Control method and system of double-station transceiving integrated coupling equipment

Abstract

The invention provides a control method and system of double-station transceiving integrated coupling equipment, and the method comprises the steps: obtaining defect coordinates and pinhole characteristic data of a target region, carrying out the analysis to obtain a transmittance abnormal region, generating an initial path of a spiral search algorithm according to the transmittance abnormal region, adjusting the search step size and angle, and obtaining a preliminary scanning boundary of a to-be-scanned region; analyzing the closeness degree of the scanning path and the defect, if the closeness degree is within a preset range, determining a power reduction condition according to the defect coordinate, adjusting the power, processing the light spot to obtain the size, and calculating a broadening coefficient; and extracting dynamic change data of the light spot size from the distributed power distribution, scanning the defect area to obtain a heat accumulation risk real-time distribution diagram, and dynamically adjusting the light spot size and power until the peak coupling power stability recovers to a preset range.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a control method and system for a dual-station transceiver integrated coupling device. Background Art

[0002] Optical communication technology, as a core pillar of modern information transmission, directly determines the efficiency and reliability of data communications. In the field of optical module manufacturing, the development of dual-station transceiver coupling equipment is particularly critical. Because it integrates both transmit and receive functions, it places extremely high demands on coupling accuracy and stability. With increasing communication speeds and the trend toward device miniaturization, the application of miniature aspheric spherical lenses is becoming increasingly widespread. The processing quality of their anti-reflection coatings has become a crucial factor influencing beam steering and coupling efficiency. Optimizing the beam steering strategy at the coupling station has become a key focus in current research and production, driving breakthroughs in optical module performance. However, existing coupling equipment often relies on static scanning paths and fixed power distributions when processing anti-reflection-coated lenses. This approach exhibits significant shortcomings when dealing with localized transmittance anomalies caused by coating defects, such as pinholes. Traditional solutions struggle to adapt to the asymmetric scanning requirements caused by defects and are unable to effectively address shifts in the peak coupling power point, which can lead to reduced efficiency and even device damage. Existing control systems often rely on passive adjustment of preset parameters and lack the ability to dynamically respond to defects, limiting the flexibility and reliability of the coupling process. In this context, the core challenges faced by the research focus on the interference of defect characteristics on beam control. Specifically, pinhole defects increase the asymmetry of the scanning area required by the coupling station, reduce the spatial stability of the peak coupling power point, and at the same time, the defect may bring potential risks due to local heat accumulation. These factors are coupled with each other, making it difficult for traditional linear scanning strategies to accurately capture the optimal coupling point, and unable to achieve effective power and spot control near the defect area, which in turn leads to the technical difficulty of how to optimize the scanning path and energy distribution in a dynamic environment. Therefore, how to design a control system after receiving the defect coordinates and characteristic data so that it can dynamically adjust the laser scanning path at the coupling station, give priority to exploring the area around the defect, and simultaneously optimize the power distribution and spot size has become a key issue in improving coupling efficiency and equipment stability. Summary of the Invention

[0003] The present invention provides a control method for a dual-station transceiver integrated coupling device, which mainly includes: obtaining defect coordinates and pinhole characteristic data of a target area, analyzing to obtain an abnormal transmittance area, and generating an initial path of a spiral search algorithm based on the data, adjusting the search step size and angle, and obtaining a preliminary scanning boundary of the area to be scanned;

[0004] Analyze the proximity of the scanning path to the defect. If it is within the preset range, determine the power reduction condition based on the defect coordinates and adjust the power. Process the spot to obtain its size and calculate the broadening coefficient.

[0005] The actual spot size is adjusted using the adjusted power and stretch factor. An initial small-step spiral scan is performed within the preset range of the defect coordinates according to the adjusted search step size and angle. During this process, real-time monitoring is performed to determine whether the heat accumulation risk exceeds the preset threshold.

[0006] If the heat accumulation risk exceeds the threshold, the corresponding area is marked as a high-risk area, the coordinate position of the high-risk area is obtained, and based on the value of the heat accumulation risk exceeding the threshold, the correction value is calculated, the spiral search radius increment is adjusted, and a new scanning path is generated to avoid the high-risk area;

[0007] Update the current scanning path according to the new scanning path and allocate the power distribution. If the actual spot size decreases after the current scanning path is updated, the spatial coordinate data of the peak coupled power point is collected and time series analysis is performed to obtain the spatial coordinate change trend;

[0008] Analyze the trend of spatial coordinate changes and calculate the drift rate of the peak coupled power point position. If the drift rate exceeds the preset stability threshold, indicating that the current positioning stability is insufficient to meet the preset accuracy requirements, increase the number of spiral search iterations.

[0009] The dynamic change data of the spot size is extracted from the allocated power distribution, the defect area is scanned, and a real-time distribution map of the heat accumulation risk is obtained. The spot size and power are dynamically adjusted until the peak coupled power stability returns to the preset range.

[0010] Furthermore, the defect coordinates and pinhole characteristic data of the target area are obtained, analyzed to obtain the transmittance abnormality area, and based on this, the initial path of the spiral search algorithm is generated. The search step size and angle are adjusted to obtain the preliminary scanning boundary of the area to be scanned. The method includes: dividing the target area into a number of sampling blocks of 10×10 pixels using a grid method. Within each sampling block, a sequence of defect coordinate points and pinhole transmittance values ​​are obtained using a micro-displacement scanning device. The transmittance values ​​are averaged and, based on a preset transmittance threshold, points within the block with a transmittance below 85% are marked as abnormal points. A photoelectric sensor array is used to scan the light intensity within the sampling block, obtaining a sequence of light intensity grayscale values. A grayscale value change rate curve is calculated, and the coordinates of the starting point of the spiral search path are calculated based on the distribution density of the abnormal points. The search step size is dynamically adjusted based on the grayscale value change rate curve, and the initial search angle is set to 5 degrees. Based on the search angle and the adjusted search step size, a spiral trajectory sampling point sequence is generated within the region sampling boundary using Euclidean distance calculation. The sampling point sequence is then grouped according to spatial distance using a hierarchical clustering method, and points are added to areas where the sampling point density is less than a preset threshold. Curve fitting is performed on the grouped sampling point sequence, and the boundary contour curve equation for each group of sampling points is calculated using cubic spline interpolation to obtain a sequence of boundary point coordinates on the contour curve. A closed polygon is constructed based on the boundary point coordinate sequence, and a convex hull algorithm is used to calculate the minimum circumscribed polygon as the target scanning boundary. The vertex coordinate sequence of this polygon is then extracted as the scanning area boundary coordinates.

[0011] Furthermore, the proximity of the scanning path to the defect is analyzed. If it is within a preset range, the power reduction condition is determined and adjusted based on the defect coordinates. The spot size is then processed to determine its size, and the broadening coefficient is calculated. This process involves calculating the shortest distance from the defect location from the scanning path point sequence using a four-neighborhood grid distance calculation method. The reflected light intensity curve of the defect region is acquired using an array photodetector at a wavelength of 800 nanometers. Points with distances less than twice the defect characteristic size are marked as path proximity points. The depth coefficient of the defect region is calculated based on the light intensity curve. When the depth coefficient exceeds a preset baseline value, a Gaussian attenuation function curve is constructed from the depth coefficient, with the function variance set proportional to the depth coefficient. The power attenuation ratio is calculated using the Gaussian function. The laser output power is adjusted based on the power attenuation ratio. A photomultiplier tube is used at a gain of 1000 to acquire energy density distribution data across the spot cross section. A grayscale image of the spot is generated from this density data. Maximum entropy segmentation is performed on the grayscale image to extract the coordinates of edge contour points. The distance from the contour point to the center of the spot is calculated to obtain a spot radius sequence. The maximum value in the radius sequence is extracted as the spot diameter value. The overlap between the actual processing area and the theoretical area is calculated using the defect area boundary curve and the spot diameter value. The overlap curve is fitted by cubic spline interpolation, and the overlap curve is integrated to obtain the broadening coefficient value.

[0012] Furthermore, the adjusted power and stretch factor are used to adjust the actual spot size. An initial small-step spiral scan is performed within a preset defect coordinate range using the adjusted search step size and angle. During this process, real-time monitoring is performed to determine whether the heat accumulation risk exceeds a preset threshold. This involves calculating the theoretical spot diameter based on the adjusted power and stretch factor. The actual spot diameter is then adjusted using a focusing lens assembly within a focal length range of 50 to 200 mm in 5-mm increments. The scan coverage radius is determined by multiplying the actual spot diameter by a factor of 1.2. A quarter-arc-length subdivision method is used to generate spiral scan trajectory points within the scan coverage area, with the spacing between adjacent trajectory points set to one-tenth of the spot diameter. A temperature distribution map of the defect area is acquired using an infrared temperature sensor array at a sampling frequency of 100 Hz. The temperature distribution map is discretized into a 10×10 micron grid, and the temperature gradient is calculated for each grid point. Heat flux density is calculated based on the Fourier heat conduction equation, and the heat accumulation per unit time is derived from the heat flux density and the grid area. A random forest algorithm is used to predict heat accumulation sequences. Input features include temperature gradient, heat flux, and scan time. The algorithm outputs a predicted heat accumulation risk value and determines whether the predicted value exceeds a preset threshold. If the predicted heat accumulation risk value exceeds the threshold, the coordinates of the temperature peak point are extracted. The coordinates of the isothermal lines within the radius of the peak point are calculated. The scan path coordinate sequence is then regenerated using the multiplicative grid method for the scan points within the isothermal range.

[0013] Furthermore, if the heat accumulation risk exceeds a threshold, the corresponding area is marked as a high-risk zone. The coordinates of the high-risk zone are obtained, and a correction value is calculated based on the value of the heat accumulation risk exceeding the threshold. The spiral search radius increment is adjusted to generate a new scanning path that avoids the high-risk zone. This includes: judging the heat accumulation risk monitoring data against a preset threshold. If the heat accumulation risk value exceeds the threshold, a temperature similarity region growing method is used to mark connected areas with a temperature difference of less than 5 degrees Celsius. A boundary tracking method is used to extract the coordinate sequence of the boundary points of the high-risk area to obtain a high-risk zone contour map. The centroid coordinates are extracted from the high-risk zone contour map as the region center point. A safety distance coefficient is calculated based on the ratio of the measured heat accumulation risk value to the threshold. The safety distance coefficient is mapped to a range of 0.5 to 2 as the spiral search path increment adjustment parameter. The original spiral path is reconstructed based on the adjustment parameter. The weighted distance value from each path sampling point to the center of the high-risk zone is calculated based on the Gaussian distance. Sampling points with a weighted distance value less than twice the spot diameter are marked. Path avoidance constraints are constructed for the marked sampling point sequence. The least squares method is used to fit a new path curve equation, and the curve nodes are adjusted based on the curvature continuity condition. The node tangent vectors are calculated based on the adjusted curve equation. The sampling interval is set to one-tenth of the spot diameter, and a new scanning path sampling point sequence is extracted. Cubic spline interpolation is performed on the sampling point sequence to obtain the coordinates of the smooth scanning trajectory.

[0014] Furthermore, the current scanning path is updated based on the new scanning path, and a power distribution is assigned. If the actual spot size decreases after the current scanning path is updated, the spatial coordinate data of the peak coupled power point is collected and time series analysis is performed to determine the spatial coordinate change trend. This includes discretizing the new scanning path using a square grid with a grid spacing of one-tenth the spot diameter. A sequence of power values ​​at the path points is obtained using an optical power detector at a sampling frequency of 100 Hz. A two-dimensional power distribution matrix is ​​constructed based on the power value sequence, and power configuration parameters are extracted from the matrix. The laser output power is adjusted based on the power configuration parameters. A spot profilometer with a resolution of 0.1 micron is used to obtain transverse and longitudinal light intensity distribution curves. The Sobel edge detection operator is used to extract the coordinate points of the spot boundary. The spot size is calculated by performing an ellipse fitting using the least squares method on the boundary coordinate points. The fitted spot size is compared with the original size. If the spot size decreases by more than 10%, a point with a power value greater than twice the average value is selected from the power distribution matrix as the coupled power peak point, and the three-dimensional coordinate data of the peak point is recorded. A time series was constructed from the peak point coordinate data at a 10-millisecond sampling interval. Coordinate sequence features were extracted using a three-hidden layer structure within a long-short-term memory network, with a time series prediction window length of 100 data points. A state-space equation was established based on the extracted time series features, and a Kalman filter was used to optimize the coordinate predictions. The coordinate drift velocity and acceleration were calculated using recursive least squares, yielding a characteristic vector for the spatial coordinate change trend.

[0015] Furthermore, the spatial coordinate trend is analyzed and the drift rate of the peak coupled power point position is calculated. If the drift rate exceeds a preset stability threshold, indicating that the current positioning stability is insufficient to meet the preset accuracy requirements, the number of spiral search iterations is increased. This includes: segmenting the spatial coordinate sequence using a time window of fixed length of 100 sampling points, calculating the three-dimensional displacement of the peak coupled power point within each window through forward difference, dividing the displacement by the corresponding time interval to obtain the instantaneous velocity sequence, and constructing a drift curve from the instantaneous velocity sequence. A Fourier transform is performed on the drift curve, with a frequency threshold set to one-tenth of the sampling frequency. The Fourier coefficients are screened, and the drift sequence is decomposed using an autoregressive sliding average model with an autoregressive order of 3 and a sliding average order of 2. From the decomposition results, components with a standard deviation greater than twice the mean are extracted as random drift, and components with periodic characteristics are extracted as periodic drift. The integrated drift rate is calculated through vector superposition and exponentially smoothed. The smoothed drift rate is determined to determine whether it exceeds the stability threshold. If so, the least squares method is used to calculate the feedback gain coefficient, and a proportional-integral controller is used to compensate the spiral search step size and angle increment. A multi-layer iterative structure is constructed based on the compensated parameters. In each iteration, the search range is subdivided into four sub-regions. New scan path nodes are recursively generated for each sub-region, and the node density is balanced using the Thiessen polygon method.

[0016] Furthermore, the dynamic change data of the spot size is extracted from the allocated power distribution, the defect area is scanned, and a real-time distribution map of the heat accumulation risk is obtained. The spot size and power are dynamically adjusted until the peak coupled power stability returns to the preset range. This includes: constructing a light intensity matrix based on the allocated power distribution data, using a two-dimensional Gaussian function to fit the spot half-maximum width parameter, interpolating the spot size sampling data using a cubic spline curve, and calculating the spot size change from the interpolation result. An infrared thermal imager with a resolution better than 0.1 degrees Celsius is used to obtain temperature field data. The spatial gradient value of the temperature field is calculated based on the central difference method. A heat diffusion rate matrix is ​​constructed through gridding processing, and the matrix data is normalized to obtain a thermal field distribution map. The sliding window method is used to extract local features from the thermal field distribution map. Feature extraction is performed based on a five-layer convolutional network structure with a convolution kernel size of 3×3, maximum pooling in the pooling layer, and a fully connected layer to output the heat accumulation risk value. The peak temperature position is determined from the heat accumulation risk value, and pulse width modulation is used to dynamically compensate the laser output power. An objective function is constructed based on the deviation between the spot radius and the temperature gradient, and the focal length compensation value is calculated through gradient iteration. A high-speed photodetector is used to acquire peak coupled power data. The sampling period is set to 1 millisecond, and the sampled data is exponentially smoothed. The ratio of the power fluctuation standard deviation to the mean is calculated as the coupling stability indicator. Based on the deviation between the coupling stability indicator and a preset threshold, a proportional-integral controller is constructed to perform closed-loop adjustment of the spot size and power parameters. Iteration is continued until the stability indicator falls below the preset threshold.

[0017] The present invention provides a control system for a dual-station transceiver integrated coupling device, which mainly includes:

[0018] The defect coordinate and pinhole characteristic data acquisition module is used to obtain the defect coordinates and pinhole characteristic data of the target area, analyze the abnormal transmittance area, and generate the initial path of the spiral search algorithm based on this data. The search step size and angle are adjusted to obtain the preliminary scanning boundary of the area to be scanned;

[0019] The module for analyzing abnormal transmittance areas and generating the initial path is used to analyze the proximity of the scanning path to the defect. If the proximity is within a preset range, the module determines the power reduction condition based on the defect coordinates and adjusts the power. The module also processes the light spot to obtain its size and calculates the broadening coefficient.

[0020] The scanning path and defect proximity analysis module is used to adjust the actual spot size using the adjusted power and stretch factor. It performs an initial small-step spiral scan in the preset range of the defect coordinates according to the adjusted search step size and angle. During this process, it monitors in real time whether the heat accumulation risk exceeds the preset threshold.

[0021] The power adjustment and spot size processing module is used to mark the corresponding area as a high-risk area if the heat accumulation risk exceeds the threshold, obtain the coordinate position of the high-risk area, and calculate the correction value based on the value of the heat accumulation risk exceeding the threshold, adjust the spiral search radius increment, and generate a new scanning path to avoid the high-risk area;

[0022] The heat accumulation risk monitoring and high-risk area marking module is used to update the current scanning path according to the new scanning path and allocate the power distribution. If the actual spot size decreases after the current scanning path is updated, the spatial coordinate data of the peak coupled power point is collected and time series analysis is performed to obtain the spatial coordinate change trend;

[0023] A new scanning path generation and power distribution allocation module is used to analyze the trend of spatial coordinate changes and calculate the drift rate of the peak coupled power point position. If the drift rate exceeds the preset stability threshold, indicating that the current positioning stability is insufficient to meet the preset accuracy requirements, the number of spiral search iterations is increased;

[0024] The spot size dynamic adjustment and peak coupled power stability recovery module is used to extract the dynamic change data of the spot size from the allocated power distribution, scan the defect area, obtain the real-time distribution map of the heat accumulation risk, and dynamically adjust the spot size and power until the peak coupled power stability is restored to the preset range.

[0025] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0026] The present invention discloses a control method for a dual-station transceiver integrated coupling device. The method first obtains the defect coordinates and pinhole characteristic data to generate an initial spiral search path. Then, the laser power and spot size are dynamically adjusted according to the defect position, and a small-step spiral scan is performed. During the scanning process, the present invention monitors the heat accumulation risk in real time, and if it exceeds the threshold, a new path is generated to avoid the high-risk area. At the same time, by analyzing the spatial coordinate change trend of the peak coupling power point, the positioning stability is evaluated, and the number of search iterations is increased if necessary. Finally, the present invention dynamically adjusts the spot size and power according to the real-time distribution map of the heat accumulation risk until the preset peak coupling power stability is achieved. This method can achieve accurate and efficient scanning of the defect area while ensuring safety, thereby improving the accuracy and reliability of laser processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 The present invention is a flowchart of a control method of a dual-station transceiver integrated coupling device.

[0028] Figure 2 This is a structural diagram of a control system of a dual-station transceiver integrated coupling device of the present invention. DETAILED DESCRIPTION

[0029] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments derived by those skilled in the art based on the embodiments in this specification without creative effort shall fall within the scope of protection of this specification.

[0030] like Figure 1 In this embodiment, a control method for a dual-station transceiver integrated coupling device may specifically include:

[0031] S101, obtaining the defect coordinates and pinhole characteristic data of the target area and analyzing them to obtain the transmittance abnormality area, based on which an initial path of the spiral search algorithm is generated and the search step size and angle are adjusted to determine the preliminary scanning boundary of the area to be scanned.

[0032] S1011. The target area is divided into multiple sampling blocks using a grid division method. The transmittance data of each block is collected using a micro-displacement scanning device and compared with a preset threshold to mark abnormal points. The grayscale value sequence is then acquired using a photoelectric sensor array and the coordinates of the starting point of the spiral search are calculated. In an embodiment of the present invention, the grid division is based on 10×10 pixels. The transmittance value is collected point by point using a micro-displacement scanning device, and points below 85% of the preset threshold are marked as abnormal points. Subsequently, an 8×8 photoelectric sensor array is used to scan the light intensity of each block, generate a grayscale value sequence and calculate the rate of change curve. The starting point coordinates are determined based on the distribution density of the abnormal points. The center of the area with the highest density is usually selected as the starting point.

[0033] S1012. Generate a spiral trajectory sampling point sequence based on the starting point coordinates and the preset search angle. Use a hierarchical clustering method to spatially group the sampling points and use an interpolation method to calculate the boundary contour curve. Then, construct a closed polygon and use a convex hull algorithm to determine the boundary coordinates of the scanning area. In this embodiment of the present invention, the initial search angle is set to 5 degrees, and the search step size is adaptively adjusted according to the grayscale value change rate, reducing it to 1 pixel in areas with drastic changes and increasing it to 3 pixels in areas with gentle changes. After the sampling point sequence is generated, hierarchical clustering is performed based on the Euclidean distance as the standard, and the clustering threshold is set to 2 pixels. Points are added to areas with insufficient density. Afterwards, the boundary curve is fitted using a cubic spline interpolation method with a node spacing of 5 pixels. Finally, the vertex coordinates of the minimum circumscribed polygon are extracted as the scanning boundary using a convex hull algorithm.

[0034] In an embodiment of the present invention, the grid division of the target area is the basis for achieving precise positioning. Taking a 100×100 pixel area as an example, it is divided into 100 10×10 pixel blocks. Each block obtains the defect coordinates and transmittance data through micro-displacement scanning. The threshold of 85% is based on experimental statistics to ensure the accuracy of marking. The grayscale value range of the photoelectric sensor array is 0 to 255, and the grayscale change rate of the defective area can reach more than 30%, which is much higher than the 5% of the normal area, providing a reliable basis for the selection of the starting point. The generation of the spiral search path fully considers accuracy and efficiency, and the dynamic adjustment of the step size ensures detailed coverage of the defective area. The combination of hierarchical clustering and interpolation method makes the boundary contour smooth and accurate. The final generated scanning boundary usually contains 8 to 12 vertices, which completely defines the spatial range of the area to be scanned.

[0035] It is understandable that the embodiments of the present invention do not impose too many restrictions on the grid size or the parameter settings of the interpolation method. Technicians can adjust them according to the actual scenario. For example, in a higher resolution scenario, the grid can be reduced to 5×5 pixels, or the interpolation node spacing can be adjusted to 3 pixels to adapt to different coupling requirements.

[0036] In the embodiment of the present invention, the above steps can effectively identify the defect area and generate a preliminary scanning boundary, which lays the foundation for subsequent beam control and power optimization and ensures the accuracy and stability of the coupling process.

[0037] S102, analyzing the proximity between the scanning path and the defect position and determining whether it is within a preset range, determining the power reduction condition according to the defect coordinates and adjusting the laser output power, and processing the spot data to obtain its size and calculate the broadening coefficient to optimize the coupling effect.

[0038] S1021. In an embodiment of the present invention, a four-neighborhood grid distance calculation method is used to extract the shortest distance value to the defect position from the scanning path point sequence and mark the proximity point sequence. The array photodetector is used to collect the reflected light intensity data of the defect area to generate a curve. The depth coefficient is calculated based on the curve and a power adjustment function is constructed to determine the attenuation ratio. The specific distance calculation uses the defect center as the reference point and expands pixel by pixel in the four directions of up, down, left and right to obtain the Euclidean distance value between the path point and the defect. If the distance is less than twice the defect feature size, the point is included in the proximity point sequence. The array photodetector operates in the form of an 8×8 unit array at a wavelength of 800 nanometers and has a response time of less than 1 microsecond. When collecting the reflected light intensity of the defect area, the intensity value in the normal area is usually 70% to 100% of the reference value, while the defect area drops to less than 50% due to the pinhole effect. The depth coefficient is calculated by analyzing the peak-to-valley difference of the light intensity curve. When it exceeds the preset baseline value of 0.5, a Gaussian attenuation function is constructed, and the variance is set to be positively correlated with the depth coefficient. For example, when the depth coefficient is 0.8, the variance is 2.0, and the calculated power attenuation ratio is between 0.6 and 0.9, which is used for power adjustment.

[0039] S1022. After the power is adjusted, the energy density distribution data of the light spot cross section is collected through the photomultiplier tube and a grayscale image is generated. The maximum entropy segmentation method is used to extract the coordinates of the edge contour points of the light spot to determine the value of the light spot diameter. Then, the overlap is calculated in combination with the boundary curve of the defect area and the broadening coefficient is obtained by fitting. In an embodiment of the present invention, the photomultiplier tube operates at a gain of 1000 times, and the energy distribution of the captured light spot shows the characteristics of high center and low edge. After digital processing, a 256-level grayscale image is generated, with the grayscale value of the central area ranging from 220 to 250 and the edge dropping to 50 to 80. The maximum entropy segmentation method determines the optimal segmentation threshold by analyzing the entropy value distribution of the grayscale histogram, for example, segmenting at a grayscale value of 120, extracting the edge contour points, and calculating the distance from each point to the center. The maximum value is taken as the light spot diameter, which is usually between 10 and 15 pixels. Based on the boundary curve and spot diameter of the defect area, the overlap ratio between the actual processing area and the theoretical area is calculated. The discrete overlap data is fitted using cubic spline interpolation to obtain a continuous curve and integrate it. The broadening coefficient results are mostly between 1.2 and 1.5, indicating the degree of match between the processing accuracy and the theoretical design.

[0040] In an embodiment of the present invention, the four-neighborhood grid distance calculation method is an efficient distance measurement method, which ensures the accuracy of the distance value by pixel-by-pixel expansion. Taking a 20×20 pixel area as an example, if the defect is located at the center point coordinate 10,10, the distance value of the adjacent points is 1, the distance of the diagonal points is about 1.414, and so on to form a distance gradient distribution, which is convenient for marking the close point sequence. The application of array photodetectors improves the measurement sensitivity of the reflected light intensity, and its fast response characteristics ensure the real-time data acquisition and provide a reliable basis for the calculation of the depth coefficient. The design of the Gaussian attenuation function fully considers the dynamic changes of the defect depth. By adaptively adjusting the variance value, the power attenuation ratio can accurately reflect the defect characteristics and avoid the risk of thermal damage caused by excessive power.

[0041] The processing of the spot size is particularly critical in the embodiments of the present invention. The energy density data collected by the photomultiplier tube is gray-scaled to intuitively reflect the spatial distribution characteristics of the spot. The maximum entropy segmentation method can accurately separate the edge of the spot under a complex background by maximizing the entropy value, avoiding the misjudgment caused by the traditional threshold method due to light intensity fluctuations. The calculated broadening coefficient not only measures the degree of overlap of the processing area, but also provides a quantitative basis for subsequent path optimization. If the broadening coefficient deviates too far from the ideal value of 1, for example, more than 1.8, it indicates that the spot size is not well matched with the defect area, and the power or scanning step size needs to be further adjusted.

[0042] In this embodiment of the present invention, the above steps achieve dynamic optimization of power and spot size by analyzing the proximity of the scanning path to the defect. This method adaptively adjusts control parameters as defect characteristics change, ensuring that the coupling station maintains efficient and stable operation even when facing pinhole defects. It is understood that the specific intensity curve analysis method or interpolation parameters can be adjusted by technicians based on actual equipment conditions, such as changing the wavelength to 850 nanometers or increasing the gain multiplier to 1200, to meet the needs of different scenarios.

[0043] S103 , optimizing the actual spot size using the adjusted parameters and performing a small-step spiral scan within a preset range of the defect coordinates, while determining whether the heat accumulation risk exceeds a preset threshold by real-time monitoring of the temperature distribution.

[0044] In this embodiment of the present invention, the theoretical spot diameter is calculated based on the adjusted power value and the stretching factor. The actual spot size is then adjusted through the focusing lens assembly. The scan coverage radius is determined by combining the preset coefficients, and a spiral scanning trajectory is generated. An infrared temperature sensor array is then used to collect temperature data and analyze heat flow characteristics to assess risk. If the risk exceeds the specified value, the temperature peak is extracted and the scanning path is replanned. Specifically, the theoretical spot diameter is derived from the power and the stretching factor using an optical formula. The focusing lens assembly is adjusted in 5 mm increments within a focal length range of 50 to 200 mm. The actual spot diameter typically fluctuates between 40 and 60 microns. This is multiplied by a factor of 1.2 to obtain the scan coverage radius, ensuring coverage of the heat-affected zone surrounding the defect. A quarter-arc-length subdivision method is used to generate the spiral scanning trajectory. The spacing between adjacent track points is set to one-tenth of the spot diameter. For example, for a spot diameter of 50 microns, the spacing is 5 microns, ensuring scanning uniformity and coverage. The infrared temperature sensor array operates at a sampling frequency of 100 Hz, with a unit response time of less than 5 milliseconds and a temperature resolution of 0.1 degrees Celsius. The collected temperature distribution reflects the thermal state of the defect area.

[0045] S1031. Discretize the temperature distribution map into a grid of points and calculate the temperature gradient and heat flux. The Fourier heat conduction equation is used to analyze heat transfer characteristics and, in conjunction with a prediction algorithm, determine the risk of heat accumulation. The temperature distribution map is divided into a 10×10 micron grid. For example, 100 monitoring points are generated for a 100×100 micron area. The temperature gradient at each point is calculated using the central difference method. Locally high-temperature areas can have gradients of 5 to 10 degrees Celsius per micron. The Fourier heat conduction equation calculates the heat flux based on the temperature difference at the grid points and the thermal conductivity of the material. Typical values ​​range from 0.1 to 0.5 watts per square micron. Heat accumulation per unit time is further estimated based on the grid area. A random forest algorithm is used for risk prediction. Input features include the temperature gradient matrix, the heat flux sequence, and the scan duration. A risk value is generated through an ensemble learning algorithm of 500 decision trees, achieving a prediction accuracy exceeding 95%. A risk value exceeding 0.8 is considered an over-threshold condition.

[0046] S1032. When the risk of heat accumulation exceeds the standard, the coordinates of the temperature peak point are extracted and the scanning path is regenerated based on the isothermal range to avoid the high-temperature area. In an embodiment of the present invention, the peak point is identified from the temperature distribution diagram. For example, when the temperature exceeds 150 degrees Celsius, it is marked as a high-temperature point. The isothermal line in the area with a radius of 2 times the diameter of the light spot is extracted with this point as the center, and 3 to 5 closed curves are generated at intervals of 0.5 degrees Celsius. The multiplication grid method adjusts the grid size within the isothermal range, gradually increasing from the initial 5 microns to 20 microns, and sparsely scans the high-temperature area to reduce heat accumulation while retaining fine coverage of the defect edge. The new path coordinate sequence is thus generated and updated to the scanning control system.

[0047] In the embodiments of the present invention, focal length adjustment of the focusing lens assembly is the core link in spot size control. Taking an initial focal length of 100 mm as an example, the spot diameter is approximately 50 microns. Step adjustment allows for precise variation of ±10 microns. The 1.2x factor, based on experimental verification, accounts for the thermal diffusion range while avoiding insufficient coverage. The quarter-arc-length subdivision method ensures a balance between computational efficiency and scanning density by evenly dividing the spiral trajectory by arc length. For example, approximately 400 trajectory points are generated within a 100×100 micron area, significantly improving path planning flexibility.

[0048] Temperature monitoring and risk assessment are crucial in the embodiments of the present invention. The high-frequency sampling of the infrared sensor array ensures the real-time nature of the temperature data, and the grid processing facilitates heat flow analysis. The application of the Fourier heat conduction equation provides a theoretical basis for heat transfer, making the calculation of heat flux density more scientific. The random forest algorithm not only improves the reliability of risk prediction through multi-dimensional feature analysis, but also provides data support for dynamic adjustment. If the risk exceeds the standard, the path replanning of the multiplication grid method can effectively disperse the heat load. For example, after the grid in the high-temperature zone is increased to 20 microns, the local heat accumulation rate can be reduced by more than 30%, while ensuring effective scanning of the defective area.

[0049] It is understandable that the embodiments of the present invention do not impose too many specific restrictions on the focal length step or grid size. Technicians can adjust according to the performance of the equipment, for example, increasing the sampling frequency to 120 Hz or reducing the grid to 8×8 microns to meet higher precision coupling requirements.

[0050] S104. If the heat accumulation risk is detected to exceed a preset threshold, the high-risk area is marked and its coordinate position is obtained. The spiral search path parameters are adjusted by calculating the correction value to generate a new scanning trajectory that avoids the high-risk area to ensure the safety and efficiency of the coupling process.

[0051] In an embodiment of the present invention, the risk status is judged by comparing the heat accumulation risk monitoring data with a preset threshold value. If the risk value exceeds the standard, the temperature similarity region growing method is used to identify the high-risk area and extract the boundary coordinates. The safety distance coefficient is calculated based on the risk level and the scanning path is reconstructed. Specifically, the heat accumulation risk monitoring data is derived from the aforementioned temperature distribution analysis. When the risk value exceeds the threshold, for example, 0.8, the high-risk area marking is triggered. The temperature similarity region growing method uses the temperature peak point as the seed, expands from this point to the surrounding areas, and marks the connected areas with a temperature difference of less than 5 degrees Celsius. The growth termination condition is that the temperature difference exceeds 5 degrees Celsius or reaches the region boundary. Subsequently, the contour is extracted by the boundary tracking method to generate a boundary point coordinate sequence of the high-risk area, which usually contains 20 to 30 key points to form a closed contour map.

[0052] S1041. Calculate the centroid coordinates from the high-risk area contour map and determine the safety distance coefficient. Calculate the weighted distance between the path point and the high-risk area using Gaussian distance, and then mark the sampling points that need to be adjusted to construct the avoidance constraint. The centroid coordinates are calculated using the geometric mean of the boundary point coordinates and serve as the center point of the high-risk area. The safety distance coefficient is determined by the ratio of the measured value of the heat accumulation risk to the threshold. For example, the ratio of the measured value of 1.2 to the threshold of 0.8 is 1.5. It is adjusted to the range of 0.5 to 2 through linear mapping and serves as the adjustment parameter for the spiral path radius increment. The Gaussian distance calculation is centered on the centroid and uses a Gaussian function to measure the distance weight between the path sampling point and the center. The weight decays as the distance increases. When the distance is less than twice the spot diameter, for example, 100 microns, these sampling points are marked as areas that need to be avoided. The number of marked points varies depending on the size of the area and usually accounts for 10% to 20% of the path points.

[0053] S1042. Construct path avoidance constraints based on the marked sampling points and fit the new path curve, and generate a smooth scanning trajectory coordinate sequence through interpolation to optimize the scanning process. The path avoidance constraint requires that the weighted distance of the marked points is greater than the minimum value determined by the safety distance coefficient, such as 150 microns. The constraint condition is implemented by fitting the path curve through the least squares method. The least squares method uses the sampling point coordinates as input, and the optimization goal is to minimize the total error of the path deviation from the original trajectory while satisfying the curvature continuity. The fitted curve equation is in the form of a quadratic or cubic polynomial. Calculate the node tangent vector along the curve, set the sampling interval to one tenth of the spot diameter, such as 5 microns, and after extracting the new path sampling point sequence, use cubic spline interpolation to smooth it to ensure the continuity and processing accuracy of the trajectory curve. The number of points in the new sequence usually increases by about 15% compared to the original path.

[0054] The introduction of a temperature-similarity region growing method improves the accuracy of high-risk area identification. Taking a 100×100 micron region as an example, if the peak temperature is 160 degrees Celsius, the growth process can mark a connected high-temperature region of approximately 30×30 microns. The boundary tracking method extracts contour points through pixel-by-pixel scanning to ensure the accuracy of region demarcation. The dynamic adjustment of the safety distance coefficient fully considers the severity of the heat accumulation risk. For example, when a ratio of 1.5 is mapped to a coefficient of 1.2, the spiral radius increment increases by 20%, effectively increasing the distance between the path and the high-temperature zone. Gaussian distance calculation quantifies spatial relationships through an exponential decay function. Compared with simple Euclidean distance, it can better reflect the gradual characteristics of thermal effects and provide a scientific basis for path reconstruction.

[0055] The path fitting and interpolation process is crucial in the embodiments of the present invention. The least squares method balances the path smoothness and avoidance requirements during fitting. For example, in the case of 20 marking points, the sum of squared errors is controlled within 5 microns, and curvature continuity is achieved by limiting the mutation of the second-order derivative. Cubic spline interpolation further optimizes the trajectory details, allowing the new path to avoid high-temperature areas while maintaining effective coverage of defective areas. For example, it can still scan with an accuracy of 2 microns at the edge of the high-temperature area. This method not only reduces the risk of thermal damage, but also improves the adaptability of the coupling station.

[0056] It is understood that the parameter settings involved in the embodiments of the present invention, such as the 5°C temperature difference threshold or the sampling interval ratio, can be adjusted by technicians based on actual equipment performance. If higher accuracy is required, the growth temperature difference can be reduced to 3°C, or the interpolation interval can be adjusted to one-fifteenth of the spot diameter to meet the needs of different scenarios.

[0057] In a possible implementation, the comparison between the heat accumulation risk monitoring data and the preset threshold value may be based on temperature field data collected in real time.

[0058] For example, assuming the preset threshold is 0.8, when the heat accumulation risk value of a monitored area reaches 0.9, the system will trigger subsequent processing. The core of the temperature similarity region growing method is to identify the connectivity of the temperature distribution.

[0059] Specifically, we can start from the point where the risk value exceeds the standard, expand to the surrounding areas, and mark the areas where the temperature difference is less than 5 degrees Celsius.

[0060] For example, within a 100×100 micron scanning area, if the temperature at a certain point is 150 degrees Celsius, the temperature will spread to adjacent points, finding connected blocks with temperatures between 145 and 155 degrees Celsius, ultimately forming an irregular high-risk area. The advantage of this method is that it can quickly locate the heat concentration area, providing a basis for subsequent boundary extraction.

[0061] It should be noted that when the boundary tracking method extracts the contour of the high-risk area, a coordinate sequence can be generated by traversing the peripheral points of the connected area in a clockwise direction.

[0062] For example, in the aforementioned area, 20 boundary points may be obtained, forming a closed contour. When extracting the center of mass coordinates from the contour, one can simply take the geometric mean position of the boundary points, assuming the center of mass is located at (50, 50) microns. This center of mass calculation method is intuitive and efficient, making it suitable as a central reference for path adjustment. The safety distance factor is calculated based on the ratio of the risk value to the threshold, for example, 0.9 / 0.8 ​​= 1.125, which is then mapped to a range of 0.5 to 2, resulting in an adjustment parameter of 1.125. This mapping design takes into account the degree of risk while ensuring the controllability of the path increment.

[0063] Specifically, the reconstruction of the spiral search path can scale the original path using the adjustment parameter.

[0064] Preferably, if the original path increment is 5 microns, multiplying it by 1.125 will result in 5.625 microns. The Gaussian distance calculation provides a weighted basis for each path point to the centroid. For example, if a point is 10 microns from the centroid and the spot diameter is 50 microns, its weighted distance is less than twice the spot diameter and is marked as a point to avoid. This marking method effectively identifies path segments near high-risk areas.

[0065] In one embodiment, the construction of the path avoidance constraint may adjust the direction through a sequence of marker points.

[0066] For example, after marking 10 points near a high-risk area, the least squares method is used to fit a new curve, avoiding these points by approximately 20 microns. Curvature continuity adjustment ensures a smooth path and avoids sudden changes during scanning.

[0067] It is understandable that after the new sampling point sequence is extracted at 5-micron intervals, cubic spline interpolation can generate a smooth trajectory, for example, a smooth transition from (40,40) microns to (60,60) microns. This smooth trajectory not only improves scanning accuracy but also reduces mechanical stress during device operation.

[0068] For example, in actual applications, boundary tracking may reveal that the outline of a high-risk area is approximately elliptical, with the center of mass biased toward the heat source. The safety distance factor is then dynamically adjusted, and the path avoidance range is more closely aligned with actual needs. This multi-faceted design, forming a closed loop from risk identification to path optimization, significantly improves the system's adaptability to complex heat distributions.

[0069] In one embodiment, if the high-risk area is large, the path increment can be further enlarged to 8 microns to ensure a balance between coverage efficiency and safety.

[0070] S105 , updating the current scanning strategy according to the path and redistributing the power distribution, while performing time series analysis by collecting spatial coordinate data of the peak coupling power point when the spot size is reduced.

[0071] A grid discretization method is used to process the new scanning path and collect power data to construct a distribution matrix. Laser output power is adjusted based on parameters extracted from the matrix. A spot profiler is then used to analyze the intensity distribution and detect dimensional changes. If the reduction exceeds the specified value, the peak point is identified and the coordinate prediction is optimized using a time series model. Specifically, the new scanning path is discretized using a grid with a spacing of one-tenth of the spot diameter, for example, 5 microns. Approximately 400 sampling points are generated within a 100×100 micron area. An optical power detector collects power values ​​at a frequency of 100 Hz, with a 10 millisecond interval between each point, forming a two-dimensional power distribution matrix. The power configuration parameters in the matrix include average power, peak power, and distribution uniformity, which are used to dynamically adjust the laser output power. The spot profiler measures the lateral and longitudinal intensity distribution curves at a resolution of 0.1 microns, revealing a Gaussian distribution with peak intensity typically 5 to 10 times that of the edge. A Sobel edge detection operator is used to extract boundary coordinates, and the spot size is calculated using a least-squares ellipse fit. If the size is reduced by more than 10%, for example, from 50 microns to 45 microns, points with power values ​​greater than 2 times the average value are selected from the matrix as coupling power peak points, and their three-dimensional coordinates are recorded.

[0072] S1051. A time series is constructed from the peak point coordinate data and temporal features are extracted using a long-short-term memory network. The predicted values ​​are optimized using a state-space equation and a Kalman filter to analyze spatial variation trends. Peak point coordinates are sampled at 10-millisecond intervals to form a time series, for example, 100 data points are collected within one second. The long-short-term memory network uses a three-hidden layer structure with 64 neurons per layer. The sequence is processed using a sliding window approach with a window length of 100 points, equivalent to one second. The training data covers linear, periodic, and random drift patterns, and the extracted features include position change rate and fluctuation period. The state-space equation uses coordinate position, velocity, and acceleration as state variables, and the observations are real-time coordinate data. The Kalman filter optimizes and estimates through prediction and update phases, with convergence time typically between 50 and 100 milliseconds. The output drift velocity is between 0.1 and 1 micron per second, and the acceleration is between 0.01 and 0.1 micron per square second, forming the trend feature vector.

[0073] The accuracy of grid discretization directly affects the reliability of power sampling. Taking a 50-micron spot as an example, a 5-micron spacing ensures that the sampling points cover path details while avoiding data redundancy. The 100-Hz sampling frequency balances time resolution and computational load. The construction of the power distribution matrix provides a quantitative basis for power adjustment. For example, when the matrix shows that the local peak power is 2.5 times the average value, it indicates uneven coupling and the output power needs to be reduced to protect the device. The high-resolution measurement of the spot profiler combined with the Sobel operator can accurately capture edge changes, and the least squares ellipse fitting effectively eliminates noise interference to ensure the accuracy of size calculation.

[0074] The reduction in spot size often reflects the impact of power distribution or path adjustment. In an embodiment of the present invention, if the size is reduced from 50 microns to 45 microns, the spot may be over-focused due to power concentration. By screening the peak points and recording the coordinates, the coupling abnormality area can be located. The application of long short-term memory networks enhances the timing analysis capability. Its multi-layer structure can capture the long-term dependence of coordinates. For example, the frequency of periodic drift detected is about 0.5 Hz, indicating that path adjustment may cause oscillation. The Kalman filter reduces the prediction error through recursive optimization. For example, in a noisy environment, the coordinate deviation is reduced from 2 microns to 0.5 microns, which improves the reliability of trend analysis. This method not only reveals the dynamic changes of the peak points, but also provides data support for subsequent power optimization.

[0075] It is understandable that the parameter settings in the embodiments of the present invention are flexible. For example, the grid spacing can be adjusted to 3 microns to improve accuracy, or the network window length can be increased to 150 points to analyze trends over a longer period of time. Technicians can optimize the configuration according to actual needs.

[0076] S106, analyzing the trend of spatial coordinate changes and calculating the position drift rate. When the drift rate exceeds a preset stability threshold, the positioning accuracy is optimized by increasing the number of spiral search iterations to meet the coupling requirements.

[0077] In an embodiment of the present invention, the three-dimensional displacement of the peak coupled power point is calculated based on the spatial coordinate sequence and an instantaneous velocity sequence is generated. The drift characteristics are decomposed through Fourier transform and autoregressive sliding average model. After extracting the random and periodic drift components, the comprehensive drift rate is calculated and smoothed. If the rate exceeds the standard, the search parameters are adjusted and a multi-layer iterative structure is constructed. Specifically, the spatial coordinate sequence is segmented into a fixed time window of 100 sampling points, corresponding to an observation time of 1 second at a sampling frequency of 100 Hz. The three-dimensional displacement of the peak point in each window is calculated by forward difference. For example, the horizontal displacement can reach 1 to 2 microns, and the vertical fluctuation is 0.1 to 0.5 microns. The displacement is divided by the time interval to obtain the instantaneous velocity sequence and construct the drift curve. Fourier transform is used to analyze the drift curve, and the frequency threshold is set to one-tenth of the sampling frequency, or 10 Hz. Low-frequency components reflect trends, and high-frequency components capture jitter. The autoregressive moving average model decomposes the sequence using third-order autoregressive and second-order moving average methods. Random drift with a standard deviation greater than twice the mean and periodic drift with a periodic characteristic of 2 to 5 Hz are extracted. The integrated drift rate is obtained through vector superposition, typically in the range of 0.5 to 2 microns per second, and then exponential smoothing is performed to reduce the impact of noise.

[0078] S1061. If the combined drift rate exceeds a stability threshold, such as 1 micron per second, feedback control is used to adjust spiral search parameters and construct an iterative structure to improve positioning stability. After comparing the drift rate with the threshold, if it exceeds the threshold, the least squares method is used to calculate the feedback gain coefficient. The optimization goal is to minimize rate deviation. For example, within a 50×50 micron area, the rate is reduced from 1.5 microns per second to 0.8 microns per second. The gain coefficient is typically between 0.3 and 0.7. A proportional-integral controller adjusts the search step size and angle increment based on the gain. The step size ranges from 50% to 150% of the original value, and the angle increment is adjusted between 2 and 8 degrees. A multi-layer iterative structure subdivides the search range into four sub-areas. The first layer is divided from 100×100 microns to 50×50 microns, and the second layer is further divided to 25×25 microns. New path nodes are recursively generated at each layer and density is balanced using the Thiessen polygon method to ensure that the sampling point spacing is between 5 and 15 microns.

[0079] In the embodiments of the present invention, drift analysis is key to ensuring coupling accuracy. The time window is selected based on matching the sampling frequency with the drift period. 100 points covering 1 second can capture typical oscillations, and Fourier transform separation of frequency components provides a basis for subsequent decomposition. The autoregressive moving average model accurately distinguishes between random and periodic drift by modeling short-term correlations and long-term trends. For example, irregular fluctuations in random drift may be caused by thermal noise, while periodic drift may be related to mechanical vibration. Vector superposition combines the spatial influences of both, making the smoothed rate more representative.

[0080] If the drift rate exceeds the specified value, for example, reaching 1.5 microns per second, it indicates that the peak position is unstable, which may cause the coupled power to drop by more than 20%. By adjusting the step size to 120% of the original value and increasing the angle increment by 4 degrees, the drift rate can be controlled within the threshold. The multi-layer iterative structure avoids overly dense or sparse paths through regional subdivision and density balancing. For example, within a 25×25 micron sub-region, the node spacing after adjustment using the Thiessen polygon method is stabilized at around 10 microns, ensuring coverage while reducing the computational burden. This method effectively addresses the positioning challenges caused by drift.

[0081] It is understandable that the parameter configuration in the embodiments of the present invention can be adjusted according to the actual scenario, such as increasing the window length to 150 points to analyze longer periods, or setting the frequency threshold to 8 Hz to focus on low-frequency drift. Technicians can flexibly optimize to adapt to different equipment requirements.

[0082] S107 , extracting the dynamic change data of the spot size from the allocated power distribution and scanning the defect area, and dynamically adjusting the spot size and power by constructing a heat accumulation risk distribution map until the stability reaches a preset range.

[0083] The spot characteristic parameters are fitted based on the power distribution data and the size change is calculated. A thermal diffusion rate matrix is ​​constructed through temperature field analysis. After extracting the heat accumulation risk characteristics, the laser power and focal length are dynamically adjusted. Ultimately, closed-loop control is used to optimize coupling stability. Specifically, the power distribution data is collected in the form of a 100×100 grid to collect light intensity. A two-dimensional Gaussian function is used to fit the half-width parameter, which typically reflects the spot distribution characteristics between 20 and 30 microns. Cubic spline interpolation is used to process the sampled data at 1 millisecond intervals to generate a smooth spot size change curve. The infrared thermal imager acquires the temperature field with a resolution of 0.1 degrees Celsius. The central difference method is used to calculate the spatial gradient value. For example, the temperature difference between adjacent points is divided by the distance to obtain a gradient of 5 to 10 degrees Celsius per micron. After gridding, a thermal diffusion rate matrix is ​​constructed and normalized to quantify the thermal field distribution characteristics.

[0084] S1071. Feature extraction is performed on the thermal diffusion rate matrix to assess thermal accumulation risk. Power and focal length parameters are optimized through pulse width modulation and gradient iteration. A five-layer convolutional neural network is used to process the matrix. Three convolutional layers use 3×3 convolution kernels to extract local features. Two pooling layers use max pooling to reduce the feature map size. A fully connected layer outputs a thermal accumulation risk value ranging from 0 to 1. Risk values ​​exceeding 0.8 are marked as high-temperature areas. Pulse width modulation adjusts the laser power at a frequency of 20 kHz. The duty cycle changes dynamically based on the risk value. For example, when the risk value is 0.9, the duty cycle is reduced by 10%. The objective function is based on the spot radius and the temperature gradient deviation. The focal length compensation value is calculated through gradient iteration with a step size of 5% of the deviation. For example, when the deviation is 5 microns, the compensation is approximately 0.25 microns.

[0085] S1072. Use a high-speed photodetector to monitor peak coupled power and assess stability, continuously adjusting parameters through closed-loop control. The detector samples power data at a 1-millisecond period. After exponential smoothing, the ratio of the standard deviation of the fluctuation to the mean is calculated as a stability indicator. For example, a ratio less than 0.05 within a window of 100 sampling points is considered stable. If the indicator deviates significantly from the threshold, a proportional-integral controller adjusts the spot size and power using a proportional coefficient of 0.5 and an integral time constant of 0.2 seconds, iterating until the requirements are met.

[0086] Spot size monitoring is the foundation for dynamic adjustment. Gaussian fitting accurately describes the light intensity distribution by minimizing residuals. For example, for a 50-micron spot, changes in the full width at half maximum reflect the trend of power concentration or diffusion, while interpolation curves reveal how the size changes over time. Temperature field analysis further reveals the impact of thermal effects on coupling. Normalized matrices make the thermal diffusion rates in different regions comparable. For example, the rate in a high-temperature area may be twice that of a normal area, providing a basis for risk assessment.

[0087] The application of convolutional networks in the embodiments of the present invention greatly enhances risk prediction capabilities. A 3×3 convolution kernel captures local thermal features through a sliding window, for example, identifying a 10×10 micron high-temperature block within a 100×100 micron area. The pooling layer compresses the data while retaining key information, and the output risk value intuitively reflects the possibility of thermal damage. Pulse width modulation combined with gradient iteration ensures real-time power adjustment and accurate focal length optimization. For example, when the temperature peak offset is 5 microns, focal length adjustment can reduce the deviation to within 1 micron, significantly reducing the risk of heat accumulation.

[0088] The high-speed detector's rapid sampling captures detailed power fluctuations, while exponential smoothing effectively filters out noise, reducing high-frequency jitter from 0.2 microwatts to 0.05 microwatts, for example. Stability indicators provide clear standards for closed-loop regulation. Proportional-integral control dynamically balances rapid response with long-term stability. For example, when power fluctuations exceed standards, the system rapidly adjusts the duty cycle and gradually optimizes the focal length, improving coupling efficiency by approximately 15%, meeting high-precision requirements.

[0089] It is understood that the parameters in the embodiments of the present invention can be adjusted according to actual scenarios. For example, the convolution kernel can be changed to 5×5 to enhance feature extraction, or the sampling period can be shortened to 0.5 milliseconds to improve real-time performance. Technicians can flexibly configure the parameters to adapt to different working conditions. Subsequent steps will further improve the control strategy based on these results, and the specific implementation will be detailed in other embodiments.

[0090] like Figure 2 The present invention provides a control system for a dual-station transceiver integrated coupling device, which mainly includes:

[0091] The defect coordinate and pinhole characteristic data acquisition module is used to obtain the defect coordinates and pinhole characteristic data of the target area, analyze the abnormal transmittance area, and generate the initial path of the spiral search algorithm based on this data. The search step size and angle are adjusted to obtain the preliminary scanning boundary of the area to be scanned;

[0092] The module for analyzing abnormal transmittance areas and generating the initial path is used to analyze the proximity of the scanning path to the defect. If the proximity is within a preset range, the module determines the power reduction condition based on the defect coordinates and adjusts the power. The module also processes the light spot to obtain its size and calculates the broadening coefficient.

[0093] The scanning path and defect proximity analysis module is used to adjust the actual spot size using the adjusted power and stretch factor. It performs an initial small-step spiral scan in the preset range of the defect coordinates according to the adjusted search step size and angle. During this process, it monitors in real time whether the heat accumulation risk exceeds the preset threshold.

[0094] The power adjustment and spot size processing module is used to mark the corresponding area as a high-risk area if the heat accumulation risk exceeds the threshold, obtain the coordinate position of the high-risk area, and calculate the correction value based on the value of the heat accumulation risk exceeding the threshold, adjust the spiral search radius increment, and generate a new scanning path to avoid the high-risk area;

[0095] The heat accumulation risk monitoring and high-risk area marking module is used to update the current scanning path according to the new scanning path and allocate the power distribution. If the actual spot size decreases after the current scanning path is updated, the spatial coordinate data of the peak coupled power point is collected and time series analysis is performed to obtain the spatial coordinate change trend;

[0096] A new scanning path generation and power distribution allocation module is used to analyze the trend of spatial coordinate changes and calculate the drift rate of the peak coupled power point position. If the drift rate exceeds the preset stability threshold, indicating that the current positioning stability is insufficient to meet the preset accuracy requirements, the number of spiral search iterations is increased;

[0097] The spot size dynamic adjustment and peak coupled power stability recovery module is used to extract the dynamic change data of the spot size from the allocated power distribution, scan the defect area, obtain the real-time distribution map of the heat accumulation risk, and dynamically adjust the spot size and power until the peak coupled power stability is restored to the preset range.

[0098] With the above embodiments of the present invention as inspiration, and through the above description, relevant personnel can make various changes and modifications without departing from the technical scope of this invention. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A control method for a dual-station transceiver integrated coupling device, characterized in that: The method comprises: Obtain the defect coordinates and pinhole characteristic data of the target area, analyze and obtain the transmittance abnormality area, and generate the initial path of the spiral search algorithm based on this data. Adjust the search step size and angle to obtain the preliminary scanning boundary of the area to be scanned. Analyze the proximity of the scanning path to the defect. If it is within the preset range, determine the power reduction condition based on the defect coordinates and adjust the power. Process the spot to obtain its size and calculate the broadening coefficient. The actual spot size is adjusted using the adjusted power and stretch factor. An initial small-step spiral scan is performed within the preset range of the defect coordinates according to the adjusted search step size and angle. During this process, real-time monitoring is performed to determine whether the heat accumulation risk exceeds the preset threshold. If the heat accumulation risk exceeds the threshold, the corresponding area is marked as a high-risk area, the coordinate position of the high-risk area is obtained, and based on the value of the heat accumulation risk exceeding the threshold, the correction value is calculated, the spiral search radius increment is adjusted, and a new scanning path is generated to avoid the high-risk area; Update the current scanning path according to the new scanning path and allocate the power distribution. If the actual spot size decreases after the current scanning path is updated, the spatial coordinate data of the peak coupled power point is collected and time series analysis is performed to obtain the spatial coordinate change trend; Analyze the trend of spatial coordinate changes and calculate the drift rate of the peak coupled power point position. If the drift rate exceeds the preset stability threshold, indicating that the current positioning stability is insufficient to meet the preset accuracy requirements, increase the number of spiral search iterations. The dynamic change data of the spot size is extracted from the allocated power distribution, the defect area is scanned, and a real-time distribution map of the heat accumulation risk is obtained. The spot size and power are dynamically adjusted until the peak coupled power stability returns to the preset range.

2. The method according to claim 1, characterized in that The defect coordinates and pinhole characteristic data of the target area are obtained, and the abnormal transmittance area is obtained by analysis. Based on the obtained data, the initial path of the spiral search algorithm is generated, and the search step size and angle are adjusted to obtain the preliminary scanning boundary of the area to be scanned, including: The target area is divided into sampling blocks using a grid method, and the transmittance value in the block is obtained by a micro-displacement scanning device. The abnormal points are marked based on the comparison result of the transmittance value with the preset transmittance threshold; According to the distribution of abnormal points, a photoelectric sensor array is used to obtain a gray value sequence in the sampling block, and the coordinates of the starting point of the search path are calculated based on the gray value sequence; Generate a spiral trajectory sampling point sequence within a sampling boundary using the starting point coordinates and a preset search angle, and group the sampling point sequence by spatial distance using a hierarchical clustering method; The cubic spline interpolation method is used to calculate the boundary contour curve equation for the grouped sampling point sequence, a closed polygon is constructed according to the boundary contour curve equation, and a convex hull algorithm is used to obtain the minimum circumscribed polygon as the boundary coordinates of the scanning area.

3. The method according to claim 1, characterized in that The analysis of the proximity between the scanning path and the defect, if within a preset range, determines the power reduction condition according to the defect coordinates and adjusts the power, processes the spot to obtain its size, and calculates the broadening coefficient, including: The four-neighborhood grid distance calculation method is used to obtain the shortest distance between the scanning path point sequence and the defect position, and the path approach point sequence is obtained; According to the path approach point sequence, an array photoelectric detector is used to obtain a reflected light intensity curve of the defect area, and a depth coefficient is calculated from the light intensity curve. If the depth coefficient is greater than a preset reference value, a Gaussian attenuation function curve is constructed to obtain a power attenuation ratio value; Adjusting the laser output power according to the power attenuation ratio value, acquiring the spot cross-section energy density distribution data through a photomultiplier tube to generate a grayscale image, performing maximum entropy segmentation on the grayscale image to extract edge contour point coordinates to obtain a spot radius sequence; The maximum value of the spot radius sequence is extracted to determine the spot diameter value, the overlap between the actual processing area and the theoretical area is calculated according to the defect area boundary curve and the spot diameter value, and the broadening coefficient value is obtained by fitting the overlap curve through cubic spline interpolation.

4. The method according to claim 1, wherein The actual spot size is adjusted by applying the adjusted power and stretch coefficient, and an initial small-step spiral scan is performed in the preset range of the defect coordinates according to the adjusted search step and angle. During this process, real-time monitoring is performed to determine whether the heat accumulation risk exceeds a preset threshold, including: The actual spot diameter is obtained by adjusting the focusing lens group within a preset focal length range, and the radius of the scanning coverage area is obtained by multiplying the actual spot diameter by a preset coefficient; A spiral scanning trajectory point sequence is generated using an arc length subdivision method for the scanning coverage area, and a temperature distribution map is obtained at a preset sampling frequency through an infrared temperature sensor array; Discretizing the temperature distribution map into a grid of preset size, calculating a temperature gradient value from the grid, and obtaining a heat flux density value based on the temperature gradient value using the Fourier heat conduction equation; If the heat flux value exceeds a preset threshold, the coordinates of the temperature peak point are extracted from the temperature distribution map, and a scanning path coordinate sequence is regenerated for the temperature peak point using a multiplication grid method.

5. The method according to claim 1, wherein If the heat accumulation risk exceeds the threshold, the corresponding area is marked as a high-risk area, the coordinate position of the high-risk area is obtained, and a correction value is calculated based on the value of the heat accumulation risk exceeding the threshold, and the spiral search radius increment is adjusted to generate a new scanning path to avoid the high-risk area, including: The heat accumulation risk monitoring data is compared with the preset threshold value to make a judgment. If the heat accumulation risk value exceeds the preset threshold, the coordinate sequence of the boundary points of the high-risk area is obtained by the temperature similarity region growing method; Extracting the centroid coordinates of the high-risk area boundary point coordinate sequence as the area center point, and obtaining the safety distance coefficient based on the ratio of the actual heat accumulation risk value to the preset threshold value; The original spiral path is reconstructed using the safety distance coefficient, and the weighted distance value from the path sampling point to the center point of the area is obtained by Gaussian distance calculation; A path avoidance constraint is constructed for the weighted distance value, a path curve equation is obtained by least square fitting, and a scanning trajectory coordinate sequence is determined according to the path curve equation.

6. The method according to claim 1, characterized in that The current scanning path is updated according to the new scanning path, and the power distribution is allocated. If the actual spot size decreases after the current scanning path is updated, the spatial coordinate data of the peak coupled power point is collected and time series analysis is performed to obtain the spatial coordinate change trend, including: Using an optical power detector to obtain a sequence of power values ​​of path points, constructing a two-dimensional power distribution matrix based on the power value sequence, and extracting power configuration parameters from the power distribution matrix; Adjusting the laser output power according to the power configuration parameters, obtaining a light intensity distribution curve through a spot profiler, and extracting the spot boundary coordinate points using a Sobel edge detection operator based on the light intensity distribution curve; If the reduction in the spot size exceeds a preset threshold, a point with a power value greater than twice the average value is selected from the power distribution matrix as a coupling power peak point, and three-dimensional coordinate data is recorded for the peak point; The time series features of the coordinate data are extracted through a long short-term memory network, a state space equation is established according to the time series features, and a Kalman filter is used to optimize the coordinate prediction value.

7. The method according to claim 1, characterized in that The analysis of the spatial coordinate change trend and calculation of the peak coupled power point position drift rate are performed. If the drift rate exceeds a preset stability threshold, indicating that the current positioning stability is insufficient to meet the preset accuracy requirement, the number of spiral search iterations is increased, including: Acquire a three-dimensional displacement of a peak coupling power point using a fixed-length time window according to a spatial coordinate sequence, obtain an instantaneous velocity sequence by dividing the three-dimensional displacement by a corresponding time interval, and construct a drift curve from the instantaneous velocity sequence; Performing Fourier transform and autoregressive moving average model decomposition on the drift curve, extracting from the decomposition results a component with a standard deviation greater than a standard deviation threshold as a random drift, and extracting a component with periodic characteristics as a periodic drift; Performing vector superposition on the random drift and the periodic drift to obtain a comprehensive drift rate, and performing exponential smoothing on the comprehensive drift rate; If the integrated drift rate exceeds the stability threshold, the least squares method is used to calculate the feedback gain coefficient, the spiral search step size and the angle increment are compensated by the feedback gain coefficient, and a multi-layer iterative structure is constructed from the compensated parameters to increase the number of spiral search iterations.

8. The method according to claim 1, characterized in that The method extracts dynamic change data of the spot size from the allocated power distribution, scans the defect area, obtains a real-time distribution map of the heat accumulation risk, and dynamically adjusts the spot size and power until the peak coupled power stability returns to a preset range, including: The light spot half-height width parameter is obtained by fitting a two-dimensional Gaussian function according to the power distribution, and the light spot size change is obtained by interpolating the light spot half-height width parameter through a cubic spline curve; receiving the spot size variation, calculating a temperature field spatial gradient value using a central difference method, and constructing a thermal diffusion rate matrix by gridding the temperature field spatial gradient value; A five-layer convolutional network structure is used to extract features from the heat diffusion rate matrix, and the fully connected layer of the feature extraction outputs the heat accumulation risk value; The laser output power is controlled by pulse width modulation according to the heat accumulation risk value, and the laser output power is calculated through gradient iteration to obtain a focal length compensation value.

9. A control system for a dual-station transceiver integrated coupling device, characterized in that: The system comprises: The defect coordinate and pinhole characteristic data acquisition module is used to obtain the defect coordinates and pinhole characteristic data of the target area, analyze the abnormal transmittance area, and generate the initial path of the spiral search algorithm based on this data. The search step size and angle are adjusted to obtain the preliminary scanning boundary of the area to be scanned; The module for analyzing abnormal transmittance areas and generating the initial path is used to analyze the proximity of the scanning path to the defect. If the proximity is within a preset range, the module determines the power reduction condition based on the defect coordinates and adjusts the power. The module also processes the light spot to obtain its size and calculates the broadening coefficient. The scanning path and defect proximity analysis module is used to adjust the actual spot size using the adjusted power and stretch factor. It performs an initial small-step spiral scan in the preset range of the defect coordinates according to the adjusted search step size and angle. During this process, it monitors in real time whether the heat accumulation risk exceeds the preset threshold. The power adjustment and spot size processing module is used to mark the corresponding area as a high-risk area if the heat accumulation risk exceeds the threshold, obtain the coordinate position of the high-risk area, and calculate the correction value based on the value of the heat accumulation risk exceeding the threshold, adjust the spiral search radius increment, and generate a new scanning path to avoid the high-risk area; The heat accumulation risk monitoring and high-risk area marking module is used to update the current scanning path according to the new scanning path and allocate the power distribution. If the actual spot size decreases after the current scanning path is updated, the spatial coordinate data of the peak coupled power point is collected and time series analysis is performed to obtain the spatial coordinate change trend; A new scanning path generation and power distribution allocation module is used to analyze the trend of spatial coordinate changes and calculate the drift rate of the peak coupled power point position. If the drift rate exceeds the preset stability threshold, indicating that the current positioning stability is insufficient to meet the preset accuracy requirements, the number of spiral search iterations is increased; The spot size dynamic adjustment and peak coupled power stability recovery module is used to extract the dynamic change data of the spot size from the allocated power distribution, scan the defect area, obtain the real-time distribution map of the heat accumulation risk, and dynamically adjust the spot size and power until the peak coupled power stability is restored to the preset range.

Images