Intelligent precision alignment method and system for optoelectronic coupling devices

Abstract

The application provides an intelligent precise alignment method and system for photoelectric coupling equipment, and relates to the technical field of photoelectric alignment. The application comprises the following steps: obtaining light intensity distribution information of a light source end and a receiving end, constructing a coupling state vector through sparse representation, constructing a mapping matrix of a multi-degree-of-freedom pose adjustment amount, calculating a difference vector inverse mapping to obtain an adjustment amount, obtaining a response mode through orthogonal decomposition of the mapping matrix, performing time-frequency evolution analysis and phase compensation to obtain an optimized adjustment amount, and adjusting a driving mechanism and repeating until an alignment condition is met. The application realizes high-precision, fast automatic alignment, and improves alignment efficiency and reliability.

Description

Technical Field

[0001] This invention relates to the field of photoelectric alignment technology, and in particular to an intelligent precision alignment method and system for photoelectric coupling devices. Background Technology

[0002] Precise alignment of optocouplers is a critical process in fields such as optical communication and laser processing. Current conventional methods typically rely on incremental adjustments based on optical power or spot centroid feedback: operators or control systems monitor the peak light intensity or total luminous flux at the receiver, sequentially performing trial-and-error scans of each degree of freedom (such as lateral displacement, axial distance, and deflection angle) between the light source and receiver until the optical power reaches a preset threshold. Some automated solutions use image sensors to acquire the spot distribution and calculate centroid coordinates or Gaussian fitting parameters to drive the actuators for closed-loop control. While these methods can meet basic alignment requirements under ideal conditions, they all rely on local optimization of a single scalar or low-dimensional feature, failing to fully utilize the complete spatial structural information of the light intensity distribution.

[0003] Conventional methods have several drawbacks. They rely solely on peak light intensity or total power as feedback signals, neglecting the subtle characteristics of the light intensity distribution (such as energy dispersion and mode distortion). When non-uniform illumination, surface defects on components, or environmental disturbances exist, a single metric can easily lead the alignment process into local extrema, failing to guarantee global convergence to the optimal coupling state. This is especially true for multimode fibers or non-ideal optical systems, where complex spot morphologies and potential deviations between peak positions and the actual optimal alignment point result in low adjustment efficiency or even misjudgments.

[0004] The strong coupling effect between multiple degrees of freedom is another prominent bottleneck of existing methods. During the adjustment process of one degree of freedom, the change in one degree of freedom will significantly affect the light intensity response characteristics of other degrees of freedom, causing the traditional sequential adjustment strategy to oscillate repeatedly. Operators or simple control algorithms often find it difficult to establish real-time correlations between various adjustment quantities, resulting in a lengthy convergence path and accuracy limited by empirical thresholds. This coupling characteristic is particularly significant under the alignment requirements of micrometer to submicrometer scales, requiring a large number of iterations to approach the target state, which severely restricts production efficiency and alignment consistency. Summary of the Invention

[0005] The present invention provides a method and system for intelligent precision alignment of optocouplers, which can solve the problems in the prior art.

[0006] A first aspect of the present invention provides a method for intelligent precision alignment of an optocoupler, comprising: To obtain light intensity distribution information between the light source end and the receiving end in an optocoupler; The light intensity distribution information is sparsely represented, an overcomplete dictionary containing multiple atomic bases is constructed, and the sparse representation coefficients of the light intensity distribution information on the overcomplete dictionary are calculated by iterative matching and tracing. The sparse representation coefficients are arranged in the order of the dictionary atoms to form a coupled state vector. Construct a mapping matrix between the coupled state vector and the pose adjustment amount of multiple degrees of freedom. Each element of the mapping matrix is ​​the change in the corresponding component of the coupled state vector caused by the unit adjustment amount of the corresponding degree of freedom. Calculate the difference vector between the current coupled state vector and the target coupled state vector. Convert the difference vector into the adjustment amount of each degree of freedom through the inverse operation of the mapping matrix. The mapping matrix is ​​orthogonally decomposed to obtain the response modes and their corresponding eigenvalues. Time-frequency evolution analysis is performed on each response mode to obtain the spectral distribution characteristics. The adjustment amount of each degree of freedom is projected onto the direction of each response mode to obtain the mode adjustment amount. Phase compensation is performed according to the spectral distribution characteristics, and the inverse transformation is used to map it into the optimized adjustment amount of each degree of freedom to drive the adjustment mechanism to perform pose adjustment. Repeat the adjustment until the light intensity distribution information meets the preset alignment conditions.

[0007] In one optional embodiment, the light intensity distribution information is sparsely represented, an overcomplete dictionary containing multiple atomic bases is constructed, and the sparse representation coefficients of the light intensity distribution information on the overcomplete dictionary are calculated by iterative matching pursuit. The sparse representation coefficients are then arranged in the order of the dictionary atoms to form a coupled state vector, including: The light intensity distribution information is normalized to obtain standardized light intensity distribution information; Multiple scale parameters and multiple direction angle parameters are set. For each combination of scale parameters and direction angle parameters, the standardized light intensity distribution information is spatially filtered at the corresponding scale, and gradient features are extracted at the corresponding direction angle. The extracted gradient features are used as an atomic basis. The atomic basis corresponding to all combinations of scale parameters and direction angle parameters are summarized to form an overcomplete dictionary. The standardized light intensity distribution information is used as the initial signal to be decomposed. The matching degree between the initial signal to be decomposed and each atomic base in the overcomplete dictionary is calculated. The atomic base with the largest matching degree is selected as the current matching atomic base. The sparse representation coefficients of the current matching atomic base are calculated and stored. Based on the current matching atomic base and the corresponding sparse representation coefficients, the contribution component of the current matching atomic base is removed from the initial signal to be decomposed to obtain the residual signal. It is then determined whether the residual signal meets the preset termination condition. If not, the residual signal is used as the new initial signal to be decomposed, and the process is repeated. The sparse representation coefficients obtained in each iteration are arranged in the order of the corresponding atomic bases in the overcomplete dictionary to form a coupled state vector.

[0008] In one optional embodiment, the normalized light intensity distribution information is used as the initial signal to be decomposed. The matching degree between the initial signal to be decomposed and each atomic base in the overcomplete dictionary is calculated. The atomic base with the highest matching degree is selected as the current matching atomic base. The sparse representation coefficients of the current matching atomic base are calculated and stored. Using the standardized light intensity distribution information as the initial signal to be decomposed, spatial gradient calculations are performed on the initial signal to be decomposed and each atomic basis in the overcomplete dictionary to obtain the spatial gradient distribution of the initial signal to be decomposed and the atomic gradient distribution corresponding to each atomic basis. The gradient angle deviation between the spatial gradient distribution and the atomic gradient distribution is calculated at each sampling point in the spatial domain. Sampling points with gradient angle deviation exceeding a preset angle threshold are removed. The product of the gradient magnitudes of the remaining sampling points is spatially integrated to obtain the matching degree of each atomic basis. Select the atomic base with the highest matching degree as the current matching atomic base; The size of the spatial search window is determined by the scale parameter of the current matching atomic base. The initial signal to be decomposed is scanned at multiple positions within the spatial search window. The local correlation peak between the initial signal to be decomposed and the current matching atomic base is calculated at each translation position. The spatial offset when the local correlation peak reaches its maximum is recorded. The maximum value of the local correlation peak and the spatial offset are combined to form the sparse representation coefficient of the current matching atomic base. The sparse representation coefficients of the current matching atomic base are associated with the index position of the current matching atomic base in the overcomplete dictionary and stored.

[0009] In one optional embodiment, a mapping matrix is ​​constructed between the coupled-state vector and the pose adjustment amount of the multi-degree-of-freedom position. Each element of the mapping matrix represents the change in the corresponding component of the coupled-state vector caused by a unit adjustment amount for the corresponding degree of freedom, including: For each degree of freedom in the pose adjustment, the control adjustment mechanism generates a unit adjustment in that degree of freedom and collects the perturbation light intensity distribution information of that degree of freedom. The perturbation intensity distribution information is sparsely represented and calculated to obtain the perturbation coupled state vector of this degree of freedom; Calculate the component differences between the perturbation coupled-state vector and the reference coupled-state vector, and arrange the component differences in order of the dimensions of the coupled-state vector to form the sensitivity vector of that degree of freedom; Calculate the cosine of the angle between any two degree of freedom sensitivity vectors. When the absolute value of the cosine exceeds a preset correlation threshold, perform Schmitt orthogonalization on the sensitivity vectors to obtain orthogonal sensitivity vectors. Assemble the orthogonal sensitivity vectors of all degrees of freedom into column vectors in order of degrees of freedom to form a mapping matrix.

[0010] In one optional embodiment, calculating the difference vector between the current coupled-state vector and the target coupled-state vector, and converting the difference vector into adjustment values ​​for each degree of freedom through the inverse operation of the mapping matrix includes: Obtain the current light intensity distribution information, perform sparse representation calculation on the current light intensity distribution information, and obtain the current coupled state vector; Calculate the component differences between the current coupled state vector and the target coupled state vector in each dimension, and arrange the component differences in each dimension in dimensional order to form a difference vector; Singular value decomposition is performed on the mapping matrix to obtain the sequence of singular values ​​of the mapping matrix. The ratio of the maximum singular value to the minimum singular value in the sequence of singular values ​​is calculated as the condition number. When the condition number exceeds the preset ill-conditioned threshold, a regularization coefficient is superimposed on the diagonal elements of the mapping matrix to obtain a regularized mapping matrix. Perform matrix inversion on the regularized mapping matrix to obtain the inverse mapping matrix; Perform matrix multiplication on the difference vector and the inverse mapping matrix to obtain the initial adjustment amount. Iterate through the adjustment amplitude corresponding to each degree of freedom in the initial adjustment amount, and compare each adjustment amplitude with the maximum allowable adjustment amplitude of the corresponding degree of freedom. When the adjustment amplitude exceeds the maximum allowable adjustment amplitude, replace the adjustment amplitude with the maximum allowable adjustment amplitude to obtain the adjustment amount of each degree of freedom.

[0011] In one optional embodiment, the mapping matrix is ​​orthogonally decomposed to obtain the response modes and corresponding eigenvalues, and time-frequency evolution analysis is performed on each response mode to obtain the spectral distribution characteristics, including: The mapping matrix is ​​orthogonally decomposed to obtain each response mode and its corresponding eigenvalue. The magnitude of each eigenvalue is calculated as the contribution of the corresponding response mode to the coupling efficiency. Multiple sets of mapping matrices are collected within a preset time window at a preset sampling interval. Orthogonal decomposition is performed on each set of mapping matrices to obtain the response modes and eigenvalues ​​at the corresponding time. The similarity of the eigenvectors of each response mode at the current time and the response modes at the previous time is calculated. The response mode pairs with the highest similarity are determined as the correspondence of the same response mode at different times. The feature values ​​of the same response mode at different times are extracted according to the correspondence, and the feature values ​​at each time are arranged in chronological order to form the dynamic evolution sequence of the response mode; A fast Fourier transform is performed on the dynamic evolution sequence to obtain the amplitude spectrum and phase spectrum in the frequency domain. The frequency components with amplitudes greater than the average amplitude in the amplitude spectrum and their corresponding phase information are extracted to form the spectral distribution characteristics of the response mode.

[0012] In one optional embodiment, the modal adjustment amount is obtained by projecting the adjustment amount of each degree of freedom onto the direction of each response mode, performing phase compensation based on the spectral distribution characteristics, and mapping it to the optimized adjustment amount of each degree of freedom through inverse transformation, thereby driving the adjustment mechanism to perform pose adjustment, including: Convert the adjustment values ​​of each degree of freedom into vector form, and calculate the projection components of the vectors on each response mode as the initial mode adjustment values; The frequency component with the largest amplitude is extracted from the spectral distribution characteristics of each response mode as the dominant frequency component. The phase value corresponding to the dominant frequency component is extracted, and the response mode with the smallest phase is selected as the phase reference mode. The phase lag of other response modes relative to the phase reference mode is calculated, and the synchronization adjustment timing of each response mode is determined by sorting the phase lag from smallest to largest. The phase prediction value for the next moment is calculated based on the phase lag and dominant frequency of each response mode. The phase prediction value is then compared with the phase of the phase reference mode to obtain the dynamic phase deviation. The initial mode adjustment amount is then adjusted by time delay based on the dynamic phase deviation to obtain the mode adjustment amount for phase alignment. The optimized mode adjustment is obtained by multiplying the phase-aligned mode adjustment amount with the contribution of the corresponding response mode. The optimized mode adjustment amount is then inversely transformed with each response mode to obtain the optimized adjustment amount for each degree of freedom, which drives the adjustment mechanism to perform pose adjustment.

[0013] A second aspect of the present invention provides an intelligent precision alignment system for an optocoupler, comprising: The light intensity acquisition unit is used to acquire light intensity distribution information between the light source end and the receiving end in the optocoupler. The sparse representation unit is used to sparsely represent light intensity distribution information. It constructs an overcomplete dictionary containing multiple atomic bases, calculates the sparse representation coefficients of light intensity distribution information on the overcomplete dictionary through iterative matching and tracing, and arranges the sparse representation coefficients according to the atomic order of the dictionary to form a coupled state vector. The mapping matrix unit is used to construct a mapping matrix between the coupled state vector and the pose adjustment amount of the multi-degree-of-freedom. Each element of the mapping matrix is ​​the change in the corresponding component of the coupled state vector caused by the unit adjustment amount of the corresponding degree of freedom. The difference vector between the current coupled state vector and the target coupled state vector is calculated. The difference vector is converted into the adjustment amount of each degree of freedom through the inverse operation of the mapping matrix. The modal analysis unit is used to perform orthogonal decomposition of the mapping matrix to obtain the response modes and corresponding eigenvalues, perform time-frequency evolution analysis on each response mode to obtain the spectral distribution characteristics, project the adjustment amount of each degree of freedom to the direction of each response mode to obtain the modal adjustment amount, perform phase compensation according to the spectral distribution characteristics, and map it to the optimized adjustment amount of each degree of freedom through inverse transformation to drive the adjustment mechanism to perform pose adjustment. The iterative adjustment unit is used to repeatedly perform adjustments until the light intensity distribution information meets the preset alignment conditions.

[0014] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0016] In this embodiment of the invention, a sparse representation method is used to extract features from the light intensity distribution information. The coupled-state vector constructed by the overcomplete dictionary and matching pursuit algorithm can effectively filter out environmental stray light and system noise while retaining core alignment information. The sparse coding method based on atoms significantly compresses the data dimension, transforming the light intensity distribution into a structured vector expression. This provides a stable and robust feature space for establishing mapping relationships, overcoming the model failure problem caused by light intensity signal interference in traditional methods. By constructing a mapping matrix between the coupled-state vector and the multi-degree-of-freedom pose adjustment quantities, and using inverse operations to convert the difference vector into adjustment quantities for each degree of freedom, the system achieves... The system accurately maps light intensity distribution deviation to pose correction; the response modes and eigenvalues ​​obtained by orthogonal decomposition reflect the independent contribution and coupling characteristics of each adjustment degree of freedom; the spectral distribution based on time-frequency evolution analysis provides a physical basis for phase compensation, effectively avoiding modal crosstalk during the adjustment process, making the adjustment amount calculation more targeted and accurate, significantly improving the dynamic response quality of pose adjustment, and ensuring stable convergence under multi-degree-of-freedom cooperative operation; by projecting the adjustment amount to the direction of each response mode and implementing phase compensation, and then inversely transforming it into the optimized adjustment amount, adaptive matching of the inherent dynamic characteristics of the system is achieved, effectively suppressing over-adjustment and oscillation phenomena during the adjustment process. Attached Figure Description

[0017] Figure 1 A flowchart illustrating the intelligent precision alignment method for optocouplers; Figure 2 The flowchart shows the logic for mapping matrix analysis and spectral characteristic extraction. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0020] Figure 1 This is a flowchart illustrating the intelligent precision alignment method for optocouplers according to an embodiment of the present invention. Figure 1 As shown, the method includes: To obtain light intensity distribution information between the light source end and the receiving end in an optocoupler; The light intensity distribution information is sparsely represented, an overcomplete dictionary containing multiple atomic bases is constructed, and the sparse representation coefficients of the light intensity distribution information on the overcomplete dictionary are calculated by iterative matching and tracing. The sparse representation coefficients are arranged in the order of the dictionary atoms to form a coupled state vector. Construct a mapping matrix between the coupled state vector and the pose adjustment amount of multiple degrees of freedom. Each element of the mapping matrix is ​​the change in the corresponding component of the coupled state vector caused by the unit adjustment amount of the corresponding degree of freedom. Calculate the difference vector between the current coupled state vector and the target coupled state vector. Convert the difference vector into the adjustment amount of each degree of freedom through the inverse operation of the mapping matrix. The mapping matrix is ​​orthogonally decomposed to obtain the response modes and their corresponding eigenvalues. Time-frequency evolution analysis is performed on each response mode to obtain the spectral distribution characteristics. The adjustment amount of each degree of freedom is projected onto the direction of each response mode to obtain the mode adjustment amount. Phase compensation is performed according to the spectral distribution characteristics, and the inverse transformation is used to map it into the optimized adjustment amount of each degree of freedom to drive the adjustment mechanism to perform pose adjustment. Repeat the adjustment until the light intensity distribution information meets the preset alignment conditions.

[0021] In one optional embodiment, the light intensity distribution information is sparsely represented, an overcomplete dictionary containing multiple atomic bases is constructed, and the sparse representation coefficients of the light intensity distribution information on the overcomplete dictionary are calculated by iterative matching pursuit. The sparse representation coefficients are then arranged in the order of the dictionary atoms to form a coupled state vector, including: The light intensity distribution information is normalized to obtain standardized light intensity distribution information; Multiple scale parameters and multiple direction angle parameters are set. For each combination of scale parameters and direction angle parameters, the standardized light intensity distribution information is spatially filtered at the corresponding scale, and gradient features are extracted at the corresponding direction angle. The extracted gradient features are used as an atomic basis. The atomic basis corresponding to all combinations of scale parameters and direction angle parameters are summarized to form an overcomplete dictionary. The standardized light intensity distribution information is used as the initial signal to be decomposed. The matching degree between the initial signal to be decomposed and each atomic base in the overcomplete dictionary is calculated. The atomic base with the largest matching degree is selected as the current matching atomic base. The sparse representation coefficients of the current matching atomic base are calculated and stored. Based on the current matching atomic base and the corresponding sparse representation coefficients, the contribution component of the current matching atomic base is removed from the initial signal to be decomposed to obtain the residual signal. It is then determined whether the residual signal meets the preset termination condition. If not, the residual signal is used as the new initial signal to be decomposed, and the process is repeated. The sparse representation coefficients obtained in each iteration are arranged in the order of the corresponding atomic bases in the overcomplete dictionary to form a coupled state vector.

[0022] In one specific implementation, after acquiring the light intensity distribution information between the light source and the receiver in the optocoupler, this information needs to be sparsely represented to construct a coupling state vector that can accurately describe the coupling state. Light intensity distribution information typically exists in a two-dimensional spatial distribution. Due to differences in the absolute values ​​of light intensity under different devices and measurement conditions, directly processing the raw light intensity data can lead to deviations in subsequent calculations due to dimensionality. Therefore, normalizing the light intensity distribution information is a necessary prerequisite. Specifically, the acquired light intensity distribution data is divided by its maximum value, mapping the light intensity values ​​of all pixels to the range of 0 to 1, thus obtaining standardized light intensity distribution information. Normalization eliminates the influence of absolute light intensity magnitude, allowing the subsequent gradient-based atomic basis extraction process to focus only on the relative distribution of light intensity, effectively improving the robustness of the sparse representation.

[0023] Constructing an overcomplete dictionary is a core step in sparse representation. Multiple scale parameters are set. (in (Total number of scale parameters) and multiple directional angle parameters (in (This represents the total number of direction and angle parameters). Each scale parameter represents the filter kernel size or Gaussian kernel standard deviation used when spatially filtering the standardized light intensity distribution information. Smaller scale parameters correspond to the capture of detailed information, while larger scale parameters correspond to the description of the overall contour structure. For each set of scale parameters... ( ) and direction angle parameters ( The combination of these parameters is used to perform spatial Gaussian filtering on the standardized light intensity distribution information, filtering out high-frequency noise above this scale while retaining spatial structure information within this scale range; subsequently, the corresponding directional angle is used... Gradient features are extracted by calculating the directional derivative of the filtered image along the specified angle, yielding the gradient response map in that direction. This gradient response map is then expanded into a column vector, serving as an atomic basis in an overcomplete dictionary. All... The combination of scale parameters and orientation angle parameters will correspond to The atomic bases are arranged in a comprehensive dictionary. Due to the number of atomic bases in the dictionary Typically much larger than the dimensions of the standardized light intensity distribution information after expansion, this dictionary is overcomplete and can redundantly describe the light intensity distribution structure from multiple angles and scales, providing a sufficient decomposition basis for subsequent sparse representation.

[0024] After the incomplete dictionary is constructed, an iterative matching pursuit algorithm is used to sparsely decompose the standardized light intensity distribution information. The standardized light intensity distribution information is expanded into a column vector, which serves as the initial signal to be decomposed. In each iteration, the current signal to be decomposed and the overcomplete dictionary are calculated. The matching degree of each atomic base is defined as the absolute value of the inner product between the current signal to be decomposed and each atomic base. The larger the absolute value of the inner product, the stronger the correlation between the atomic base and the current signal to be decomposed. The atomic base with the highest matching degree is selected as the current matching atomic base. subscript This is the index number of the atomic base in the overcomplete dictionary. Calculate the sparse representation coefficients corresponding to the currently matching atomic base. Its value is equal to the projection length of the current signal to be decomposed onto the current matching atomic base direction, i.e. ,in This represents the inner product operation. The calculated sparse representation coefficients... and its corresponding atomic base index Store the data for later use in arranging the coupled-state vectors.

[0025] After obtaining the current matching atomic base and its sparse representation coefficients, the contribution component of that atomic base is removed from the current signal to be decomposed. Specifically, the product of the current matching atomic base and its sparse representation coefficients is subtracted from the current signal to be decomposed to obtain the residual signal. The residual signal represents the portion of the current signal to be decomposed that has not yet been interpreted by the selected atomic bases. A termination condition is determined for the residual signal: if the norm of the residual signal is below a preset threshold, or the number of iterations has reached a preset maximum number of iterations, the preset termination condition is met, and iteration stops; otherwise, the residual signal is used as the new initial signal to be decomposed, and the next round of iterations continues, repeatedly calculating new matching degrees, selecting new matching atomic bases, calculating new sparse representation coefficients, and updating the residual signal. Through multiple rounds of iteration, the atomic base combinations and corresponding coefficients that best represent the structural characteristics of the standardized light intensity distribution information are gradually extracted, achieving a sparse approximate description of the light intensity distribution information.

[0026] After all iterations are completed, the sparse representation coefficients obtained in each iteration are sorted according to the corresponding atomic basis in an overcomplete dictionary. The indexes in the dictionary are arranged in order. For the first index in an overcomplete dictionary... Atom base ( If an atomic base is selected during the iteration, the component value at its corresponding position is the sparse representation coefficient calculated in that iteration; if the atomic base is not selected, the component value at its corresponding position is zero. The coefficients corresponding to all atomic bases are arranged sequentially according to the above rules, forming a dimensional array. The coupled-state vector encodes the structural features of light intensity distribution information in a multi-scale, multi-directional feature space in a sparse form. The positions and values ​​of the non-zero components reflect the main characteristic directions and intensities of the light intensity distribution, respectively. Since the atomic basis of the overcomplete dictionary covers multiple scales and directions, the coupled-state vector can comprehensively characterize the spatial distribution characteristics of the coupling state between the light source and the receiver, providing a structured and quantifiable input representation for subsequently constructing the mapping relationship between the coupled-state vector and the multi-degree-of-freedom pose adjustment quantities.

[0027] In practical applications, the scale parameter range is typically set based on the spatial resolution of the light intensity distribution information and the spot size, while the orientation angle parameter is uniformly sampled between 0° and 180° to ensure that the dictionary atomic base can cover all possible light intensity gradient directions. The residual signal norm threshold in the preset termination condition can be adaptively set according to the measurement noise level, thereby achieving a balance between sparse representation accuracy and computational efficiency. Through the complete sparse representation process described above, the constructed coupled-state vector has the characteristics of fixed dimensions and clear physical meaning, laying a solid foundation for subsequent alignment adjustment calculations.

[0028] In one optional embodiment, the normalized light intensity distribution information is used as the initial signal to be decomposed. The matching degree between the initial signal to be decomposed and each atomic base in the overcomplete dictionary is calculated. The atomic base with the highest matching degree is selected as the current matching atomic base. The sparse representation coefficients of the current matching atomic base are calculated and stored. Using the standardized light intensity distribution information as the initial signal to be decomposed, spatial gradient calculations are performed on the initial signal to be decomposed and each atomic basis in the overcomplete dictionary to obtain the spatial gradient distribution of the initial signal to be decomposed and the atomic gradient distribution corresponding to each atomic basis. The gradient angle deviation between the spatial gradient distribution and the atomic gradient distribution is calculated at each sampling point in the spatial domain. Sampling points with gradient angle deviation exceeding a preset angle threshold are removed. The product of the gradient magnitudes of the remaining sampling points is spatially integrated to obtain the matching degree of each atomic basis. Select the atomic base with the highest matching degree as the current matching atomic base; The size of the spatial search window is determined by the scale parameter of the current matching atomic base. The initial signal to be decomposed is scanned at multiple positions within the spatial search window. The local correlation peak between the initial signal to be decomposed and the current matching atomic base is calculated at each translation position. The spatial offset when the local correlation peak reaches its maximum is recorded. The maximum value of the local correlation peak and the spatial offset are combined to form the sparse representation coefficient of the current matching atomic base. The sparse representation coefficients of the current matching atomic base are associated with the index position of the current matching atomic base in the overcomplete dictionary and stored.

[0029] In one specific implementation, after the overcomplete dictionary is constructed, the standardized light intensity distribution information is expanded into a column vector as the initial signal to be decomposed, and then the first iteration of iterative matching pursuit is initiated. Standardization typically involves zeroing the mean and normalizing the variance of the original light intensity distribution image to ensure the comparability of light intensity distribution information under different acquisition conditions. The initial signal to be decomposed carries the intensity distribution characteristics of the light field at each sampling point in space under the current coupling state and serves as the basic input for subsequent sparse decomposition.

[0030] When calculating the spatial gradient of the initial signal to be decomposed, a two-dimensional Sobel or Scharr operator is used for convolution in the horizontal and vertical directions to obtain the gradient magnitude and gradient direction angle of the initial signal at each sampling point. These two values ​​together constitute the spatial gradient distribution of the initial signal to be decomposed. Simultaneously, the same spatial gradient calculation process is performed on each atomic basis in the overcomplete dictionary to obtain the atomic gradient distribution corresponding to each atomic basis, including the gradient magnitude and gradient direction angle of each atomic basis at each sampling point. Since the number of atomic bases in the overcomplete dictionary is equal to the product of the total number of scale parameters and the total number of orientation angle parameters, i.e. The computational cost of this step, which involves building up a number of atomic bases, is relatively high. However, since the gradient distribution of each atomic base depends only on the dictionary itself and is independent of the signal to be decomposed, it can be pre-calculated and cached during the dictionary construction stage, thereby avoiding repeated calculations in each iteration and significantly improving the running efficiency.

[0031] At each sampling point in the spatial domain, the gradient direction angle of the initial signal to be decomposed and the gradient direction angle of each atomic basis are taken, and the gradient angle deviation between the two is calculated. The gradient angle deviation reflects the degree of consistency between the signal and the atomic basis in local structural orientation: when the gradient directions at a sampling point are highly consistent, the angle deviation is close to zero; when the local structural orientations differ significantly, the angle deviation is large. The gradient angle deviation is set to exceed a preset angle threshold. The sampling points are marked and discarded. This threshold is usually set according to the anisotropy of the light field distribution, typically ranging from 15° to 45°. Discarding sampling points with inconsistent structural orientations effectively suppresses the interference of edge noise and structural misalignment on the matching degree calculation, allowing the matching process to focus more on regions with similar structural orientations. For the remaining sampling points, the gradient magnitude of the initial signal to be decomposed is multiplied point by point by the gradient magnitude of the corresponding atomic base. Then, the product of all remaining sampling points is spatially integrated (i.e., summed in the discrete case) to obtain the matching degree of the atomic base relative to the initial signal to be decomposed. Match degree The larger the value, the higher the match between the atomic base and the initial signal to be decomposed, both in terms of structural orientation and amplitude distribution.

[0032] Iterate through all entries in the complete dictionary Each atomic base is used to calculate its matching degree. The atomic base with the highest matching degree is selected as the current matching atomic base, and its index number in the overcomplete dictionary is denoted as . The corresponding atomic base is This selection strategy ensures that the most energy-concentrated and structurally relevant components are extracted from the signal to be decomposed in each iteration, which conforms to the basic principle of greedy search in the matching pursuit algorithm. At the same time, the introduction of gradient direction constraints makes the selection process more sensitive to the physical characteristics of the light field structure.

[0033] Determine the current matching atomic base Then, using its corresponding scale parameters Determine the size of the spatial search window. Scale parameter. This describes the spatial extent of the atomic base; larger atomic bases correspond to wider search windows, and smaller atomic bases correspond to narrower search windows. The size of the spatial search window is typically set as a scale parameter. Integer multiples, for example, taking A rectangular region is defined to cover the main energy distribution range of the atomic base. Within this spatial search window, a multi-position translational scan is performed on the initial signal to be decomposed, that is, the two-dimensional arrangement of the initial signal to be decomposed is aligned point by point with the current matching atomic base at different spatial offset positions, and the local correlation peak of the two at each translation position is calculated. Local correlation peak It measures the degree of local linear correlation between the initial signal to be decomposed and the current matching atomic base under a specific translation amount. It is calculated by normalizing the sum of the products of the two in the corresponding region.

[0034] Recording local correlation peaks Spatial offset corresponding to the maximum , This is a two-dimensional vector, representing the translation amounts in the horizontal and vertical directions (units are the sampling point spacing). Spatial offset. This precisely describes the relative positional relationship between the currently matched atomic bases and the initial signal to be decomposed, providing crucial information for locating the fine structure of the optical field distribution. The maximum value of the local correlation peaks... Spatial offset Combined, they together constitute the current matching atomic base. sparse representation coefficients .in The contribution magnitude of this atomic base to the initial signal to be decomposed was characterized. Both characterize the spatial location of the contribution and are indispensable, together fully describing the sparse component characteristics of the current matched atomic base in the initial signal to be decomposed.

[0035] The sparse representation coefficients of the current matching atomic base Rather than the index number in an overcomplete dictionary Associative storage is used to form a sparse component record. The purpose of associative storage is to ensure that the coupled-state vector is constructed in dictionary-ordered atomic sequence. Sparse representation of coefficients, if a certain index position If no atomic basis is selected as the current matching basis, the corresponding component should be filled with zero; if selected, the corresponding sparse representation coefficient should be filled in. This associative storage method preserves sparsity (a large number of components are zero) while ensuring a one-to-one correspondence between each component of the coupled state vector and the dictionary atoms, providing a well-structured data foundation for the subsequent construction of the mapping matrix and the calculation of the difference vector.

[0036] After completing one round of matching pursuit, the contribution of the current matching atomic basis weighted by the sparse representation coefficients is subtracted from the initial signal to be decomposed to obtain the residual signal, and the next round of iteration begins. This process is repeated until the energy of the residual signal is lower than a preset threshold or the maximum number of iterations is reached, at which point the iteration terminates, and all sparse representation coefficients are arranged by index number to form a complete coupled-state vector.

[0037] In one optional embodiment, a mapping matrix is ​​constructed between the coupled-state vector and the pose adjustment amount of the multi-degree-of-freedom position. Each element of the mapping matrix represents the change in the corresponding component of the coupled-state vector caused by a unit adjustment amount for the corresponding degree of freedom, including: For each degree of freedom in the pose adjustment, the control adjustment mechanism generates a unit adjustment in that degree of freedom and collects the perturbation light intensity distribution information of that degree of freedom. The perturbation intensity distribution information is sparsely represented and calculated to obtain the perturbation coupled state vector of this degree of freedom; Calculate the component differences between the perturbation coupled-state vector and the reference coupled-state vector, and arrange the component differences in order of the dimensions of the coupled-state vector to form the sensitivity vector of that degree of freedom; Calculate the cosine of the angle between any two degree of freedom sensitivity vectors. When the absolute value of the cosine exceeds a preset correlation threshold, perform Schmitt orthogonalization on the sensitivity vectors to obtain orthogonal sensitivity vectors. Assemble the orthogonal sensitivity vectors of all degrees of freedom into column vectors in order of degrees of freedom to form a mapping matrix.

[0038] In one specific implementation, the accurate construction of the mapping matrix is ​​a core step in achieving coordinated adjustment of multi-degree-of-freedom poses during the precision alignment of optocouplers. The mapping matrix describes the linear response relationship between the adjustment amount of each degree of freedom and the change in the coupled-state vector; each column element corresponds to the change in each component of the coupled-state vector when a unit adjustment is made in a certain degree of freedom. To ensure that the physical meaning of the mapping matrix is ​​clear and the numerical conditions are favorable, the sensitivity information of each degree of freedom needs to be obtained through experimental calibration, and orthogonalization is performed when there is a linear correlation between the sensitivity vectors.

[0039] For each degree of freedom in the multi-degree-of-freedom pose adjustment, the control adjustment mechanism applies a preset unit adjustment amount to that degree of freedom while keeping the other degrees of freedom unchanged. The "unit adjustment amount" referred to here is not a strictly physical unit quantity, but a small perturbation that is small enough to cause an observable change in the coupled-state vector within the linear response range. Taking a six-degree-of-freedom alignment platform as an example, if the current degree of freedom is a lateral translation along the direction perpendicular to the optical axis, the corresponding piezoelectric actuator is controlled to apply a lateral displacement of a preset step size. The light intensity distribution information between the light source and the receiver is collected at this time and recorded as the perturbation light intensity distribution information for that degree of freedom. To improve calibration accuracy, positive and negative perturbations of the same degree of freedom can be collected separately, and half of the difference between the two perturbation coupled-state vectors is taken as the response corresponding to the unit adjustment amount, thereby eliminating the influence of nonlinear errors and system bias.

[0040] The perturbation light intensity distribution information is processed using the same sparse representation calculation method as the main process. Utilizing the constructed overcomplete dictionary and iterative matching pursuit algorithm, the perturbation coupled-state vector corresponding to that degree of freedom is obtained. Both the perturbation coupled-state vector and the reference coupled-state vector are represented within the same overcomplete dictionary framework; therefore, their dimensions are completely identical, allowing for direct component-by-component difference calculation. The reference coupled-state vector is the coupled-state vector obtained by acquiring light intensity distribution information and performing sparse representation when the adjustment mechanism is in its initial pose state without any perturbation. It represents the reference reference in the current alignment state.

[0041] Let a certain degree of freedom The perturbation coupled state vector is The reference coupled state vector is Then the sensitivity vector of that degree of freedom Defined as: ; in Each component corresponds to the change in response along a certain dictionary atom direction in the coupled-state vector, arranged in dimensional order according to the coupled-state vector. The physical meaning of the sensitivity vector is: when the first... When a unit adjustment is made in each degree of freedom, the corresponding changes occur in each component of the coupled-state vector. The magnitude of the sensitivity vector reflects the degree of influence of that degree of freedom on the coupled state; the larger the magnitude, the more sensitive the adjustment of that degree of freedom is to the optical coupling state.

[0042] In practical optocoupler systems, there may be strong linear correlations between sensitivity vectors of different degrees of freedom. For example, tilting and lateral translation along the optical axis may produce similar patterns of change in light intensity distribution, causing the cosine of the angle between the corresponding sensitivity vectors to be close to 1 or -1. If these linearly correlated sensitivity vectors are directly assembled into column vectors of a mapping matrix, the mapping matrix will be close to singular, and its inverse operation (or pseudo-inverse operation) will introduce large numerical errors, resulting in severely distorted calculations of pose adjustment.

[0043] Therefore, for any two degrees of freedom and Sensitivity vector and Calculate the cosine of the included angle. : ; in Represents the sensitivity vector and The angle between them This represents the Euclidean norm. When... Exceeding the preset correlation threshold When the value is 0.85 (for example), it is assumed that there is a significant linear correlation between the sensitivity vectors of the two degrees of freedom, and it is necessary to perform Schmidt orthogonalization on them.

[0044] The Schmidt orthogonalization process is performed sequentially according to the order of the degrees of freedom. The sensitivity vector for the first degree of freedom... It is directly normalized and used as the first orthogonal sensitivity vector. Sensitivity vector for the second degree of freedom First subtract its value. The projection components along the direction are used to obtain the residual vector. The residual vector is then normalized to obtain the second orthogonal sensitivity vector. And so on, for the first... Sensitivity vector with one degree of freedom Its orthogonal sensitivity vector The calculation process is as follows: ; ; in The unnormalized intermediate vector after orthogonalization. This is an index of the degrees of freedom that have been processed during the orthogonalization process. This is the index of the degrees of freedom currently being processed. It's important to note that Schmidt orthogonalization is not performed unconditionally on all degrees of freedom, but only when a correlation exceeding a threshold is detected. Only when the relevant degree of freedom is processed, for sensitivity vectors whose absolute value of the included angle cosine is lower than the threshold, only normalization is performed without changing their direction, so as to preserve the information of the original physical sensitivity direction.

[0045] After completing the orthogonal sensitivity vector calculation for all degrees of freedom, the orthogonal sensitivity vectors of each degree of freedom are... Arranged sequentially according to degrees of freedom, they form column vectors, constituting a mapping matrix. : ; in The total number of degrees of freedom involved in alignment adjustment. The number of rows equals the dimension of the coupled-state vector, and the number of columns equals the number of degrees of freedom. Because the column vectors have been orthogonalized, The column vectors are linearly independent, the condition number of the matrix is ​​effectively controlled, and the numerical stability is significantly improved when the difference vector is converted into adjustment quantities of each degree of freedom through inverse operation or pseudo-inverse operation.

[0046] In practical engineering implementation, when the working environment of the optocoupler changes (such as temperature drift, mechanical creep, etc.) and the system response characteristics change, it is necessary to periodically re-execute the above mapping matrix calibration process to update the sensitivity vector and orthogonalization results. This is to ensure that the mapping matrix always accurately reflects the true response relationship between each degree of freedom and the coupling state vector under the current system state, thereby ensuring the accuracy of subsequent pose adjustment calculations.

[0047] In one optional embodiment, calculating the difference vector between the current coupled-state vector and the target coupled-state vector, and converting the difference vector into adjustment values ​​for each degree of freedom through the inverse operation of the mapping matrix includes: Obtain the current light intensity distribution information, perform sparse representation calculation on the current light intensity distribution information, and obtain the current coupled state vector; Calculate the component differences between the current coupled state vector and the target coupled state vector in each dimension, and arrange the component differences in each dimension in dimensional order to form a difference vector; Singular value decomposition is performed on the mapping matrix to obtain the sequence of singular values ​​of the mapping matrix. The ratio of the maximum singular value to the minimum singular value in the sequence of singular values ​​is calculated as the condition number. When the condition number exceeds the preset ill-conditioned threshold, a regularization coefficient is superimposed on the diagonal elements of the mapping matrix to obtain a regularized mapping matrix. Perform matrix inversion on the regularized mapping matrix to obtain the inverse mapping matrix; Perform matrix multiplication on the difference vector and the inverse mapping matrix to obtain the initial adjustment amount. Iterate through the adjustment amplitude corresponding to each degree of freedom in the initial adjustment amount, and compare each adjustment amplitude with the maximum allowable adjustment amplitude of the corresponding degree of freedom. When the adjustment amplitude exceeds the maximum allowable adjustment amplitude, replace the adjustment amplitude with the maximum allowable adjustment amplitude to obtain the adjustment amount of each degree of freedom.

[0048] In one specific implementation, the light intensity distribution information of the optocoupler at the current moment is obtained, expanded into a column vector, and processed according to the same standardized process as constructing an overcomplete dictionary. Then, iterative matching and tracing calculations are performed to obtain the sparse representation coefficient sequence of the current light intensity distribution information in the overcomplete dictionary. This sequence is then assembled into the current coupled-state vector according to the dictionary's atomic arrangement order. The target coupled-state vector corresponds to the sparse representation coefficient sequence when the light source and receiver reach the optimal coupling state. This sequence can be obtained during the system calibration phase by performing the same sparse representation process on the light intensity distribution information under the known optimal alignment pose and stored as a reference value. The current coupled-state vector and the target coupled-state vector have the same dimension, both equal to the total number of atomic bases in the overcomplete dictionary.

[0049] Calculate the component differences between the current coupled-state vector and the target coupled-state vector dimension by dimension-by-dimensional calculation. This involves subtracting the coefficients of the two vectors in each dimension and arranging the resulting dimensional differences in the same order as the coupled-state vectors to form the difference vector. Difference vector This fully describes the deviation between the current coupling state and the target coupling state in the sparse representation space. The larger the absolute value of each component, the more significant the deviation in the corresponding atomic base direction. This difference vector will serve as the input for subsequent inverse operations, and will be mapped to the adjustment amount of each degree of freedom through the inverse operation of the mapping matrix.

[0050] For the mapping matrix Perform singular value decomposition, decomposing it into the product of a left singular matrix, a singular value diagonal matrix, and a right singular matrix. The diagonal elements of the singular value diagonal matrix form the sequence of singular values, and the largest singular value in the sequence is denoted as . The smallest singular value is denoted as Calculate the ratio of the two. As the condition number of the mapping matrix . condition number This reflects the amplification effect of the mapping matrix on input perturbations during inversion: the larger the condition number, the closer the matrix is ​​to singularity, and the more sensitive the inversion result is to small errors in the difference vector. Direct inversion will cause large fluctuations in the adjustment of each degree of freedom, affecting the stability of the alignment process.

[0051] Let the preset ill-condition threshold be... , when the condition number Exceed When the mapping matrix is ​​considered to be in an ill-conditioned state, regularization is required. The regularization process involves applying a regularization method to the mapping matrix... Regularization coefficients are superimposed on the diagonal elements. That is, constructing a regularization mapping matrix. ,in To and Identity matrices of the same order This is the regularization coefficient. The selection needs to take into account two aspects: if If it is too small, its effect on improving the pathological problem will be limited; if If the value is too large, it will introduce a significant systematic bias, causing the inverse operation result to deviate from the actual adjustment amount. In practical applications, it can be based on the condition number. Size adaptive determination The value of , for example, let Proportional to and The ratio of the condition number to the condition number allows for reasonable regularization under varying degrees of ill-conditioned conditions. Not exceeding When the mapping matrix is ​​considered well-state, no regularization is needed, and it can be directly expressed as... It participates in the subsequent inversion operation.

[0052] For regularization mapping matrix (or, in the case of a well-formed system, the original mapping matrix) Perform the matrix inversion operation to obtain the inverse mapping matrix. In practical calculations, if the mapping matrix is ​​a square matrix with full rank, the standard matrix inversion can be performed directly. If the mapping matrix is ​​not square or has a numerical rank deficiency, a pseudo-inverse can be constructed using the singular value decomposition result. Components with singular values ​​less than a set truncation threshold are set to zero before calculating the pseudo-inverse to further improve numerical stability. (Inverse mapping matrix) A reverse mapping relationship was established from the difference vector in the sparse representation space to the adjustment space of each degree of freedom.

[0053] Difference vector With inverse mapping matrix Perform matrix multiplication to obtain the preliminary adjustment vector. ,Right now Vector Each component corresponds to an initial adjustment amplitude for one degree of freedom, with its sign indicating the adjustment direction and its absolute value indicating the adjustment magnitude. Because some components of the difference vector may be large during the inverse operation, or the local sensitivity of the mapping matrix may be low, the adjustment amplitude of individual degrees of freedom in the initial adjustment may exceed the physical limits or safe movement range of the adjustment mechanism. Therefore, amplitude limiting processing is required for the initial adjustment.

[0054] Traversing the initial adjustment vector The adjustment amplitude corresponding to each degree of freedom, let the first degree be... The initial adjustment amplitude for each degree of freedom is The corresponding maximum allowable adjustment range is .Will and Comparison: If If the adjustment amount for that degree of freedom remains unchanged; if Then replace the adjustment value with At the same time, the sign of the original adjustment direction is retained, that is, the adjustment amount of that degree of freedom is set to... After the amplitude limiting process described above, the resulting adjustment values ​​for each degree of freedom constitute the final adjustment vector, ensuring that the single adjustment amplitude of each degree of freedom is within the allowable range of the adjustment mechanism, thus avoiding mechanical shock or a sudden drop in coupling efficiency due to excessive adjustment. Maximum allowable adjustment amplitude. The physical stroke, driving accuracy, and system dynamic characteristics of the adjustment mechanism can be preset during the calibration stage according to each degree of freedom, and remain fixed during the alignment iteration process.

[0055] Through the above process, the coupled state difference information in the sparse representation space is reliably converted into effective adjustment quantities for each degree of freedom, taking into account both the numerical stability of the mapping matrix in the ill-conditioned case and the physical constraints of the adjustment mechanism, providing accurate and safe control commands for subsequent pose adjustment.

[0056] like Figure 2 The diagram shows the logical flowchart for mapping matrix analysis and spectral characteristic extraction.

[0057] In one optional embodiment, the mapping matrix is ​​orthogonally decomposed to obtain the response modes and corresponding eigenvalues, and time-frequency evolution analysis is performed on each response mode to obtain the spectral distribution characteristics, including: The mapping matrix is ​​orthogonally decomposed to obtain each response mode and its corresponding eigenvalue. The magnitude of each eigenvalue is calculated as the contribution of the corresponding response mode to the coupling efficiency. Multiple sets of mapping matrices are collected within a preset time window at a preset sampling interval. Orthogonal decomposition is performed on each set of mapping matrices to obtain the response modes and eigenvalues ​​at the corresponding time. The similarity of the eigenvectors of each response mode at the current time and the response modes at the previous time is calculated. The response mode pairs with the highest similarity are determined as the correspondence of the same response mode at different times. The feature values ​​of the same response mode at different times are extracted according to the correspondence, and the feature values ​​at each time are arranged in chronological order to form the dynamic evolution sequence of the response mode; A fast Fourier transform is performed on the dynamic evolution sequence to obtain the amplitude spectrum and phase spectrum in the frequency domain. The frequency components with amplitudes greater than the average amplitude in the amplitude spectrum and their corresponding phase information are extracted to form the spectral distribution characteristics of the response mode.

[0058] In one specific implementation, the mapping matrix Orthogonal decomposition can be performed using eigenvalue decomposition, which decomposes the response into combinations of several response modes and their corresponding eigenvalues. Specifically, for After performing eigenvalue decomposition, a set of eigenvectors is obtained, and each eigenvector corresponds to a response mode, denoted as the eigenvector i. The feature vectors of the response modes are The corresponding feature value is In a physical sense, response modes represent the independent motion components of an optocoupler system in a multi-degree-of-freedom space. Different response modes are orthogonal to each other and do not interfere with one another. The magnitudes of each eigenvalue are calculated. As the contribution of a corresponding response mode to coupling efficiency, a larger modulus indicates a more significant impact of that response mode on the light intensity distribution during coupling adjustment, and it should be given higher weight in subsequent adjustments. Response modes with smaller modulus values ​​reflect lower system sensitivity in that direction, meaning that small changes in the corresponding adjustment amount will not cause a significant response in the coupled-state vector. By ranking the contributions, the key response modes that dominate the changes in coupling efficiency can be identified, providing a basis for optimizing subsequent mode adjustment amounts.

[0059] Building upon single-moment feature decomposition, a time dimension is further introduced within a preset time window. Within the preset sampling interval Collect multiple sets of mapping matrices, denoted as ,in The total number of samples within the time window, satisfying For each mapping matrix, orthogonal decomposition is performed to obtain the set of eigenvectors and eigenvalues ​​for the response modes at each time step. Since the mapping matrices differ slightly at different times, the order of their eigenvectors may change. Directly following the decomposition order would cause jumps in the time series for the same physical response mode, failing to accurately reflect its dynamic evolution. To address this issue, eigenvector similarity is calculated for response modes at adjacent times. Specifically, the similarity is calculated for the eigenvectors of the current time step. Response mode feature vectors The absolute value of the inner product between the response mode and the eigenvectors of all response modes at the previous time step is used to determine the correspondence between the response modes at adjacent time steps. The formula for calculating the absolute value of the inner product is as follows: ,in Indicates the current time. The response mode is the same as the previous time step. Similarity between response modes This indicates the inner product operation. When... When the value is close to 1, it indicates that the two response modes are highly similar and can be regarded as the manifestations of the same physical mode at different times; when A significant deviation from 1 indicates a substantial change in the response mode, and it is necessary to record the bifurcation or merging events of that mode.

[0060] Based on the above correspondence, each response mode is traced along the time axis, and its feature values ​​at each time point are extracted. Arranging these eigenvalues ​​in chronological order constitutes the dynamic evolution sequence of this response mode. The dynamic evolution sequence fully records the change of the contribution of the response mode over time. The change of the real part of the eigenvalue reflects the dynamic fluctuation of the mode's contribution to the coupling efficiency gain, while the change of the imaginary part reflects the drift of the phase response characteristics represented by the mode over time. In practical optocoupler systems, due to factors such as mechanical vibration, thermal drift, and environmental disturbances, the eigenvalues ​​of each response mode often exhibit periodic or quasi-periodic time evolution patterns. By extracting these patterns, the future coupling state of the system can be predicted, providing a basis for feedforward compensation.

[0061] For dynamic evolution sequences Performing a Fast Fourier Transform converts the time-domain sequence to the frequency domain, yielding the amplitude spectrum of the response mode. and phase spectrum ,in Represents the frequency variable, with a frequency resolution of The highest resolvable frequency is the Nyquist frequency. Amplitude spectrum This reflects the energy distribution and phase spectrum of the response mode at each frequency component. This describes the initial phase information of each frequency component. To extract the dominant frequency component with practical physical meaning from the amplitude spectrum, the average amplitude of the amplitude spectrum is calculated. , will satisfy The frequency components are selected, and the set of these frequency components is denoted as . The corresponding phase information set is The selected frequency components and their phase information together constitute the spectral distribution characteristics of the response mode, which are used for subsequent phase compensation calculations.

[0062] The following technical details need to be considered when extracting spectral distribution characteristics: Sampling interval. The selection of the sampling frequency should satisfy the Nyquist sampling theorem, that is, the sampling frequency should not be less than twice the highest mechanical vibration frequency in the system, in order to avoid frequency aliasing and spectral distortion. Time window The selected value should be long enough to ensure frequency resolution. The time window should be able to distinguish adjacent vibration frequency components, but it should not be too long to avoid non-stationary changes in the mapping matrix within the time window, which would compromise the effectiveness of the spectral analysis. In practical engineering applications, the range of major interference frequencies can be estimated in advance based on the known vibration characteristics of the system, and a suitable frequency range can be selected accordingly. and Parameter combination. In addition, before performing a Fast Fourier Transform on the dynamically evolving sequence, a Hanning window or other smoothing window function should be applied to weight both ends of the sequence to reduce spectral leakage and improve the frequency resolution accuracy and amplitude estimation accuracy of the amplitude spectrum.

[0063] Through the above orthogonal decomposition and time-frequency evolution analysis process, the corresponding spectral distribution characteristics of each response mode were obtained, including the dominant frequency set. The amplitude and phase information corresponding to each frequency are also included. These spectral distribution characteristics will be directly used in the subsequent phase compensation stage. By applying phase lead or lag compensation corresponding to the spectral distribution characteristics to each modal adjustment, the dynamic delay effect of the system's mechanical response is offset, thereby significantly improving the convergence speed and steady-state accuracy of multi-degree-of-freedom pose adjustment.

[0064] In one optional embodiment, the modal adjustment amount is obtained by projecting the adjustment amount of each degree of freedom onto the direction of each response mode, performing phase compensation based on the spectral distribution characteristics, and mapping it to the optimized adjustment amount of each degree of freedom through inverse transformation, thereby driving the adjustment mechanism to perform pose adjustment, including: Convert the adjustment values ​​of each degree of freedom into vector form, and calculate the projection components of the vectors on each response mode as the initial mode adjustment values; The frequency component with the largest amplitude is extracted from the spectral distribution characteristics of each response mode as the dominant frequency component. The phase value corresponding to the dominant frequency component is extracted, and the response mode with the smallest phase is selected as the phase reference mode. The phase lag of other response modes relative to the phase reference mode is calculated, and the synchronization adjustment timing of each response mode is determined by sorting the phase lag from smallest to largest. The phase prediction value for the next moment is calculated based on the phase lag and dominant frequency of each response mode. The phase prediction value is then compared with the phase of the phase reference mode to obtain the dynamic phase deviation. The initial mode adjustment amount is then adjusted by time delay based on the dynamic phase deviation to obtain the mode adjustment amount for phase alignment. The optimized mode adjustment is obtained by multiplying the phase-aligned mode adjustment amount with the contribution of the corresponding response mode. The optimized mode adjustment amount is then inversely transformed with each response mode to obtain the optimized adjustment amount for each degree of freedom, which drives the adjustment mechanism to perform pose adjustment.

[0065] In one specific implementation, after obtaining the adjustment values ​​for each degree of freedom, they are arranged in order of degree of freedom to form an adjustment vector in column vector form, denoted as . This vector integrates the adjustment requirements of all degrees of freedom involved in alignment adjustment, but it does not yet consider the frequency characteristic differences of each response mode during dynamic coupling. To match the adjustment commands with the system's response characteristics, it is necessary to... Projected onto the directions of each response mode, it is decomposed into the initial adjustment components of each mode.

[0066] For the There are response modes, and their eigenvectors are: Adjust the vector The projection component in the direction of this response mode is defined as the initial mode adjustment amount. The calculation method is as follows This projection operation decomposes the multi-degree-of-freedom coupled adjustment quantities into the orthogonal response mode spaces, allowing each response mode to independently undertake the adjustment task in its corresponding direction, thus avoiding interference between adjustment commands from different degrees of freedom. The projection operation is performed on each response mode sequentially to obtain the initial modal adjustment quantity sequence. This provides input for subsequent phase compensation.

[0067] From the spectral distribution characteristics of each response mode, the frequency component with the largest amplitude is extracted as the dominant frequency component of that mode. Specifically, for the ... There are several response modes, and their amplitude spectra are as follows: When the amplitude is greater than the average amplitude frequency component set In the middle, find the Frequency of achieving the maximum value ,Right now ,Will Determined as the number The dominant frequency of each response mode. Correspondingly, from the phase spectrum Extract The phase value at that point is denoted as This phase value reflects the phase state of the response mode at the current moment.

[0068] After obtaining the dominant frequency phase values ​​of each response mode Then, the response mode with the smallest phase value is selected as the phase reference mode, and its index is denoted as . The corresponding phase is denoted as Using the phase reference mode as a benchmark, calculate the phase lag of each other response mode relative to the phase reference mode. For all response modes, according to the phase lag... By sorting the responses from smallest to largest, the synchronization timing of each mode is obtained. Response modes ranked earlier have smaller phase lags, indicating that their adjustments should be triggered relatively early; response modes ranked later have larger phase lags, meaning their adjustments should be executed sequentially after the preceding modes have completed their adjustments. This timing arrangement ensures that the adjustments of each response mode are coordinated and consistent in the time dimension, avoiding dynamic conflicts between degrees of freedom adjustments caused by phase asynchrony.

[0069] Based on the phase hysteresis of each response mode and dominant frequency Calculate the phase prediction values ​​of each response mode at the next adjustment time. Let the current time be... The next adjustment time is Then the first The phase prediction value of each response mode at the next time step is ,in The preset sampling interval is used to predict the phase values. Phase prediction of the phase reference mode at the next time step Perform interpolation to obtain the dynamic phase deviation. Dynamic phase deviation reflects the real-time phase difference between each response mode and the reference mode at the next adjustment time, and is the core basis for time delay adjustment.

[0070] Based on dynamic phase deviation Initial mode adjustment Perform time delay adjustment. Convert the dynamic phase deviation into a corresponding time delay. Apply a time delay to the initial mode adjustment amount, so that the first... The adjustment instructions for each response mode are shifted backward in timing. This ensures phase alignment with the adjustment action of the phase reference mode during actual execution. The mode adjustment amount after time delay adjustment is denoted as... This refers to the phase-aligned mode adjustment. This adjustment is numerically similar to the initial mode adjustment. To maintain consistency, only the execution timing was modified to ensure that the adjustment actions of each response mode are synchronized at the phase level.

[0071] After obtaining the modal adjustment amount for phase alignment Subsequently, the contribution of each response mode is further introduced to weight and correct the adjustment amount. The contribution of each response mode is then used. The eigenvalues ​​corresponding to this mode Normalization yields the result, specifically calculated as follows: Response modes with larger eigenvalues ​​play a more dominant role in the overall alignment adjustment, and their contribution weight is correspondingly higher. The modal adjustment amount for phase alignment... With contribution Perform the product operation to obtain the optimized modal adjustment amount. This weighting operation allows high-contribution response modes to play a more significant role in the adjustment process, while low-contribution modes have their adjustment magnitude reduced accordingly, thereby improving the overall stability and convergence speed of the adjustment.

[0072] Optimize modal adjustment amount With the corresponding response mode feature vector Perform an inverse transformation to map the optimization adjustment values ​​in each modal space back to their respective degrees of freedom spaces, obtaining the optimization adjustment vectors for each degree of freedom. The inverse transform is calculated as follows: The inverse transformation process involves superimposing the optimization adjustments of each response mode along the eigenvector direction to restore the actual adjustment commands for each degree of freedom. This inverse transformation process ensures the consistency between the mapping relationship from mode space back to degree-of-freedom space and the forward decomposition, thus ensuring the physical executability of the adjustment commands.

[0073] Obtain the optimization adjustment vector for each degree of freedom Then, each component is sent to the corresponding adjustment mechanism driver, which drives the adjustment mechanism to perform pose adjustment actions according to the optimized adjustment amount. After receiving the drive command, the adjustment mechanism activates the adjustment actions corresponding to each response mode in sequence according to the synchronous adjustment timing and the phase lag amount, so that the light source and the receiver move in a coordinated manner in multiple degrees of freedom, gradually approaching the target pose state. After the adjustment is completed, the light intensity distribution information is re-acquired to determine whether the preset alignment conditions are met. If not, the iterative adjustment process is entered again until the light intensity distribution information reaches the preset alignment requirements.

[0074] A second aspect of the present invention provides an intelligent precision alignment system for an optocoupler, comprising: The light intensity acquisition unit is used to acquire light intensity distribution information between the light source end and the receiving end in the optocoupler. The sparse representation unit is used to sparsely represent light intensity distribution information. It constructs an overcomplete dictionary containing multiple atomic bases, calculates the sparse representation coefficients of light intensity distribution information on the overcomplete dictionary through iterative matching and tracing, and arranges the sparse representation coefficients according to the atomic order of the dictionary to form a coupled state vector. The mapping matrix unit is used to construct a mapping matrix between the coupled state vector and the pose adjustment amount of the multi-degree-of-freedom. Each element of the mapping matrix is ​​the change in the corresponding component of the coupled state vector caused by the unit adjustment amount of the corresponding degree of freedom. The difference vector between the current coupled state vector and the target coupled state vector is calculated. The difference vector is converted into the adjustment amount of each degree of freedom through the inverse operation of the mapping matrix. The modal analysis unit is used to perform orthogonal decomposition of the mapping matrix to obtain the response modes and corresponding eigenvalues, perform time-frequency evolution analysis on each response mode to obtain the spectral distribution characteristics, project the adjustment amount of each degree of freedom to the direction of each response mode to obtain the modal adjustment amount, perform phase compensation according to the spectral distribution characteristics, and map it to the optimized adjustment amount of each degree of freedom through inverse transformation to drive the adjustment mechanism to perform pose adjustment. The iterative adjustment unit is used to repeatedly perform adjustments until the light intensity distribution information meets the preset alignment conditions.

[0075] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0076] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0077] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent precision alignment of optocoupler equipment, characterized in that, include: To obtain light intensity distribution information between the light source end and the receiving end in an optocoupler; The light intensity distribution information is sparsely represented, an overcomplete dictionary containing multiple atomic bases is constructed, and the sparse representation coefficients of the light intensity distribution information on the overcomplete dictionary are calculated by iterative matching and tracing. The sparse representation coefficients are arranged in the order of the dictionary atoms to form a coupled state vector. Construct a mapping matrix between the coupled state vector and the pose adjustment amount of multiple degrees of freedom. Each element of the mapping matrix is ​​the change in the corresponding component of the coupled state vector caused by the unit adjustment amount of the corresponding degree of freedom. Calculate the difference vector between the current coupled state vector and the target coupled state vector. Convert the difference vector into the adjustment amount of each degree of freedom through the inverse operation of the mapping matrix. The mapping matrix is ​​orthogonally decomposed to obtain the response modes and their corresponding eigenvalues. Time-frequency evolution analysis is performed on each response mode to obtain the spectral distribution characteristics. The adjustment amount of each degree of freedom is projected onto the direction of each response mode to obtain the mode adjustment amount. Phase compensation is performed according to the spectral distribution characteristics, and the inverse transformation is used to map it into the optimized adjustment amount of each degree of freedom to drive the adjustment mechanism to perform pose adjustment. Repeat the adjustment until the light intensity distribution information meets the preset alignment conditions.

2. The method according to claim 1, characterized in that, The light intensity distribution information is sparsely represented by constructing an overcomplete dictionary containing multiple atomic bases. The sparse representation coefficients of the light intensity distribution information on the overcomplete dictionary are calculated through iterative matching and tracing. These sparse representation coefficients are then arranged according to the atomic order of the dictionary to form a coupled state vector, including: The light intensity distribution information is normalized to obtain standardized light intensity distribution information; Multiple scale parameters and multiple direction angle parameters are set. For each combination of scale parameters and direction angle parameters, the standardized light intensity distribution information is spatially filtered at the corresponding scale, and gradient features are extracted at the corresponding direction angle. The extracted gradient features are used as an atomic basis. The atomic basis corresponding to all combinations of scale parameters and direction angle parameters are summarized to form an overcomplete dictionary. The standardized light intensity distribution information is used as the initial signal to be decomposed. The matching degree between the initial signal to be decomposed and each atomic base in the overcomplete dictionary is calculated. The atomic base with the largest matching degree is selected as the current matching atomic base. The sparse representation coefficients of the current matching atomic base are calculated and stored. Based on the current matching atomic base and the corresponding sparse representation coefficients, the contribution component of the current matching atomic base is removed from the initial signal to be decomposed to obtain the residual signal. It is then determined whether the residual signal meets the preset termination condition. If not, the residual signal is used as the new initial signal to be decomposed, and the process is repeated. The sparse representation coefficients obtained in each iteration are arranged in the order of the corresponding atomic bases in the overcomplete dictionary to form a coupled state vector.

3. The method according to claim 2, characterized in that, Using the standardized light intensity distribution information as the initial signal to be decomposed, the matching degree between the initial signal to be decomposed and each atomic basis in the overcomplete dictionary is calculated. The atomic basis with the highest matching degree is selected as the current matching atomic basis. The sparse representation coefficients of the current matching atomic basis are calculated and stored, including: Using the standardized light intensity distribution information as the initial signal to be decomposed, spatial gradient calculations are performed on the initial signal to be decomposed and each atomic basis in the overcomplete dictionary to obtain the spatial gradient distribution of the initial signal to be decomposed and the atomic gradient distribution corresponding to each atomic basis. The gradient angle deviation between the spatial gradient distribution and the atomic gradient distribution is calculated at each sampling point in the spatial domain. Sampling points with gradient angle deviation exceeding a preset angle threshold are removed. The product of the gradient magnitudes of the remaining sampling points is spatially integrated to obtain the matching degree of each atomic basis. Select the atomic base with the highest matching degree as the current matching atomic base; The size of the spatial search window is determined by the scale parameter of the current matching atomic base. The initial signal to be decomposed is scanned at multiple positions within the spatial search window. The local correlation peak between the initial signal to be decomposed and the current matching atomic base is calculated at each translation position. The spatial offset when the local correlation peak reaches its maximum is recorded. The maximum value of the local correlation peak and the spatial offset are combined to form the sparse representation coefficient of the current matching atomic base. The sparse representation coefficients of the current matching atomic base are associated with the index position of the current matching atomic base in the overcomplete dictionary and stored.

4. The method according to claim 1, characterized in that, A mapping matrix is ​​constructed between the coupled-state vector and the pose adjustment amount of multiple degrees of freedom. Each element of the mapping matrix represents the change in the corresponding component of the coupled-state vector caused by a unit adjustment amount for the corresponding degree of freedom, including: For each degree of freedom in the pose adjustment, the control adjustment mechanism generates a unit adjustment in that degree of freedom and collects the perturbation light intensity distribution information of that degree of freedom. The perturbation intensity distribution information is sparsely represented and calculated to obtain the perturbation coupled state vector of this degree of freedom; Calculate the component differences between the perturbation coupled-state vector and the reference coupled-state vector, and arrange the component differences in order of the dimensions of the coupled-state vector to form the sensitivity vector of that degree of freedom; Calculate the cosine of the angle between any two degree of freedom sensitivity vectors. When the absolute value of the cosine exceeds a preset correlation threshold, perform Schmitt orthogonalization on the sensitivity vectors to obtain orthogonal sensitivity vectors. Assemble the orthogonal sensitivity vectors of all degrees of freedom into column vectors in order of degrees of freedom to form a mapping matrix.

5. The method according to claim 1, characterized in that, Calculate the difference vector between the current coupled-state vector and the target coupled-state vector, and convert the difference vector into adjustment values ​​for each degree of freedom through the inverse operation of the mapping matrix, including: Obtain the current light intensity distribution information, perform sparse representation calculation on the current light intensity distribution information, and obtain the current coupled state vector; Calculate the component differences between the current coupled state vector and the target coupled state vector in each dimension, and arrange the component differences in each dimension in dimensional order to form a difference vector; Singular value decomposition is performed on the mapping matrix to obtain the sequence of singular values ​​of the mapping matrix. The ratio of the maximum singular value to the minimum singular value in the sequence of singular values ​​is calculated as the condition number. When the condition number exceeds the preset ill-conditioned threshold, a regularization coefficient is superimposed on the diagonal elements of the mapping matrix to obtain a regularized mapping matrix. Perform matrix inversion on the regularized mapping matrix to obtain the inverse mapping matrix; Perform matrix multiplication on the difference vector and the inverse mapping matrix to obtain the initial adjustment amount. Iterate through the adjustment amplitude corresponding to each degree of freedom in the initial adjustment amount, and compare each adjustment amplitude with the maximum allowable adjustment amplitude of the corresponding degree of freedom. When the adjustment amplitude exceeds the maximum allowable adjustment amplitude, replace the adjustment amplitude with the maximum allowable adjustment amplitude to obtain the adjustment amount of each degree of freedom.

6. The method according to claim 1, characterized in that, Orthogonal decomposition of the mapping matrix yields the response modes and their corresponding eigenvalues. Time-frequency evolution analysis of each response mode yields its spectral distribution characteristics, including: The mapping matrix is ​​orthogonally decomposed to obtain each response mode and its corresponding eigenvalue. The magnitude of each eigenvalue is calculated as the contribution of the corresponding response mode to the coupling efficiency. Multiple sets of mapping matrices are collected within a preset time window at a preset sampling interval. Orthogonal decomposition is performed on each set of mapping matrices to obtain the response modes and eigenvalues ​​at the corresponding time. The similarity of the eigenvectors of each response mode at the current time and the response modes at the previous time is calculated. The response mode pairs with the highest similarity are determined as the correspondence of the same response mode at different times. The feature values ​​of the same response mode at different times are extracted according to the correspondence, and the feature values ​​at each time are arranged in chronological order to form the dynamic evolution sequence of the response mode; A fast Fourier transform is performed on the dynamic evolution sequence to obtain the amplitude spectrum and phase spectrum in the frequency domain. The frequency components with amplitudes greater than the average amplitude in the amplitude spectrum and their corresponding phase information are extracted to form the spectral distribution characteristics of the response mode.

7. The method according to claim 6, characterized in that, The modal adjustment is obtained by projecting the adjustment values ​​of each degree of freedom onto the direction of each response mode. Phase compensation is performed based on the spectral distribution characteristics, and the result is mapped to the optimized adjustment values ​​of each degree of freedom through inverse transformation. The adjustment mechanism is then driven to perform pose adjustment, including: Convert the adjustment values ​​of each degree of freedom into vector form, and calculate the projection components of the vectors on each response mode as the initial mode adjustment values; The frequency component with the largest amplitude is extracted from the spectral distribution characteristics of each response mode as the dominant frequency component. The phase value corresponding to the dominant frequency component is extracted, and the response mode with the smallest phase is selected as the phase reference mode. The phase lag of other response modes relative to the phase reference mode is calculated, and the synchronization adjustment timing of each response mode is determined by sorting the phase lag from smallest to largest. The phase prediction value for the next moment is calculated based on the phase lag and dominant frequency of each response mode. The phase prediction value is then compared with the phase of the phase reference mode to obtain the dynamic phase deviation. The initial mode adjustment amount is then adjusted by time delay based on the dynamic phase deviation to obtain the mode adjustment amount for phase alignment. The optimized mode adjustment is obtained by multiplying the phase-aligned mode adjustment amount with the contribution of the corresponding response mode. The optimized mode adjustment amount is then inversely transformed with each response mode to obtain the optimized adjustment amount for each degree of freedom, which drives the adjustment mechanism to perform pose adjustment.

8. An intelligent precision alignment system for optocouplers, used to implement the method as described in any one of claims 1-7, characterized in that, include: The light intensity acquisition unit is used to acquire light intensity distribution information between the light source end and the receiving end in the optocoupler. The sparse representation unit is used to sparsely represent light intensity distribution information. It constructs an overcomplete dictionary containing multiple atomic bases, calculates the sparse representation coefficients of light intensity distribution information on the overcomplete dictionary through iterative matching and tracing, and arranges the sparse representation coefficients according to the atomic order of the dictionary to form a coupled state vector. The mapping matrix unit is used to construct a mapping matrix between the coupled state vector and the pose adjustment amount of the multi-degree-of-freedom. Each element of the mapping matrix is ​​the change in the corresponding component of the coupled state vector caused by the unit adjustment amount of the corresponding degree of freedom. The difference vector between the current coupled state vector and the target coupled state vector is calculated. The difference vector is converted into the adjustment amount of each degree of freedom through the inverse operation of the mapping matrix. The modal analysis unit is used to perform orthogonal decomposition of the mapping matrix to obtain the response modes and corresponding eigenvalues, perform time-frequency evolution analysis on each response mode to obtain the spectral distribution characteristics, project the adjustment amount of each degree of freedom to the direction of each response mode to obtain the modal adjustment amount, perform phase compensation according to the spectral distribution characteristics, and map it to the optimized adjustment amount of each degree of freedom through inverse transformation to drive the adjustment mechanism to perform pose adjustment. The iterative adjustment unit is used to repeatedly perform adjustments until the light intensity distribution information meets the preset alignment conditions.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images