before and after the switch. The absolute value of the difference between the two averages is used as the directional angle reversal amplitude of that crossing point. A larger reversal amplitude indicates a greater crossing depth when the spot asymmetric vector crosses zero, and a higher confidence level that the crossing event is a true crossing. The geometric relationship between amplitude and axial positioning accuracy is handled separately in the subsequent weighting stage of the spatial distribution of crossing points; each dimension has its own function. The six-dimensional coordinates of each crossing point and the corresponding reversal amplitude together constitute a complete description of each crossing event in the inaccurate crossing record. The reversal amplitude of the crossing point marked by the time sequence deviation in the crossing position coordinates is included in the inaccurate crossing record summary calculation as a weighting coefficient. The weighting coefficient is inversely proportional to the time sequence alignment error; this is the only weighting for the time sequence deviation crossing point. When multiple devices at adjacent workstations operate simultaneously, vibrations are transmitted via ground coupling. The scanning platform experiences brief displacement jitter near the crossing point. During this jitter, the beam direction angle sequence θ within the mean window before and after the crossing is affected by θ offset caused by misalignment motion, resulting in an overestimation of the reversal amplitude. For crossing points where the standard deviation of the displacement component of the crossing position coordinates within the mean window before and after the crossing point exceeds a vibration threshold (e.g., 0.1 micrometers), a jitter correction label is added. The reversal amplitude of these crossing points is replaced by the correction amplitude after deducting the displacement standard deviation and the equivalent θ offset, thus preventing the overestimation from being carried over to the subsequent zero-crossing axis positioning. Batches with consistently low reversal amplitudes in the crossing records are mostly due to excessively large scanning step sizes and axes crossing the optimal alignment zone with large step sizes. These batches are marked with a coarse step size warning in the record, prompting the operator to narrow the step size in the next scanning batch, thereby improving the positioning resolution of the crossing position.

[0032] In some embodiments, determining the estimated amount of adhesive dot shrinkage based on the side profile center offset direction analyzed from the adhesive dot morphology data includes: extracting the side height profile and bottom contact line coordinates from the adhesive dot morphology data to generate a side profile distribution; calculating the profile height plus the weighted center coordinates of the side profile distribution to obtain the height center offset; analyzing the volume distribution asymmetry axis deviation angle based on the height center offset to establish a shrinkage feature set; and calculating the maximum asymmetry amplitude of the side height based on the shrinkage feature set to obtain the estimated amount of adhesive dot shrinkage.

[0033] Side profile distribution is generated by extracting the side height contour and bottom contact line coordinates from adhesive dot morphology data. After image enhancement and edge detection, the pixel coordinate sequence of the upper contour line of the adhesive dot and the contact line coordinate of the bottom contact area with the substrate are extracted. Aligning these two coordinates in the same pixel coordinate system constitutes the basic geometric description of the side profile distribution. The side height contour is determined by the ordinate of the highest adhesive pixel in each horizontal pixel column, and the bottom contact line is determined by the lateral range of the lowest adhesive surface in contact with the substrate. During the leveling process after adhesive application, the contact line expands outward. The width of the contact line in the side profile distribution reflects the final leveling range. A narrow width usually indicates insufficient adhesive, while an excessively wide width indicates excessive adhesive and a risk of adhesive creep. When the contact line width exceeds the upper limit of the process specification, an over-specification is recorded in the corresponding entry of the side profile distribution. This annotation triggers contact constraint area correction during the constraint correction factor calculation stage, thus accurately reflecting the constraint effect of the actual wetting boundary on the shrinkage direction. If the side view angle of the glue dot morphology data deviates from the normal of the glue dot, the horizontal projection of the height profile will introduce a systematic proportional error. The tilt deviation is measured during the camera calibration stage and written into the configuration parameters as a correction factor. The coordinates of each contour in the side contour distribution are converted into actual dimensions after being corrected by the correction factor and then used in the subsequent centroid calculation, thus unifying the physical dimensions of the spatial coordinates. For local segments in the glue dot morphology data where the edge detection sub-pixel accuracy is worse than 0.3 pixels, low-precision annotations are added within the side contour distribution. The contour coordinates of the low-precision annotation segments are replaced by extrapolation interpolation of adjacent high-precision segments to prevent local detection deviations from being mixed into the subsequent centroid statistics as jump points.

[0034] The height center of gravity offset is obtained by calculating the weighted center coordinates of the side profile distribution. The profile height value h(x) at each horizontal position x is used as the weight for the weighted center of gravity calculation. The difference between the horizontal weighted mean X_c and the nominal center of symmetry X_ref constitutes the horizontal height center of gravity offset, calculated by the following formula: X_c=Σ[x×h(x)] / Σh(x), ΔX=X_c-X_ref, where X_c is the horizontal coordinate of the weighted center of gravity of the side profile distribution; h(x) is the profile height value corresponding to position x; X_ref is the coordinate of the nominal center of symmetry of the adhesive dot, taken as the midpoint between the two ends of the bottom contact line; ΔX is the horizontal height center of gravity offset, a positive value indicates that the height center of gravity is biased towards the positive X direction, and a negative value indicates that it is biased towards the negative X direction. Positions where h(x) is zero are not included in the weighted calculation to prevent the asymmetrical expansion of the contact line from interfering with the weighted center of gravity. In the side profile distribution, the h(x) of the low-precision labeled segment is used in the calculation with the interpolated substitute value, but the weight of the corresponding position is reduced according to the interpolation confidence in the X_c calculation to ensure that the influence of the low-precision segment on the height centroid offset is appropriately suppressed. When adhesive dots are applied within the narrow gap between the optical fiber and the lens assembly, lateral capillary forces cause the dots to concentrate towards the side with the smaller gap. The side profile distribution in this scenario exhibits a noticeable unilateral height bulge, with the direction of ΔX corresponding to the side with the smaller gap. After curing and shrinkage, the collimating lens will shift in the ΔX direction. The height centroid offset is simultaneously recorded as the vertical centroid offset ΔZ, calculated using the following formula: Z_c = Σ[z×b(z)] / Σb(z), ΔZ = Z_c - Z_ref, where Z_c is the vertical centroid coordinate weighted by the width b(z), b(z) is the width of the inner contour of the contact line corresponding to position z (the sign b is used to distinguish it from the weight w of the crossing point later), and Z_ref is the ordinate of the nominal symmetry center of the adhesive dot. ΔX and ΔZ together constitute a complete planar centroid offset vector for subsequent volume asymmetry axis analysis.

[0035] For example, the step of establishing a shrinkage feature set based on the deviation angle of the asymmetric axis of the volume distribution for the height center of gravity offset includes: calculating the deviation angle of the principal inertial axis of the volume distribution for the height center of gravity offset to obtain the volume offset principal axis; using the volume offset principal axis to calculate the shrinkage constraint weights of adjacent glue points and wetting boundaries to form a constraint correction factor; adjusting the shrinkage principal axis from the offset direction to the constrained direction through the constraint correction factor to construct the offset shrinkage axis; and deriving the direction cosine and amplitude of the shrinkage components of each degree of freedom based on the offset shrinkage axis to obtain the shrinkage feature set.

[0036] The volume offset principal axis is obtained by calculating the deviation angle of the principal inertial axis of the volume distribution based on the height centroid offset. The height centroid offset determines the offset vector of the volume centroid of the adhesive point relative to the nominal center of symmetry. Using the contour height of each position on the side contour distribution as the area weight, a second-order moment matrix is ​​calculated around the centroid position indicated by the height centroid offset. Eigenvalue decomposition is performed on the matrix, and the direction of the eigenvector corresponding to the largest eigenvalue is the direction of the volume distribution principal axis. The angle between this direction and the assembly axis is defined as the deviation angle α. After normalization, the eigenvector constitutes the direction description of the volume offset principal axis. Eigenvalue decomposition requires high accuracy of the contour coordinates. Low-precision labeled sections are weighted and reduced when participating in moment calculation. The difference in eigenvalues ​​of the matrix after reduction is sometimes lower than the resolution threshold. In this case, the volume centroid offset direction revealed by the height centroid offset is still relatively reliable, but the principal axis direction cannot be reliably decomposed from the matrix. In this case, a low-resolution label is added to the volume offset principal axis, and the direction of the line connecting the asymmetry centers of the bottom contact line is used to replace the principal inertial axis direction, using contact geometry information to fill the resolution gap of the volume moment analysis. When the amount of adhesive is too small and the height of the adhesive dots is too low, the number of effective height points decreases, weakening the statistical basis of moment calculation and increasing the directional uncertainty of the volume offset principal axis. For adhesive dots with fewer than 30 effective height points, the volume offset principal axis is labeled with sparse contours. During the constraint correction factor calculation stage, these sparsely labeled adhesive dots are supplemented with historical constraint correction data from neighboring adhesive dots. This ensures that the sparse contours of individual points do not drag down the overall directional estimation of the shrinkage feature set. The direction vector of the volume offset principal axis is described in the XZ plane by both the deviation angle α and the normal component. When the normal component is too large, adhesive dot morphology data in the frontal direction needs to be collected and verified and corrected along with the height centroid offset.

[0037] A constraint correction factor is formed by calculating the shrinkage constraint weights of adjacent adhesive dots and the wetting boundary using the volume offset principal axis. Taking the most common left-right symmetrical two-point adhesive application scheme as an example: if the shrinkage eccentricity of the current adhesive dot happens to point to the right adhesive dot, the reverse constraint force formed when the right adhesive dot cures will pull the shrinkage path towards the line connecting the two points. The closer the volume offset principal axis is to the line connecting adjacent adhesive dots, the more significant this shrinkage principal axis deflection effect. The wetting boundary is the outer edge of the contact area between the bottom surface of the adhesive dot and the substrate. When the shape of the wetting boundary is asymmetrical, the lateral component of the bottom adhesion force constrains the shrinkage direction. The constraint correction factor is calculated by multiplying the cosine of the angle between the direction of the line connecting adjacent adhesive dots and the direction of the volume offset principal axis by the attenuation weight of the distance between adjacent adhesive dots (the closer the line direction is to the shrinkage principal axis and the closer the distance, the stronger the constraint), plus the normalization index of the degree of asymmetry of the wetting boundary. The sum of the two is normalized to form the quantitative value of the constraint correction factor. For adhesive dots with excessive adhesive content, the wetting boundary term recalculates the constraint area based on the measured contact line width, thus incorporating the additional constraint on the shrinkage direction from the excessively extended wetting range. For adhesive dots with sparse contour markings, the constraint correction factor is not calculated from their own contours; instead, it directly uses the historical constraint correction average of adjacent adhesive dots in the same batch to fill the gap. The constraint correction factor for adjacent adhesive dots is dynamically weighted according to the curing state of each adhesive dot in the adhesive application process sequence, with synchronous curing scenarios having the highest weight. Adjacent adhesive dots that have already cured participate in the constraint calculation based on their actual positions after curing and shrinkage. When the substrate surface roughness is uneven, local pinning occurs at the wetting boundary. The constraint strength at the pinning location is significantly higher than in smooth areas. The wetting boundary term of the constraint correction factor detects the sharp indentation features of the contact line. If pinning features are detected, the constraint weight at the corresponding boundary position is increased, thus providing a more accurate representation of the non-uniformity of the actual bottom surface constraint distribution by the volume offset principal axis.

[0038] The shrinkage principal axis is adjusted from the offset direction to the constrained direction by a constraint correction factor to construct an offset shrinkage axis. The original direction of the volume offset principal axis rotates under the action of the constraint correction factor. The rotation angle is the directed angle of the cross product of the constraint direction and the volume offset principal axis direction. The direction of the rotated axis is the direction of the offset shrinkage axis, which represents the main direction of the final shrinkage displacement of the adhesive dot under the actual curing constraint conditions. When the constraint correction factor is close to zero, the offset shrinkage axis and the volume offset principal axis almost coincide. The curing constraint has a weak intervention on the shrinkage direction, and the dominant shrinkage direction is still determined by the volume offset. When the constraint correction factor is large, the offset shrinkage axis deviates significantly from the volume offset principal axis. The constraint force of the adjacent adhesive dot or the wetting boundary has substantially torn the shrinkage direction. If the volume offset principal axis direction is still used to replace the offset shrinkage axis in subsequent pre-offset compensation, the compensation direction will deviate from the actual shrinkage path and introduce a non-negligible residual error. When multiple adhesive dots are cured simultaneously, the constraint correction factor of each adhesive dot needs to be calculated iteratively. Taking a four-corner symmetrical distribution as an example, after the bias shrinkage axis of each point is independently estimated in the first round, the direction of the already corrected shrinkage axis of the adjacent points is included in the constraint force recalculation in the second round. The spatial distribution of the constraint force gradually converges to the curing equilibrium state. In practice, the direction set usually stabilizes after 3 to 5 rounds. The iteration termination condition is that the sum of the changes in the bias shrinkage axis direction angle of each adhesive dot in two adjacent rounds is less than 0.5°. The iteration does not converge, which is often seen when the constraint correction factors of two symmetrically distributed adhesive dots cancel each other out and the direction angle oscillates back and forth between the two extremes. In such cases, the result of the odd number of iterations shall be taken as the standard, and an iteration instability mark shall be left on the bias shrinkage axis construction result.

[0039] The shrinkage feature set is obtained by deriving the direction cosines and amplitudes of the shrinkage components for each degree of freedom based on the offset shrinkage axis. The projection components of the unit direction vector of the offset shrinkage axis onto the X, Y, and Z axes are the direction cosines of each degree of freedom. The direction cosines and the asymmetric amplitude estimation calculated from the side profile are retained separately. The direction cosines and amplitude estimation together constitute the core content of the shrinkage feature set. If the offset shrinkage axis has a component of rotation about the assembly axis, the corresponding rotational degree of freedom will have angular displacement shrinkage. The rotational component is converted into angular displacement shrinkage by multiplying the projection of the offset shrinkage axis direction vector in the tangential direction by the equivalent force arm length, and written into the rotational component field of the shrinkage feature set. The direction cosine components with strict assembly tolerance requirements are highlighted separately in the shrinkage feature set. The highlighted components are given priority in being included in the feasible interval constraints during the subsequent pre-offset compensation pose determination stage. For biased shrinkage axes with iterative instability annotations, the corresponding shrinkage feature set components are supplemented with the statistical average of the historical shrinkage of similar adhesive dots along that axis. The historical average sample size must exceed 20 pieces to be used. If the sample size is insufficient, the corresponding shrinkage feature set component is marked with an underestimation annotation. The underestimation annotation component is subject to a more conservative upper limit of amplitude in the calculation of the feasible range of the pre-bias compensation pose. The shrinkage feature set version number is bound to the corresponding volume bias principal axis and constraint correction factor version number. After adjusting the adhesive application parameters or changing the adhesive batch, the shrinkage feature set must be re-extracted. This binding record allows for quick location of the corresponding batch's morphology data and constraint correction history during assembly defect analysis, enabling the reconstruction of the specific source of shrinkage prediction deviation.

[0040] The estimated shrinkage of the adhesive dot is obtained by calculating the maximum asymmetry amplitude of the side height based on the shrinkage feature set. The cosine of the shrinkage direction of each degree of freedom in the shrinkage feature set is converted into linear shrinkage components in each axis by the preliminary amplitude estimate. The conversion benchmark is the product of the maximum asymmetry amplitude of the profile height in the side profile distribution and the linear shrinkage rate of the material. The linear shrinkage rate of the material is taken from the material database according to the adhesive type and written into the configuration parameters. It is usually around 1.2% for commonly used UV adhesives. The maximum asymmetry amplitude of the profile height is quantified by the absolute value of the difference between the integrals of the heights on both sides of the center of gravity in the side profile distribution. The larger the difference, the more serious the volume distribution eccentricity, and the larger the amplitude of the estimated shrinkage of the adhesive dot. When the amplitude approaches zero, it means that the volume of the adhesive dot is close to symmetrical and the components of curing shrinkage in each direction tend to cancel each other out. In this case, the estimated shrinkage of the adhesive dot is based on the isotropic shrinkage benchmark value of the material. Even perfectly symmetrical adhesive dots will still have slight directional shrinkage due to local uneven heating during actual curing. The minimum value is the margin left for this part. After the adhesive batch is changed, the linear shrinkage rate varies between batches. The direction cosine of the shrinkage feature set is fixed by the adhesive dot morphology and does not change with the adhesive batch. The adhesive dot shrinkage prediction has a batch calibration mechanism for the linear shrinkage rate parameter. After each new batch of adhesive arrives, the actual linear shrinkage rate is deduced from the measured displacement of standard parts before and after curing, and then the material database is updated. When the deviation of the linear shrinkage rate between batches exceeds 0.3 percentage points, a batch deviation warning mark is added to the adhesive dot shrinkage prediction calculation result to remind the operator that the absolute accuracy of the current prediction is affected by batch differences. After the first batch of assemblies is cured, the consistency between the actual shrinkage and the prediction must be verified. The adhesive dot shrinkage prediction is output as a vector combination of shrinkage components of each axis. The vector direction is determined by the direction cosine of each degree of freedom stored in the shrinkage feature set (i.e., the offset shrinkage axis direction, which has been included in the constraint correction), and the vector amplitude is based on the converted linear shrinkage.

[0041] Step S102: Identify the boundary of the asymmetric vector reversal region based on the misalignment crossing record to form the zero crossing axis, and use the zero crossing axis to fit the position of the maximum coupling efficiency in the zero-crossing scan to obtain the optimal alignment pose.

[0042] In some embodiments, the step of identifying the boundary of the asymmetric vector reversal region and forming a zero crossing axis based on the inaccurate crossing record includes: extracting a weighted distribution of six-dimensional coordinates of crossing points from the inaccurate crossing record to obtain a spatial distribution of crossing points; performing a rotation-displacement coupling decoupling transformation on the spatial distribution of crossing points to obtain a decoupled coordinate distribution; performing principal inertial direction calculation on the decoupled coordinate distribution to establish a principal crossing direction vector; and mapping the principal crossing direction vector to the platform motion coordinate system to form a zero crossing axis.

[0043] The spatial distribution of crossing points is obtained by extracting the weighted distribution of the six-dimensional coordinates of crossing points from inaccurate crossing records. The six-dimensional coordinates of each crossing event in the inaccurate crossing records are weighted by the reciprocal of the direction angle reversal amplitude. A smaller reversal amplitude indicates the platform is closer to the zero crossing axis and has higher coordinate accuracy, thus receiving the highest weight; a larger reversal amplitude indicates the platform has made large leaps and the coordinates are far from the true axis, resulting in a correspondingly reduced weight. Authenticity verification is completed during the inaccurate crossing record generation stage; the reciprocal weight only considers the geometric proximity dimension in this stage. For crossing points carrying low-confidence annotations, the reversal weight is multiplied by an additional 0.5 reduction factor, while for jitter-corrected annotation crossing points, the weight is calculated by taking the reciprocal of the corrected reversal amplitude. Before implementing the amplitude reversal weighting, the extreme case where the reversal amplitude approaches zero must be handled. When the absolute value of θ is extremely small near the crossing time and close to the lower limit of sensor quantization noise, the reciprocal value may tend to infinity. The lower limit of noise is used as the lower limit of the denominator for truncation. The upper limit of the truncated weight is taken as the 5th quartile of the distribution of the smallest effective reversal amplitude in the historical batches. The weight of the crossing point with the smallest reversal amplitude is thus kept within the physically reliable range. The six-dimensional coordinates of each crossing point and the amplitude reversal weight are arranged side by side to form the spatial distribution of the crossing points. The weighted center coordinates of the spatial distribution of the crossing points represent the spatial position of all crossing points that is closest to the geometric expectation of the true zero crossing axis. The subsequent decoupling transformation is performed with this weighted center as the origin. In the inaccurate crossing record, the timing deviation marking crossing points use the weighting coefficient already determined in the summary stage, and no further reduction is required in this step; when the assembly scanning step size is too large, the reversal amplitude in the inaccurate crossing record is generally too large, and the corresponding crossing point has a generally low weight in the crossing point spatial distribution. Therefore, the weighted center is closer to the true zero crossing axis than the mean center of gravity. The axial positioning error of the batch with large step size is partially compensated by the reverse amplitude weighting.

[0044] A pre-applied rotation-displacement coupling decoupling transformation is applied to the spatial distribution of crossing points to obtain a decoupled coordinate distribution. The decoupling transformation uses the Jacobian matrix of the motion platform as the operator, converting the rotational coordinate components of each crossing point in the spatial distribution into equivalent end displacements, which are then merged with the original displacement components. After the transformation, each crossing point is described in pure displacement form within the decoupled space. The decoupled coordinate distribution consists of all transformed coordinates and their inverse amplitude weights. Directly calculating the principal inertial axis in the undecoupled six-dimensional coordinate space results in a direction vector that is already deformed by kinematic coupling; subsequent correction is merely an approximate compensation for the deformed result, leaving linearization truncation errors. By pre-transforming the entire system to the decoupled space before calculating the principal inertial axis, the direction vector itself is in the uncoupled physical reality space. The Jacobian matrix is ​​measured axis-by-axis during the assembly system calibration phase and written into the configuration parameters. The calibration accuracy directly determines the quality of the decoupling transformation. When the condition number of the Jacobian matrix is ​​too large, the equivalent displacement conversion error in certain rotation-displacement coupling directions is amplified, and the coordinate accuracy in the corresponding direction of the decoupling coordinate distribution is low. The numerical instability of this dimension, if left unchecked, will dominate subsequent eigenvalue decomposition. When the condition number exceeds 100, a high condition number label is registered in the corresponding coordinate dimension of the decoupling coordinate distribution, and regularization constraints are applied to the covariance matrix calculation in this dimension. When the coupling coefficient between the rotation axis and the displacement axis is small, the difference in point cloud morphology between the spatial distribution of crossing points and the decoupling coordinate distribution is minimal; when the coupling is strong, the difference between the two is significant. The improvement in the accuracy of the principal direction estimation is positively correlated with the magnitude of the coupling coefficient. The introduction of the decoupling coordinate distribution has the most prominent value in strongly coupled assembly configurations. When the low-confidence weight crossing points in the spatial distribution of crossing points are transformed into the decoupling coordinate distribution, the original reduced weight is inherited. The decoupling transformation does not change the weight magnitude, only the coordinate description space is changed.

[0045] The principal inertial direction is calculated for the decoupled coordinate distribution to establish the traversal principal direction vector. The decentralized coordinate vector qᵢ is obtained by subtracting the weighted center from the coordinates of each traversal point in the decoupled coordinate distribution. The weighted covariance matrix C_w is then calculated for each point using the amplitude weight wᵢ, as follows: In the formula, C_w is the weighted covariance matrix, qᵢ is the decoupled coordinate vector of the i-th crossing point after decentering, and wᵢ is the amplitude weight of that point. The summation in both places traverses all valid crossing points. This is the cross product of the vector and itself. Eigenvalue decomposition is performed on C_w. The direction of the eigenvector corresponding to the largest eigenvalue is the main crossing direction vector. This direction represents the linear direction with the largest variance in the crossing point set within the decoupled space, i.e., the geometric direction of the zero crossing axis. The quality of linear clustering is quantified by the ratio of the largest eigenvalue to the second largest eigenvalue. When the ratio is much greater than 1, the crossing point set is highly linearly clustered, and the direction of the main crossing direction vector has high reliability. When the ratio is close to 1, the point cloud is isotropic, and the main crossing direction vector is labeled with low-resolution annotations. When the high condition number dimension participates in the calculation of C_w, a regularization term λ_r is superimposed on the diagonal elements of this dimension, λ_r = β_k × σ_k², where β_k is the regularization ratio coefficient set during the initialization phase, and σ_k² is the coordinate variance of this dimension within the current statistical window. Regularization suppresses the influence of numerical instability on eigenvalue decomposition and prevents the high condition number dimension from dominating the eigenvector direction, causing the main crossing direction vector to deviate from the true zero crossing axis. If there are a few outlier crossing points deviating from the main line in the decoupled coordinate distribution, the outlier points have a limited contribution to the C_w calculation due to their low inverse amplitude weight. The suppression effect of the inverse amplitude weighting mechanism on outlier interference is reflected here again. When the non-parallelism of the normals is large in precision assembly, such as the cosine of the crossing main direction vector X-axis direction is 0.71 and the cosine of the crossing main direction vector Z-axis direction is 0.71, the main axis of the decoupled coordinate distribution point cloud is tilted at 45°. Pushing the X-axis or Z-axis alone can only scan the oblique section rather than the real zero crossing trajectory. Zero crossing must be performed in conjunction along the XZ diagonal direction.

[0046] The zero-crossing axis is formed by mapping the main crossing direction vector to the platform motion coordinate system. The main crossing direction vector is expressed as a pure displacement direction cosine in the decoupled coordinate space. After multiplying by the inverse Jacobian matrix, it is mapped back to the driving ratio of each axis in the platform motion coordinate system. The driving ratio of each axis is normalized according to the set scan step size, forming a combination of single-step motion increments for each axis of the platform. This combination constitutes the executable expression of the zero-crossing axis. The inverse Jacobian transformation restores the direction cosine of the decoupled space to the command components of the platform's physical axes. The centroid displacement on the far-field image plane must first be converted to platform displacement using the image plane-platform calibration magnification before participating in this mapping. If the driving ratio of a certain axis exceeds the platform's rated travel limit during the restoration, that axis component is truncated to the travel boundary, and the components of other axes are scaled proportionally. Truncation changes the precise correspondence between the zero-crossing axis and the main crossing direction vector. The directional deflection angle caused by truncation is written into the metadata. When the deflection angle exceeds 2°, a travel-limited label is attached. The travel-limited label triggers a segmentation strategy during the zero-crossing scan phase, and the complete zero-crossing region is covered by segmented splicing. When the traversing main direction vector carries a low-resolution annotation, the zero-crossing axis synchronously inherits this annotation. During the zero-crossing scan phase, the zero-crossing axis of the low-resolution annotation employs a dual strategy of narrowing the step size and expanding the search range. The batch consistency of the direction cosine of the zero-crossing axis across multiple batches is quantified by the standard deviation of the direction cosine of each axis. Axis axes with a persistently large standard deviation indicate insufficient repeatability of the fixture positioning mechanism for that axis. Batch sequences with a standard deviation below 0.02 indicate that the assembly tooling has entered a stable operating state, and the directional accuracy of the zero-crossing axis has converged to the lower limit of the tooling positioning accuracy. During long-term operation of the assembly system, mechanical wear causes the Jacobian matrix to drift, resulting in a systematic deflection of the mapping result from the traversing main direction vector to the zero-crossing axis. The monotonic drift trend of the direction cosine of the zero-crossing axis in adjacent batches serves as the trigger for Jacobian matrix recalibration. When the cumulative deflection exceeds 1°, a matrix recalibration warning is recorded in the zero-crossing axis batch record.

[0047] The optimal alignment pose is obtained by fitting the position of maximum coupling efficiency during zero-crossing scanning using the zero-crossing axis. The motion platform is driven to advance point by point along the scan path planned along the zero-crossing axis with equal-interval steps (the step size is appropriately narrowed and the search range expanded when carrying low-resolution annotations along the zero-crossing axis). At each scan position, far-field spots are simultaneously acquired and coupling efficiency is calculated in real time. The maximum value region in the coupling efficiency-position relationship curve formed by progressive scanning represents the coarse positioning range of the optimal alignment pose. After coarse positioning is determined, a local fine scan is performed near the maximum value position with a smaller step size. The fine scan result is fitted with a Gaussian curve, and the peak position is used as the final six-dimensional coordinates of the optimal alignment pose. Fitting the peak value has a higher position resolution than directly taking the highest sampling point, and the peak positioning accuracy can be refined to below the sampling interval. After the optimal alignment pose is confirmed, the centroid coordinates of the far-field spot in the current frame are simultaneously stored as the initial centroid reference for this batch and written into the batch configuration. The instantaneous deviation calculation during the solidification stage directly calls this reference. When carrying travel-limited annotations along the zero-crossing axis, the zero-crossing scan is performed segmentally. The coupling efficiency sequences of each segment are stitched together to form a complete curve. If an efficiency jump occurs at the stitching position, the search for the maximum value of the optimal alignment pose must exclude data points within a 3-step range near the stitching jump position, thus eliminating false maxima caused by continuation errors. During the fine scan, when thermal drift exceeds 0.05 micrometers, the data quality degrades. The thermal drift-affected annotation is recorded along with the optimal alignment pose. The optimal alignment pose carrying this annotation is appropriately relaxed in the tolerance judgment of thermal drift-sensitive dimensions during the pre-bias compensation pose calculation stage. When the Gaussian fit goodness of fit (R², i.e., coefficient of determination) of the fine scan is lower than 0.95, the optimal alignment pose replaces the fitted peak with the coordinates of the original sampling point with the highest efficiency and adds an annotation indicating insufficient fitting quality. The operator uses this to determine whether to re-execute the zero-crossing axis crossing point fitting.

[0048] Step S103: Perform reverse pre-bias compensation on the optimal alignment pose to determine the pre-bias compensation pose, verify the acceptableness of the spot index for the pre-bias compensation pose to obtain the pre-bias verification result, and determine the multi-point symmetrical glue application position and establish the glue dot distribution scheme based on the pre-bias verification result.

[0049] In some embodiments, the step of applying the estimated adhesive shrinkage amount to the optimal alignment pose to determine the pre-bias compensation pose includes: analyzing the shrinkage amount allocation value of each degree of freedom direction for the estimated adhesive shrinkage amount to obtain the axial shrinkage component; calculating the equal reverse bias amount of each axis using the axial shrinkage component to form an axial pre-bias vector; identifying axial pre-bias combinations within the spot quality tolerance based on the axial pre-bias vector to establish a feasible pre-bias interval; and synthesizing the optimal alignment pose and the optimal pre-bias amount of each axis based on the feasible pre-bias interval to construct the pre-bias compensation pose.

[0050] The axial shrinkage component is obtained by analyzing the shrinkage distribution values ​​in each degree of freedom for the estimated glue dot shrinkage. The estimated glue dot shrinkage is decomposed into X, Y, and Z displacement components and three-axis rotational components by multiplying the cosine of each axial shrinkage direction by the total shrinkage amplitude. The shrinkage distribution values ​​of the six degrees of freedom together constitute a complete axial shrinkage component vector group. The dimensions of each component are consistent with the corresponding degree of freedom, with the displacement component in length and the rotational component in angle. When the projection of the cosine of the estimated glue dot shrinkage direction on a certain degree of freedom is close to zero, the value of the axial shrinkage component in that degree of freedom is extremely small. Whether to treat it as zero depends on the upper limit of the batch fluctuation of the material linear shrinkage rate. Batch fluctuation may cause the component that is originally close to zero to have a considerable absolute value in actual curing. The degree of freedom component whose absolute value of the axial shrinkage is lower than the equivalent displacement of the batch fluctuation is marked as a noise level component. The noise level component is assigned zero when constructing the axial pre-bias vector. The estimation error already covers this value. Simply applying it to the reverse compensation will only introduce meaningless axial offset. The rotational degree of freedom of the axial shrinkage component is converted into an equivalent end displacement using the equivalent force arm length at the end of the motion platform. This is compared with the displacement component using the same dimensions. The equivalent force arm length is determined during the assembly tooling design and written into the configuration parameters. After conversion, all six degrees of freedom participate in the spot quality assessment of the pre-biased feasible interval in the form of equivalent end displacements. The influence of the two types of compensation components, rotation and displacement, on the spot index can be compared on the same scale. Components carrying low-discrimination or sparse contour annotations in the glue dot shrinkage prediction are marked with uncertainty amplification annotations at the corresponding positions of the axial shrinkage component. Components with insufficient estimation annotations from the shrinkage feature set retain insufficient estimation annotations to mark the source while adding uncertainty amplification annotations, forming a double annotation. Components with increased uncertainty annotations are applicable to a wider feasible range when establishing the pre-biased feasible interval. Components carrying insufficient estimation annotations are further suppressed by lowering the upper limit of candidate amplitudes, thus separately accommodating the risks of directional deviation due to insufficient morphological accuracy and amplitude overestimation.

[0051] The axial pre-bias vector is formed by calculating the equal reverse offset of each axial component using the axial shrinkage component. The axial pre-bias vector is constructed by superimposing a systematic correction after reversing the orientation of each degree of freedom component of the axial shrinkage component, using the following formula: V_pre(d) = -C(d) + δ_sys(d), where V_pre(d) is the axial pre-bias of degree d, with positive values ​​indicating pre-biasing in the positive direction and negative values ​​indicating pre-biasing in the negative direction; C(d) is the shrinkage distribution value of the axial shrinkage component in degree d, with displacement degrees of freedom expressed in length and rotation degrees of freedom converted to equivalent end displacements based on the equivalent force arm length (with the same dimensions as displacement degrees of freedom), positive values ​​corresponding to shrinkage in the positive direction, and negative values ​​representing the basic reverse pre-bias; δ_sys(d) is the systematic correction for shrinkage deflection caused by the curing process, determined by the statistical mean of the difference between the measured shrinkage and the estimated value of the axial shrinkage component in historical curing batches. When δ_sys(d) is zero, the axial pre-bias vector degenerates into a pure equal reverse offset. The theoretical premise of equal-amount reverse bias is that the curing shrinkage amount and the axial shrinkage component are completely consistent. In actual assembly, the difference between the two constitutes the compensation for residual error. The uncertainty of each component of the axial pre-bias vector is estimated by the product of the standard deviation of the material batch shrinkage rate and the corresponding component of the axial shrinkage component. The component with larger uncertainty is applicable to a wider feasible range in the pre-bias feasible interval identification stage to accommodate the prediction deviation caused by batch differences. The uneven spatial distribution of UV irradiation intensity will lead to the difference in curing rate on different sides of the adhesive dot. The shrinkage of the side that cures first occurs earlier and is constrained by the side that cures later. The actual shrinkage axial deflection amount systematically deviates from the axial shrinkage component. δ_sys(d) is written into the axial pre-bias vector after statistically analyzing the shrinkage deflection law caused by the curing process based on historical batch data. This makes the axial pre-bias vector not only reflect the shrinkage direction predicted by the morphology, but also include empirical corrections for the curing process deflection. In the axial shrinkage component, the noise level component takes zero at the degree of freedom corresponding to the axial pre-bias vector C(d), and V_pre(d) retains only the correction amount δ_sys(d). This type of degree of freedom is not subject to active pre-biasing, and its shrinkage direction error is passively offset by the force balance mechanism of multi-point symmetrical gluing.

[0052] Based on the axial pre-bias vector, a feasible pre-bias interval is established by identifying axial pre-bias combinations within the spot quality tolerance. The bias of each degree of freedom of the axial pre-bias vector is expanded in both positive and negative directions based on the rated value, and the expansion range of the degree of freedom carrying the uncertainty expansion label is widened by one level. Candidate bias combinations are generated with a fixed sampling density within the expanded multidimensional interval, and the key labeled components in the single column of the feature set are shrunk, with the sampling density of their corresponding degrees of freedom increased by one level and the tolerance boundary verified first. After the candidate bias combinations are superimposed on the optimal alignment pose, a candidate pre-bias compensation pose is formed. The predicted spot ellipticity at the candidate pre-bias compensation pose is evaluated by the ellipticity-pose quadratic response surface model fitted by the measured data of each step in the alignment scanning stage (the fitting residual of the model within 3 steps around the optimal position is usually less than 5% of the tolerance). Candidate bias combinations with predicted ellipticity lower than the upper limit of the process tolerance are identified as feasible combinations. The distribution range of all feasible combinations in the multidimensional bias space is the feasible pre-bias interval. For batches where the optimal alignment pose carries the influence of thermal drift, the tolerance for thermal drift-sensitive dimensions is relaxed by one level, and the drift uncertainty of the reference coordinates themselves is included in the boundary of the feasible pre-bias interval. For batches with large fiber end face cutting angles, the ellipticity of the optimal alignment pose itself is high, and the margin from the upper limit of the tolerance is small, which significantly narrows the feasible pre-bias interval and reduces the number of feasible offset combinations. When the feasible pre-bias interval is severely narrowed, the construction of the pre-bias compensation pose must prioritize the offset combination with the lowest predicted ellipticity, rather than simply pursuing the closest approximation between the offset amount of each axis and the nominal value of the axial pre-bias vector.

[0053] The optimal alignment pose and the optimal pre-bias amount for each axis are synthesized based on the feasible pre-bias interval to construct the pre-bias compensation pose. The optimal bias combination is selected from the feasible pre-bias interval according to the joint scoring criterion of the lowest predicted spot ellipticity and the bias amount for each axis being closest to the nominal value of the axial pre-bias vector. The joint score J = w_e × Δε + w_v × ΔV, where Δε is the residual of the spot ellipticity predicted by the response surface model for the candidate bias combination (approaching zero indicates that the spot is close to circular symmetry and is dimensionless), ΔV is the normalized deviation amplitude of the bias amount for each axis of the combination from the nominal value of the axial pre-bias vector (normalized to the nominal value and is dimensionless), w_e and w_v are the scoring weights corresponding to the ellipticity term and the deviation term, respectively, and w_e + w_v = 1. The rotational components of the optimal offset combination are restored to angular quantities by the equivalent force arm and then superimposed onto the coordinates corresponding to the optimal alignment pose. This is the pre-offset compensation pose. The pre-offset compensation pose also records the optimal offset combination used and the corresponding joint score. If the pre-offset verification result fails, the score is used to backtrack and the second-best offset combination is selected from the feasible pre-offset interval to reconstruct the pre-offset compensation pose. The entire pre-offset calculation does not need to be repeated from scratch. When the offset of the optimal offset combination in a certain degree of freedom falls exactly at the boundary of the feasible pre-offset interval, the feasible margin of that degree of freedom is extremely limited. The pre-offset compensation pose is marked with insufficient boundary margin next to the coordinates of the corresponding degree of freedom. The axial spot index has low tolerance to environmental disturbances. Even slight temperature changes or platform positioning jitter may push the actual position out of the boundary of the feasible pre-offset interval. The operator must pay close attention to the ellipticity change of this axis. The difference between the coordinates of each degree of freedom of the pre-bias compensation pose and the corresponding coordinates of the optimal alignment pose is the final executed reverse pre-bias amount of each axis. The ratio of the absolute value of the reverse pre-bias amount of each axis to the rated value of the axial pre-bias vector reflects the actual execution ratio of the final compensation scheme to the theoretical optimal compensation amount. This ratio is used as a reference quantity in the glue dot distribution scheme determination stage to evaluate the shrinkage offset effect of multi-point symmetrical glue application. An axial axis with a low ratio indicates that the pre-bias feasible interval constraint has suppressed the compensation execution amount of that axis, and the gap needs to be filled by adjusting the spatial distribution of glue application points in the glue dot distribution scheme.

[0054] The pre-biasing verification results are obtained to verify the acceptable tolerance of the spot index in the pre-biasing compensation pose. After the motion platform moves to the pre-biasing compensation pose coordinates, the far-field spot of the current frame is acquired. Two indices, ellipticity and spot centroid offset, are extracted from the spot image. Ellipticity is converted from the ratio of the major and minor axes fitted by the second-order central moments. Centroid offset is the difference between the current centroid and the initial centroid of the optimal alignment pose. The upper limits of the tolerances for both are back-calculated based on the device coupling efficiency index and written into the process configuration. The pre-biasing verification result is considered acceptable only if both fall within the upper limit of the process tolerance. The pre-biasing compensation pose has multiple axial offsets relative to the optimal alignment pose. After the offset, the ellipticity will inevitably increase. The pre-biasing verification result must distinguish between the acceptable ellipticity increment caused by the pre-biasing and the excessive ellipticity caused by the estimation deviation of the adhesive point shrinkage. The acceptable increment is taken as the ratio of the difference between the ellipticity of the optimal alignment pose and the upper limit of the tolerance, generally taken as 55% of the difference. If the increment of the current ellipticity relative to the optimal alignment pose reference falls within the acceptable range, the ellipticity item is considered acceptable. Excessive centroid offset of the optical spot is mostly due to an overestimation of the displacement component of the adhesive dot shrinkage estimate and an excessive deviation of the pre-offset compensation pose along the displacement axis. The pre-offset verification results, with the direction of the excess and the magnitude of the offset both pointing to the degree of freedom with the largest predicted deviation in the adhesive dot shrinkage estimate, require the operator to make targeted adjustments to the corresponding component of the adhesive dot shrinkage estimate based on this directional information, and then re-execute the pre-offset compensation verification. Two unqualified pre-offset verification results will trigger an assembly batch warning: the adhesive dot morphology data analysis results of this batch of lens-fiber assemblies have a systematic deviation from the actual curing characteristics, requiring re-collection of adhesive dot morphology data and verification of the calculation path of the adhesive dot shrinkage estimate.

[0055] The pre-bias verification results determine the multi-point symmetrical adhesive application positions and establish an adhesive dot distribution scheme. After the pre-bias verification results are qualified and the pre-bias compensation pose and spot index are confirmed to meet the process requirements, multiple adhesive application positions are selected around the lens using the pre-bias compensation pose as a spatial reference. Taking one point symmetrically placed at the front and back of the lens as an example: the front adhesive dot generates a backward contraction force during curing, and the rear adhesive dot generates a forward contraction force. The rotational components cancel each other out, and the resultant force is along the optical axis, which is the pre-set main contraction direction of the adhesive dot distribution scheme. After curing, the lens rebounds towards the optimal alignment pose along this direction. The direction information of the centroid offset in the pre-bias verification results is directly referenced in the selection of the symmetry axis of the adhesive dot distribution scheme. The centroid offset direction gives the actual offset direction after the pre-bias is executed. The symmetry axis is oriented along this direction, and the resultant contraction force cancels out the centroid offset in the opposite direction, resulting in the minimum residual offset of the lens after curing. The execution ratio of each axis recorded in the pre-bias compensation pose determination stage is also referenced here: for axes with a low execution ratio, the adhesive dots are densified along that axis or offset as a whole, and the contraction force that the pre-bias cannot compensate for is compensated by the adhesive layout. Due to structural limitations, the space for applying adhesive around the lens may not be suitable for perfectly symmetrical application positions. The adhesive dot distribution scheme must strike a balance between symmetry constraints and spatial accessibility constraints. When constraints conflict, the minimization of the shrinkage force direction error should be the priority when selecting an approximately symmetrical scheme. In batches where the ellipticity increment in the pre-bias verification results is close to the upper limit of the tolerance, the available margin for pre-bias compensation pose is already too small. In such batches, the adhesive dot distribution scheme should appropriately reduce the number of application points. The interference of curing stress at each point on the final alignment state decreases along with the number of points, and the amount of adhesive per point is increased accordingly, while the total amount of adhesive must still be sufficient for bonding strength.

[0056] Step S104: Based on the glue dot distribution scheme, the real-time offset of the centroid of the UV curing spot is recorded to generate a curing drift record. The curing drift record is used to identify the moment of sudden increase in offset rate to obtain the gelation initiation parameter set. The offset that can be compensated in the soft glue state is calculated from the gelation initiation parameter set, and a transient compensation command is output.

[0057] Specifically, based on the adhesive dot distribution scheme, the real-time offset of the centroid of the UV curing spot is recorded to generate a curing drift record. After dispensing adhesive at each application position determined by the adhesive dot distribution scheme, UV irradiation is immediately started. The symmetrical layout of the adhesive dot distribution scheme determines the design direction of the curing shrinkage force, and drift monitoring is the actual verification of this direction. Before starting UV irradiation, the centroid coordinates of the far-field spot of the current frame are stored as the curing drift zero reference for this batch and written into the batch configuration. During irradiation, the CCD continuously acquires far-field spot frames, and the centroid coordinates of each frame are extracted through intensity moment calculation. The difference between the centroid coordinates and the curing drift zero reference constitutes the real-time centroid offset of that frame. The offsets of each frame throughout the process are arranged according to the frame number to form a curing drift record. Low UV irradiation power results in slow curing of the adhesive dots, leading to a higher total number of frames and longer frame intervals in the gelation stage of the curing drift record in low-power batches. This provides ample window for setting the acceleration threshold to identify sudden increases. Conversely, higher power results in faster curing, with the gelation stage occupying only a very short frame interval in the curing drift record. The acceleration threshold needs to be adjusted to be more sensitive, and batches compressed into very few frames by rapid gelation must be detected to avoid missed detections. The far-field spot ellipticity is recorded synchronously in each frame of the curing drift record: the centroid offset increases rapidly, and the ellipticity increases synchronously, mostly due to the lens undergoing actual displacement driven by the combined force of the contraction of the adhesive dots in the adhesive dot distribution scheme. Centroid jitter caused by platform vibration exhibits high-frequency, low-amplitude oscillations, and the ellipticity does not increase synchronously. Frame segments with ellipticity changes exceeding 15% of the pre-curing baseline are marked with a high ellipticity warning. These frames are prioritized for inspection during the offset rate sequence generation stage as key gelation candidate moments.

[0058] In some embodiments, the step of using the solidified drift record to identify the moment of sudden increase in offset rate and obtain the gelation initiation parameter set includes: calculating the centroid displacement difference component of adjacent frames from the solidified drift record to generate an offset rate sequence; analyzing the rate of change of the rate in the offset rate sequence to determine the rate acceleration sequence; identifying the moment of the first frame where the acceleration continuously exceeds the threshold and the drift is unidirectional and irreversible based on the rate acceleration sequence to establish a sudden increase event set; and locking the start time of the first sudden increase event and the offset at that time according to the sudden increase event set to obtain the gelation initiation parameter set.

[0059] The offset rate sequence is generated by calculating the centroid displacement difference components between adjacent frames from the curing drift record. The difference between the centroid offset coordinates of each frame in the curing drift record and the adjacent previous frame is divided by the frame time interval to obtain the centroid movement rate in the X and Y directions. The synthesized rate vector magnitude and direction angle are arranged according to the frame number to form the offset rate sequence. In the initial stage of normal UV curing, the glue dot leveling has ended, and the lens position is constrained by the viscoelasticity of the soft glue, resulting in a slow change. The offset rate sequence is in a stable state with low rate and random direction in the early stage of curing. Once gelation occurs, the rate value increases sharply and the direction becomes more unidirectional. This change is the key signal source for the rate acceleration sequence to capture the moment of gelation initiation. The offset rate sequence corresponding to frames with high ellipticity warning annotations in the curing drift record generally has higher rate values. These frames are marked with high ellipticity inheritance annotations in the offset rate sequence. When calculating the rate acceleration sequence, the high ellipticity inheritance annotation frames are given an extended mean window, thus suppressing the rate fluctuations introduced by ellipticity changes. In batches where the UV irradiation controller trigger frame rate is fixed, the frame interval is constant. In batches where the curing drift record acquisition frame rate is not completely synchronized with the irradiation trigger frame rate, the frame interval must be calculated based on the difference in the actual timestamps of each frame. The rate calculation directly reads the timestamp difference without relying on the nominal frame rate, and the timing accuracy of the offset rate sequence is not affected by the frame rate drift. When a single frame centroid anomalous jump occurs in the curing drift record, the corresponding frame rate value is significantly higher, and the frame rates before and after return to normal. The offset rate sequence includes isolated jump markers for single frame rate values ​​exceeding 6 times the average of adjacent frames. Isolated jump marker frames are replaced by the average frame rates before and after during the rate acceleration sequence calculation stage. The identification of the gelation initiation time relies on the continuous trend and cannot withstand the interference of single frame false peaks.

[0060] The rate acceleration sequence is determined by analyzing the rate difference rate in the offset rate sequence. The rate magnitude of the offset rate sequence is extracted frame by frame. The rate difference between adjacent frames is divided by the frame interval to obtain the rate change rate, i.e., the acceleration value, for each frame. The acceleration values ​​of all frames are arranged by frame number to form the rate acceleration sequence. In the early stages of curing, the offset rate sequence exhibits low rate and random direction. The inter-frame difference of the rate magnitude in this stage appears as a low-amplitude alternating positive and negative noise sequence. The mean of the rate acceleration sequence is close to zero in the early stages of curing. After gelation occurs, the rate magnitude of the offset rate sequence increases monotonically, and the difference between adjacent frames remains positive. The rate acceleration sequence shows a sustained positive surge during the gelation stage. Accurate location of the starting frame of this surge is crucial for establishing the surge event set. In the offset rate sequence, isolated jump marker frames have been replaced with the mean. No false acceleration peaks appear at the corresponding positions in the rate acceleration sequence due to single-frame anomalies. The acceleration of high ellipticity inherited marker frames is averaged using an extended mean window before being included. The rate acceleration sequence marks continuous positive value segments lasting longer than 0.2 seconds with sustained sudden increases (whether this exceeds the lower limit of the acceleration threshold is determined in the next stage of the sudden event set). Isolated positive peaks are not triggered, thus separating the rate jitter of brief platform vibrations from the unidirectional acceleration of actual gelation. The vibration frequency of the platform micro-vibration is much higher than the rate rise frequency caused by gelation. After the rate acceleration sequence is filtered by a band-stop window corresponding to the vibration frequency, the high-frequency acceleration components of the vibration frequency band are suppressed, and the low-frequency gelation signal stands out more clearly in the rate acceleration sequence. The frame values ​​of the filtered rate acceleration sequence are used as the primary criterion in the sudden event set identification stage. The unfiltered original sequence is retained in the additional fields of the sudden event set for reference when analyzing the vibration environment state before gelation.

[0061] A burst event set is established based on the first frame moment of acceleration that continuously exceeds the threshold and drifts irreversibly in one direction, using rate acceleration sequences. The segments in the rate acceleration sequence carrying continuous burst markers constitute a candidate pool for the burst event set. The acceleration values ​​of each frame in the candidate segments are compared frame-by-frame with the lower limit of the threshold. If the first frame of a segment simultaneously satisfies the condition that the corresponding offset rate sequence direction angle remains within a single-direction sector, it is considered the start of a burst event. The start moments of all burst events and their corresponding acceleration peaks together constitute the burst event set. The acceleration threshold is taken as the 95th quantile of the statistical distribution of rate acceleration sequences from historical normal curing batches. If the threshold is too low, the low-amplitude acceleration residual from platform vibration will be mistakenly identified as a burst event; if it is too high, the gelation acceleration characteristics of mild glue batches will be weak, leading to missed detections. The directional uniformity is determined by the standard deviation of the angular dispersion of the offset rate sequence direction angle within consecutive frames being less than 15°. A larger dispersion indicates that the lens displacement direction is still randomly changing and is not a unidirectional movement driven by gelation shrinkage. Some adhesive varieties exhibit a stress relaxation phase. During this period, the rate acceleration sequence briefly shows a moderate acceleration value before declining. The unidirectional directional condition effectively separates the random displacement acceleration of stress relaxation from the fixed acceleration of gel shrinkage. When multiple candidate events exist within a batch, the event with the smallest directional dispersion is identified as the true gel burst, while the remaining candidate events are marked as pre-burst candidates and added to the burst event set. A brief acceleration pulse may occur in the rate acceleration sequence when UV irradiation switches to the final curing power. This pulse originates from thermal expansion caused by a sudden increase in irradiation intensity. The burst event set skips 10 frames before and after the irradiation power switch in time, preventing the thermal expansion pulse from being included in the identification of the gel burst.

[0062] The gelation initiation parameter set is obtained by locking the start time and offset of the first burst event in the burst event set. The frame number of the earliest real gelation burst event in the burst event set is used as the locking coordinate of the gelation initiation time, and the centroid offset of the solidification drift record corresponding to that frame is used as the lens displacement state at the time of gelation. Together, they constitute the core content of the gelation initiation parameter set. The first burst is locked instead of the maximum burst because gelation is irreversible once it begins. The earliest identified gelation initiation time is the last effective window for implementing transient compensation in the soft glue state. Continuous judgment and filtering itself consume frames. When the recognition confirmation time is more than 20 frames behind the starting frame recorded in the burst event set, or when it is close to the acceleration peak time, the gelation degree has already deepened, and the compensable range of the soft glue has narrowed significantly. In this scenario of recognition lag, the gelation initiation parameter set adds a gelation deepening warning to prompt the operator to speed up the compensation decision. The difference between the start time of the candidate events for sudden increase in the sudden increase event set and the actual start time of the sudden increase in gelation is written into the gelation initiation parameter set. A large difference indicates that the batch of glue has a significant stress relaxation precursor stage. Based on this, the operator can appropriately advance the start time of the monitoring window in the next batch, thus providing more time margin for compensation operations. The centroid offset of the gelation time in the gelation initiation parameter set is simultaneously decomposed into X and Y axis components and stored. The correspondence between the offset direction of the two axis components and the pre-offset direction of each axis relative to the optimal alignment pose of the pre-offset compensation pose is the basis for subsequent real-time deviation judgment: if the offset direction is completely opposite to the pre-offset direction, the curing shrinkage is correcting in the opposite direction as expected; if the offset direction is the same as the pre-offset direction, the shrinkage direction deviates significantly from the estimate. In this case, the gelation initiation parameter set adds an abnormal shrinkage direction annotation, triggering a review of the glue dot shrinkage estimate.

[0063] In some embodiments, the step of calculating the compensable offset in the soft gel state from the gelation initiation parameter set and outputting a transient compensation command includes: obtaining a drift step length reference by calculating the maximum single-step displacement of the centroid of the monitoring window spot before gelation using the gelation initiation parameter set; determining the instantaneous deviation by calculating the centroid alignment deviation at the gelation moment based on the gelation initiation parameter set; obtaining a compensable coefficient by analyzing the ratio between the drift step length reference and the instantaneous deviation and reserving a curing rebound margin; and outputting a transient compensation command by synthesizing the compensation amount allocated to each axis based on the compensable coefficient.

[0064] The drift step length reference is obtained by calculating the maximum single-step displacement of the centroid of the monitoring window spot before gelation using the gelation initiation parameter set. The frame number at the start of gelation is read from the gelation initiation parameter set, and the frame range corresponding to the length of the monitoring window is traced backward. Within this traceback window, the absolute value sequence of the difference between the centroid offsets of adjacent frames in the curing drift record is extracted. The maximum value among the absolute differences of each frame is the drift step length reference. The monitoring window length is set to one-third of the average duration of the soft gel stage. If the window is too long, it will include the thermal expansion displacement frames from the initial UV irradiation in the early gelation stage, resulting in a larger single-step centroid difference during the thermal expansion stage and an artificially inflated drift step length reference. If the window is too short, the number of effective frames is insufficient, and the statistical representativeness of the single-step difference is not adequately supported. The physical meaning of the drift step length reference is the maximum displacement that the lens can move freely within one acquisition frame in the soft gel state, reflecting the fluidity of the current soft gel state: lower glue viscosity and earlier curing process result in a larger drift step length reference, providing ample room for transient compensation. As curing progresses, if the absolute value of the difference between adjacent frames in the curing drift record decreases frame by frame in the last 10 frames before the gelation initiation parameter set locks in, it indicates that the adhesive has transitioned from a low-viscosity flow dynamic to a viscoelastic thickening stage. The gelation initiation parameter set adds a decreasing trend annotation to the drift step reference value. Batches carrying this annotation have increased rebound margin when calculating the compensable coefficient, ensuring sufficient operational allowance. The drift step reference value is calculated separately for the X and Y axis components, and the smaller value of the two axes is used as a conservative estimation benchmark. The constraint of the compensable coefficient on the compensation amount then covers the most unfavorable axis.

[0065] The instantaneous deviation is determined by calculating the centroid alignment deviation at the gelation initiation time based on the gelation initiation parameter set. The centroid offset at the gelation initiation time, locked by the gelation initiation parameter set, is superimposed on the curing drift zero reference to restore the absolute centroid coordinates. This is then subtracted from the initial centroid coordinates under the optimal alignment pose to obtain the current deviation of the lens relative to the optimal alignment pose at the gelation initiation time, i.e., the instantaneous deviation. The initial centroid coordinates under the optimal alignment pose are immediately recorded and written into the batch configuration after alignment is completed in S102. The instantaneous deviation calculation directly uses this recorded value without rescanning, thus preventing temperature drift during UV irradiation from contaminating the reference. The angle between the direction of the instantaneous deviation and the expected shrinkage direction of the glue dot shrinkage reflects whether the actual curing shrinkage develops according to the predicted trajectory. Taking the systematic deflection of the glue batch shrinkage direction as an example, if the measured instantaneous deviation direction deviates from the predicted axis by more than 45°, it indicates that there is a systematic deviation between the current batch curing shrinkage behavior and the morphology prediction, and the glue dot shrinkage prediction must be checked. When the angle exceeds 45°, the instantaneous deviation is marked with a direction deviation. The instantaneous deviation marked with a direction deviation uses the vector projection component instead of the absolute value when calculating the compensable coefficient. Only the component along the expected shrinkage direction participates in the compensation calculation. The component deviating from the expected direction does not participate in this transient compensation command. Compensating in the wrong direction will only push the lens away from the optimal alignment pose. When the absolute value of the instantaneous deviation is less than 0.1 micrometers, it is considered that the lens is close enough to the optimal alignment pose. At this time, the substantial impact of curing shrinkage on the alignment accuracy is within the tolerance range, the compensable coefficient is set to zero, and an empty compensation mark is output in the transient compensation command.

[0066] The compensable coefficient is obtained by analyzing the ratio between the drift step length reference value and the instantaneous deviation value, and by reserving a curing springback margin. The ratio of the drift step length reference value to the instantaneous deviation value quantifies the proportion of deviation that can be corrected by a single-step compensation operation. Combined with the curing springback margin coefficient, the compensable coefficient is formed and calculated by the following formula: k_comp(d)=min(1.0,D_step(d) / D_inst(d))×(1-r_rb), where k_comp(d) is the compensable coefficient for degree of freedom d; D_step(d) is the component of the drift step length reference value in degree of freedom d, representing the maximum displacement that can be withstood in a single step in the soft rubber state; D_inst(d) is the component of the instantaneous deviation value in degree of freedom d. For conventional batches, the absolute value component of the deviation value in that degree of freedom is taken, and the carrying direction is considered. The batch deviation annotation takes the vector projection component of the deviation along the expected shrinkage direction, which represents the effective correction distance of the current spot centroid deviating from the optimal alignment pose; when the ratio of D_step(d) / D_inst(d) is greater than 1, it is truncated to 1.0, which means that the single-step compensation amount can completely cover the instantaneous deviation amount. The truncation prevents overcompensation from pushing the lens past the optimal alignment pose; r_rb is the curing springback margin coefficient, which is written into the configuration parameters during the initialization stage based on the statistical average of the rebound amplitude of the adhesive dots in the later stage of curing of historical batches. The (1-r_rb) factor ensures that the compensable coefficient does not exceed the set proportion of the displacement that the soft glue can withstand, and reserves a margin for springback caused by stress release in the later stage of curing shrinkage. When k_comp(d) approaches 1, such as when D_step=0.5μm, D_inst=0.2μm, and r_rb=0.15, k_comp≈0.85, it indicates that the soft rubber has sufficient margin and the rebound margin meets the requirements. This compensation can completely cover the current deviation with high confidence. When k_comp(d) is low, it usually corresponds to the gelation start time being late and the soft rubber step size being significantly narrowed. The instantaneous deviation may not be completely corrected. The transient compensation command is executed with the partial compensation amount corresponding to k_comp(d). The residual deviation is handled by the residual displacement correction mechanism of S105 after curing.

[0067] The transient compensation command is output based on the compensation amount allocated to each axis according to the compensable coefficient. The compensation amount for each axis component k_comp(d) of the compensable coefficient is multiplied by the corresponding axis component D_inst(d) of the instantaneous deviation (D_inst(d) of the direction deviation labeling batch is the vector projection component) to obtain the compensation amount allocated to each axis. This is then negative and converted using a calibration multiplier to form the motion platform command increment. The all-axis command increments are superimposed on the current pre-bias compensation pose coordinates, and the resulting motion platform target coordinates constitute the execution parameters of the transient compensation command. The compensation amounts allocated to each axis must be executed synchronously, not sequentially. When advancing axis by axis, the axis completed first has already moved the lens position, and the compensation amount for the axis executed later is still calculated based on the old reference, naturally resulting in inaccuracies. Therefore, the transient compensation command is issued in a multi-axis synchronous motion mode, with the adjustment speeds of each axis matching each other, reaching the target coordinates simultaneously. When the compensation amount k_comp(d) of a certain axis in the compensable coefficient is zero, that axis is output in the transient compensation command while maintaining its original position. The transient compensation command for the empty compensation labeling batch outputs in the original position across all axes, and the motion platform does not perform any displacement. After the transient compensation command is executed, the solidification drift record continues to be acquired. If subsequent frames show that the centroid offset briefly reverses and then stabilizes after the compensation is executed, it indicates that the soft rubber has generated viscoelastic rebound after the compensation displacement. The rebound amplitude is mutually verified with the selection of r_rb. If the rebound amplitude in historical batches is consistently higher than the r_rb allowance, the r_rb parameter is automatically adjusted upwards, and the rebound margin reserved by the compensable coefficient in subsequent batches becomes more conservative. The residual deviation of the solidification drift record after the transient compensation command is executed is written into the same batch record. The S105 solidification stress identification stage uses this as the reference starting point for judging the current relatively optimal alignment pose deviation.

[0068] Step S105: Based on the transient compensation command, analyze the slope of the servo drive current increment of each axis in the later stage of curing to form a curing stress record. Use the curing stress record to identify the axis of sudden increase in current increment slope to obtain the principal stress axis. Calculate the axial residual displacement correction amount from the principal stress axis and output the precision assembly command.

[0069] In some embodiments, the step of analyzing the slope of the servo drive current increment of each axis in the later stage of curing based on the transient compensation command to form a curing stress record includes: using the transient compensation command to identify the steady-state reading of the servo current of each axis after compensation to obtain a current reference value; calculating the timing sequence of the servo current increment of each axis in the later stage of curing based on the current reference value to form a current increment sequence; analyzing the rate of change of the current increment per unit time of each axis for the current increment sequence to establish a slope record for each axis; and aggregating the maximum slope value of each axis with the corresponding time to form a curing stress record based on the slope record of each axis.

[0070] The current reference value is obtained by identifying the steady-state readings of the servo current of each axis after the transient compensation command is completed. After the transient compensation command is issued, the servo drive of each axis completes the position adjustment and enters steady-state holding. Steady-state confirmation is determined by the condition that the difference between the actual position and the target position of each axis is less than 0.03 micrometers for 8 consecutive sampling cycles. The average value of the servo current readings of each axis within the first complete sampling window after the steady-state confirmation time is the current reference value of that axis. During the execution of the transient compensation command, each axis is in a dynamic adjustment process, and the drive current fluctuates violently due to the influence of acceleration and deceleration control. The steady-state window must be collected after the motion has completely stopped and a sufficient number of stable frames have been confirmed. If the window start time is determined too early, the current reference value will be mixed with residual oscillation components of the motion, resulting in an overestimation of the reference value. When the transient compensation command is completed, the adhesive dots are still in the soft adhesive stage, and the curing shrinkage stress has not yet had a significant impact on the servo current. The current reference value truly reflects the static holding current required for each axis when the lens is constrained by the soft adhesive at the pre-bias position. If the steady-state confirmation time has entered the gelation acceleration stage, the current reference value may show a monotonically increasing trend within the steady-state window. For this batch, the current reference value is replaced by the lowest current point within the steady-state window to capture the true static reference before stress accumulation. If the reference is artificially high, the incremental value of the subsequent current increment sequence will be systematically suppressed. The current reference value of each axis is also recorded at the lens position coordinates at the steady-state confirmation time. These coordinates are retained as a comparison benchmark for the displacement evolution in the later stage of curing: the correction increment of the precision assembly command is superimposed on the actual coordinates at the time of curing completion, and the steady-state coordinates are used to verify the reconciliation between the cumulative displacement in the later stage of curing and the final residual deviation.

[0071] The current increment sequence is constructed by calculating the servo current increment of each axis during the later stage of curing based on the current reference value. The current increment of the current in the current sampling period is obtained by subtracting the corresponding axis current reference value from the real-time servo current reading of each axis. The current increment of each axis is arranged in time sequence for the entire sampling period to form the current increment sequence of each axis. The current increment sequence of each axis is displayed side by side to form a multi-axis matrix that can be compared laterally to show the stress development process. In the early stage of curing, the current of each axis is still close to the current reference value and the increment is close to zero. As the UV irradiation time increases and the curing degree deepens, the current increment sequence of the axis bearing shrinkage stress gradually rises. The axis with a faster rise rate corresponds to a stronger component of shrinkage force in that axis direction. Taking the front and rear adhesive dots as an example, the front adhesive dot cures first, and the shrinkage force pushes the lens backward along the optical axis. The Z-axis servo must continuously resist this pushing force to maintain its position. The Z-axis current increment sequence rises first, while the X and Y axis increments are still close to zero. When the front and rear adhesive dots cure simultaneously, the bidirectional resultant force partially cancels out, and the rise amplitude of the Z-axis decreases accordingly. The rise sequence and amplitude of the multi-axis matrix directly correspond to the actual curing process and shrinkage force spatial distribution of each adhesive dot. If multiple adhesive dots around the lens cure asynchronously during the curing process, the shrinkage stress of the first cured adhesive dot will rise first relative to the current reference value along the corresponding axis in the current increment sequence. The stress induced by the later cured adhesive dots will then be superimposed in the current increment sequence. The multi-stage rise pattern of the current increment sequence can identify the curing sequence of each adhesive dot by comparing historical batches. The fixture fluctuation annotation frame is replaced by linear interpolation in the current increment sequence. There are no abrupt steps during fixture interference to disrupt the trend estimation of the slope records of each axis. The current increment sequence of the stress invalid axis annotation axis is still recorded completely, but it is marked as a reference sequence in the slope calculation stage and does not participate in the main criterion for curing stress recording.

[0072] To analyze the current increment sequence, the rate of change of current increment per unit time for each axis is used to establish the slope record for each axis. The current increment sequence for each axis is pushed forward point by point using a fixed-length sliding window. Within each window, a least-squares linear regression is performed with the time coordinates of each sampling point (taken as integer multiples of the sampling period) as the independent variable and the current increment as the dependent variable. The slope of the resulting fitted line is the current increment slope at that sampling moment, with the dimension being the ratio of current increment to time. After sliding point by point, the slope sequence of each axis changing with time is obtained. The sliding window length is determined in the initialization stage according to the ratio of the typical rise time constant of curing stress to the sampling period. For batches with fast adhesive curing and steep stress rise, a shorter window is used to shorten the response delay; for batches with slow curing and relatively prominent current noise, a longer window is used to smooth out random fluctuations. The goodness of fit is output simultaneously for each window regression. A goodness of fit below a set threshold indicates that the current increment within the window deviates from linearity, mostly due to fixture fluctuations or sampling noise disturbances. The slope of this sampling point is marked as suspicious and is not included in the peak comparison. A positive slope indicates that the current increment of the corresponding axis is on an upward trend. When the slope approaches zero, the current increment of the corresponding axis enters a plateau. The larger the absolute value of the slope, the faster the accumulation rate of curing stress in that axis direction. The slopes of each sampling point across all axes are arranged by a dual index of axis number and time to form the slope record for each axis. The selection of the window length must balance noise smoothing and response speed. If the window is too short, the slope will be too sensitive to the noise of the current at a single sampling point, and random spikes in the servo current will be easily misread as stress increases. If the window is too long, the response to newly emerging stress surges will be sluggish, and the peak time will be shifted backward as a whole. The slope of the stress invalid axis is written as a reference label in the slope record of each axis. The slope value is fully retained but does not participate in the threshold comparison for subsequent principal stress axis identification.

[0073] Based on the slope records of each axis, the maximum slope value of each axis and the corresponding time are aggregated to form a curing stress record. The slope time sequence in the slope record of each axis is scanned one axis at a time. After extracting the maximum value of the slope of each axis in the entire late curing period and the time of its occurrence, it is combined with the absolute value of the increment of the current increment sequence of the corresponding axis at the time of the maximum slope to form the curing stress feature item of that axis. The curing stress feature items of all axes are summarized to form the curing stress record. The order of the maximum slope times given in the slope records for each axis reflects the relative progress of the curing stress reaching its peak rate in each axis. Taking the asynchronous curing of two adhesive dots before and after the lens as an example, the adhesive dot that is first exposed to UV radiation shrinks first, and the current increment slope of the axis containing the principal direction of its shrinkage force reaches its peak first. The maximum slope time of the corresponding axis is significantly earlier than that of the other axes. The axis with the earliest maximum slope time is mostly the axis with the largest component of the principal direction of curing shrinkage, and this axis has the highest candidate weight in the principal stress axis identification stage. The time sequence is stored in the curing stress record along with the curing stress feature entries, and is used to add time sequence weight when synthesizing the principal stress axes. The axis with a significantly delayed maximum slope time is mostly because the shrinkage force component on that axis is small, or the stress of the later cured adhesive dot is delayed by elastic transmission from the adjacent axis. When comparing the maximum slope value of each axis in the curing stress record with the threshold of the principal stress axis identification stage, the maximum slope value corresponding to the fixture fluctuation annotation frame in each axis slope record is replaced by the corrected value after window smoothing. The peak slope of each axis aggregated by the curing stress record only contains the actual stress accumulation, without the contribution of fixture interference.

[0074] The principal stress axis is obtained by identifying the axis of sudden increase in current increment slope using the curing stress record. The maximum slope value of each axis in the curing stress record is compared axis by axis with the sudden increase identification threshold. Axis axes exceeding the threshold are identified as axes of sudden slope increase. The sudden increase identification threshold is taken as the 75th quantile of the maximum slope value distribution of each axis in historical normal curing batches. Only axes significantly exceeding the normal curing stress rate are considered as candidates for sudden increase axes. When multiple axes are simultaneously identified as having sudden slope increases, the direction vector of the principal stress axis is synthesized by normalizing the maximum slope values ​​of each axis within the set of sudden increase axes. The synthesis weight is based on the normalized slope values ​​with a time-series factor added. For example, if the maximum slope of the X-axis occurs 15 sampling points earlier than that of the Y-axis, the X-axis weight is increased by approximately 15% on top of the normalized slope, while the Y-axis weight is correspondingly decreased. The principal stress axis thus deflects towards the X-axis direction where stress accumulated earlier, taking into account both intensity and time-series information, pointing to the direction where the curing stress is most concentrated among the axial components. Significant differences in curing shrinkage rates exist between different adhesive models. Batches with higher linear shrinkage rates show generally higher maximum slope values ​​across all axes in the curing stress records. If the threshold is fixed and not updated with each batch, most axes in high-shrinkage batches will exceed the threshold, resulting in an excessively wide principal stress axis identification range and decreased directional resolution. Before each new batch of adhesive is used, the sudden increase identification threshold is updated based on the peak values ​​of the slope records for each axis from the standard parts tested in that batch. Only then can the absolute value of the threshold keep up with the curing stress characteristics of the current batch. The stress-ineffective axis marked in the curing stress record does not participate in the slope comparison and synthesis calculation of the principal stress axes, but its slope reference value is retained to verify the perpendicularity of the principal stress axis direction vector to the stress-ineffective axis. When the angle between the two deviates from 90° by more than 8°, the curing stress transmission path does not match the expectation, and a directional deviation warning mark is added to the principal stress axis.

[0075] The precision assembly command is output by calculating the axial residual displacement correction amount from the principal stress axis. The direction vector of the principal stress axis is jointly analyzed with the final value of the centroid offset recorded in the last 6 frames after UV curing. The projection component of the residual deviation of the final value relative to the optimal alignment pose in the principal stress axis direction represents the residual displacement in that direction. This residual represents the portion of the lens position that has not returned to the optimal alignment pose after the execution of the transient compensation command and the completion of the subsequent curing process. The difference between the final value and the residual deviation written after S104 compensation gives the newly added drift amount in the later stages of curing. Whether the direction of the newly added drift matches the projection direction of the principal stress axis also verifies the direction judgment of the stress analysis. The negative value of the residual displacement amount constitutes the position correction increment for each axis. The correction increment is superimposed on the actual coordinates of the lens at the time of curing completion to obtain the correction target coordinates. The command for the motion platform to move towards the correction target coordinates is the core execution parameter of the precision assembly command. After the adhesive dots have fully cured, the shear modulus of the adhesive layer between the lens and the substrate is much higher than that of the soft adhesive. When the motion platform executes precision assembly instructions, it must advance at an extremely low speed. If the speed is too high, the shear stress of the adhesive layer will exceed the elastic deformation range, and microcracks will appear inside the adhesive layer. The upper limit of the speed is calculated based on the ratio of the shear modulus after the adhesive has cured to the allowable deformation. The motion speed parameters of each axis in the precision assembly instructions are written with this allowable speed as the upper limit. When the principal stress axis carries the warning label for directional deviation, the directional reliability of the residual displacement decreases. In this case, the precision assembly instructions aim to minimize the absolute value of the residual deviation and appropriately reduce the correction increment of each axis. The reduction coefficient is positively correlated with the sine value of the directional deviation angle. If the directional information is uncertain, a conservative small correction is performed first, and a decision on whether to supplement the complete correction amount is made after the next round of spot acquisition and verification.

[0076] To implement the above-described method embodiment, a precision assembly method for a miniature collimating lens and optical fiber is provided to achieve the corresponding functions and technical effects. See also... Figure 2 , Figure 2 This diagram illustrates a structural block diagram of a precision assembly device 200 for a miniature collimating lens and an optical fiber according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The precision assembly device 200 for a miniature collimating lens and an optical fiber according to an embodiment of this application includes:

[0077] Data acquisition module 201 is used to acquire far-field spot data and glue dot morphology data, identify the moment when the asymmetric vector direction of the spot is reversed based on the far-field spot data to generate a misalignment crossing record, and analyze the side contour center offset direction based on the glue dot morphology data to determine the glue dot shrinkage estimate.

[0078] The alignment search module 202 is used to identify the boundary of the asymmetric vector reversal region based on the misalignment crossing record to form a zero crossing axis, and to obtain the optimal alignment pose by fitting the position of the maximum coupling efficiency in the zero-crossing scan using the zero crossing axis.

[0079] The pre-biased gluing module 203 is used to implement the glue dot shrinkage estimation reverse pre-biased compensation to determine the pre-biased compensation posture based on the optimal alignment posture, verify the light spot index tolerance of the pre-biased compensation posture to obtain the pre-biased verification result, and determine the multi-point symmetrical gluing position and establish the glue dot distribution scheme through the pre-biased verification result.

[0080] The curing monitoring module 204 is used to collect the real-time offset of the centroid of the UV curing process according to the glue dot distribution scheme to generate a curing drift record, use the curing drift record to identify the moment of sudden increase in offset rate to obtain the gelation initiation parameter set, and calculate the offset that can be compensated in the soft glue state from the gelation initiation parameter set to output a transient compensation command.

[0081] The stress quality control module 205 is used to analyze the slope of the servo drive current increment of each axis in the later stage of curing based on the transient compensation command to form a curing stress record, use the curing stress record to identify the axis of sudden increase in current increment slope to obtain the principal stress axis, and calculate the axial residual displacement correction amount from the principal stress axis to output a precision assembly command.

[0082] The aforementioned precision assembly device 200 for a miniature collimating lens and an optical fiber can implement a precision assembly method for a miniature collimating lens and an optical fiber according to the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining contents of this application embodiment can be referred to the contents of the above method embodiments, and will not be repeated in this embodiment.

[0083] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0084] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A method for precision assembly of a miniature collimating lens and an optical fiber, characterized in that, include: Collect far-field spot data and glue dot morphology data. Based on the far-field spot data, identify the moment when the asymmetric vector direction of the spot reverses and generate a misalignment crossing record. Based on the glue dot morphology data, analyze the side profile center offset direction to determine the glue dot shrinkage estimate. Based on the misalignment crossing record, the boundary of the asymmetric vector reversal region is identified to form a zero crossing axis. The optimal alignment pose is obtained by fitting the position of maximum coupling efficiency in the zero-crossing scan using the zero crossing axis. The optimal alignment pose is used to implement the glue dot shrinkage prediction amount reverse pre-bias compensation to determine the pre-bias compensation pose. The pre-bias compensation pose is used to verify the light spot index tolerance to obtain the pre-bias verification result. The pre-bias verification result is used to determine the multi-point symmetrical glue application position and establish the glue dot distribution scheme. Based on the glue dot distribution scheme, the real-time offset of the centroid of the UV curing spot is recorded to generate a curing drift record. The curing drift record is used to identify the moment of sudden increase in offset rate to obtain the gelation initiation parameter set. The offset that can be compensated in the soft glue state is calculated from the gelation initiation parameter set, and a transient compensation command is output. Based on the transient compensation command, the slope of the servo drive current increment of each axis in the later stage of curing is analyzed to form a curing stress record. The curing stress record is used to identify the axis of sudden increase in current increment slope to obtain the principal stress axis. The axial residual displacement correction amount is calculated from the principal stress axis and a precision assembly command is output.

2. The method according to claim 1, characterized in that, The step of generating a misaligned crossing record based on the moment when the asymmetric vector direction of the light spot reverses, identified by the far-field light spot data, includes: The beam direction angle sequence is obtained by extracting the beam ellipse direction angle frame by frame from the far-field beam data; The set of reversal times is determined by identifying the alternating switching times of the positive and negative signs of the light spot direction angle sequence; The crossing position coordinates are established by recording the six-dimensional coordinates of the motion platform at each moment using the inverted time set; Based on the coordinates of the crossing locations, the inaccurate crossing record is generated by summarizing the directional angle reversal magnitude of each crossing point.

3. The method according to claim 1, characterized in that, The step of identifying the boundary of the asymmetric vector reversal region based on the misaligned crossing record to form the zero crossing axis includes: The spatial distribution of crossing points is obtained by extracting the six-dimensional coordinate weighted distribution of the crossing points from the inaccurate crossing records; A rotation-displacement coupling decoupling transformation is applied to the spatial distribution of the crossing points to obtain a decoupled coordinate distribution; Perform principal inertial direction calculation on the decoupled coordinate distribution to establish the traversal principal direction vector; The zero crossing axis is formed by mapping the main crossing direction vector to the platform motion coordinate system.

4. The method according to claim 1, characterized in that, The step of determining the estimated amount of adhesive dot shrinkage based on the side profile center offset direction analyzed from the adhesive dot morphology data includes: The side profile distribution is generated by extracting the side height contour and bottom contact line coordinates using the glue dot morphology data; The height centroid offset is obtained by calculating the profile height and weighted centroid coordinates for the side profile distribution. A shrinkage feature set is established based on the asymmetric axis deviation angle of the analytical volume distribution for the aforementioned height centroid offset. Based on the shrinkage feature set, the maximum asymmetric amplitude of the side height is calculated to obtain the estimated amount of glue dot shrinkage.

5. The method according to claim 1, characterized in that, The step of applying the adhesive dot shrinkage prediction estimate to the optimal alignment pose to determine the pre-bias compensation pose includes: The axial shrinkage component is obtained by analyzing the shrinkage distribution values ​​of each degree of freedom for the estimated shrinkage of the adhesive dot. The axial pre-offset vector is formed by calculating the equal reverse offset of each axial component using the axial contraction component. Based on the axial pre-bias vector, identify the axial pre-bias combinations within the spot quality tolerance to establish a feasible pre-bias interval; Based on the feasible pre-bias interval, the optimal alignment pose and the optimal pre-bias amount for each axis are synthesized to construct the pre-bias compensation pose.

6. The method according to claim 1, characterized in that, The step of using the solidification drift record to identify the moment of sudden increase in offset rate to obtain the gelation initiation parameter set includes: The offset rate sequence is generated by calculating the centroid displacement difference components of adjacent frames from the solidified drift record; The rate acceleration sequence is determined by analyzing the differential rate of change in the rate time series of the offset rate sequence; Based on the rate acceleration sequence, a sudden increase event set is established at the first frame moment when the acceleration continuously exceeds the threshold and the drift is unidirectional and irreversible. Based on the burst event set, the starting time of the first burst event and the offset at that time are used to obtain the gelation initiation parameter set.

7. The method according to claim 1, characterized in that, The process of generating curing stress records based on the slope of the servo drive current increment for each axis during the later stages of curing, derived from the transient compensation command analysis, includes: The current reference value is obtained by identifying the steady-state reading of the servo current of each axis after the compensation is completed using the transient compensation command. Based on the current reference value, the timing sequence of the servo current increment of each axis in the later stage of curing is calculated to form the current increment sequence. For the current increment sequence, analyze the rate of change of current increment per unit time of each axis and establish the slope record of each axis; Based on the slope of each axis, the maximum slope value of each axis and the corresponding time are used to form a curing stress record.

8. The method according to claim 1, characterized in that, The step of calculating the compensable offset in the soft gel state from the gelation initiation parameter set and outputting a transient compensation command includes: The drift step length reference is obtained by calculating the maximum single-step displacement of the centroid of the monitoring window spot before gelation using the gelation initiation parameter set. The instantaneous deviation is determined by calculating the centroid alignment deviation at the gelation moment based on the set of gelation initiation parameters. Based on the analysis of the ratio between the drift step size reference and the instantaneous deviation, and with the allowance for curing rebound margin, a compensable coefficient is obtained. Based on the compensable coefficient, the compensation amount for each axis is synthesized, and a transient compensation command is output.

9. The method according to claim 4, characterized in that, The establishment of a shrinkage feature set based on the asymmetric axis deviation angle of the analytical volume distribution for the height centroid offset includes: The volume offset principal axis is obtained by calculating the deviation angle of the principal inertial axis of the volume distribution based on the aforementioned height center of gravity offset. The constraint correction factor is formed by calculating the shrinkage constraint weights of adjacent glue points and wetting boundary using the volume offset principal axis. The constraint correction factor is used to adjust the contraction axis from the offset direction to the constrained direction to construct an offset contraction axis. Based on the biased contraction axis, the direction cosine and amplitude of the contraction components of each degree of freedom are derived to obtain the contraction feature set.

10. A precision assembly device for a miniature collimating lens and an optical fiber, characterized in that, include: The data acquisition module is used to acquire far-field spot data and glue dot morphology data. Based on the far-field spot data, it identifies the moment when the asymmetric vector direction of the spot reverses and generates a misalignment crossing record. Based on the glue dot morphology data, it analyzes the offset direction of the side contour center to determine the estimated amount of glue dot shrinkage. The alignment search module is used to identify the boundary of the asymmetric vector reversal region based on the misalignment crossing record to form a zero crossing axis, and to obtain the optimal alignment pose by fitting the position of the maximum coupling efficiency in the zero-crossing scan using the zero crossing axis. The pre-biased gluing module is used to perform reverse pre-biased compensation on the estimated amount of glue dot shrinkage in the optimal alignment pose to determine the pre-biased compensation pose, verify the compliance of the light spot index with the pre-biased compensation pose to obtain the pre-biased verification result, and determine the multi-point symmetrical gluing position and establish the glue dot distribution scheme through the pre-biased verification result. The curing monitoring module is used to collect the real-time offset of the centroid of the UV curing spot according to the glue dot distribution scheme to generate a curing drift record. The curing drift record is used to identify the moment of sudden increase in offset rate to obtain the gelation initiation parameter set. The offset that can be compensated in the soft glue state is calculated from the gelation initiation parameter set and a transient compensation command is output. The stress quality control module is used to analyze the slope of the servo drive current increment of each axis in the later stage of curing based on the transient compensation command to form a curing stress record. The curing stress record is used to identify the axis of sudden increase in current increment slope to obtain the principal stress axis. The axial residual displacement correction amount is calculated from the principal stress axis and a precision assembly command is output.

Images