A method, apparatus, device and medium for determining an optical processing dwell time

Abstract

The application discloses a kind of optical processing dwell time determination method, device, equipment and medium, it is related to optical element processing field, comprising: obtaining target removal amount, tool influence function and multiple path points on machining path;Determine the mapping weight between each path point and the local grid node of pre-established auxiliary rule grid;According to the current dwell time of each path point and mapping weight, the dwell time of each path point is distributed to the corresponding grid node of auxiliary rule grid, and the obtained grid residence amount is convolved with tool influence function Operation, obtain predicted removal amount;Using mapping weight, the residual between target removal amount and predicted removal amount is transmitted from auxiliary rule grid to each path point, to obtain the back propagation amount of each path point, to update the dwell time of each path point;Wherein convolution parameter and back propagation parameter are consistent.

Description

Technical Field

[0001] This invention relates to the field of optical component processing, and in particular to a method, apparatus, equipment, and medium for determining the dwell time during optical processing. Background Technology

[0002] In the field of optical component processing, during computer-controlled optical surface shaping (CCOS) processes such as polishing, grinding, or lapping, it is usually necessary to calculate the dwell time or equivalent feed rate of the tool at each path point so that the cumulative removal amount generated after the tool moves along the path is as close as possible to the target correction amount.

[0003] Traditional methods for calculating dwell time are mainly divided into regular grid methods and explicit matrix methods. Regular grid methods utilize the Fast Fourier Transform (FFT) to improve computational speed, but require dwell points to be located on regular grid nodes, making them unsuitable for real-world processing paths such as spiral paths, random paths, non-sequential paths, or locally refined paths. While explicit matrix methods can preserve dwell points on arbitrary paths, they require establishing a removal relation matrix between each dwell point and surface sampling point. As surface resolution and the number of dwell points increase, the computational complexity of constructing, storing, and multiplying the removal matrix increases rapidly, resulting in slow dwell time calculation speeds and high memory consumption, making it difficult to meet the rapid computational requirements of high-resolution optical components. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus, device, and medium for determining the dwell time of optical processing, which can avoid explicitly removing the construction and storage of matrices while retaining the dwell time variable of any processing path, thereby reducing memory usage and computational load.

[0005] To address the aforementioned technical problems, this invention provides a method for determining the dwell time of optical processing, comprising: Obtain the target removal amount, tool influence function, and multiple path points on the processing path; Determine the mapping weights between each path point and the local grid nodes of the pre-established auxiliary rule grid; Based on the current dwell time and the mapping weight of each path point, the dwell time of each path point is allocated to the corresponding grid node of the auxiliary rule grid, and the grid dwell time obtained after allocation is convolved with the tool influence function to obtain the predicted removal amount; The residual between the target removal amount and the predicted removal amount is backpropagated from the auxiliary rule grid to each of the path points using the mapping weights to obtain the backpropagation amount of each of the path points, so as to update the dwell time of each of the path points; wherein, the convolution parameters used in the convolution operation are the same as the backpropagation parameters used in the backpropagation.

[0006] To address the aforementioned technical problems, the present invention also provides an apparatus for determining the dwell time of optical processing, comprising: The acquisition module is used to acquire the target removal amount, the tool's influence function, and multiple path points on the processing path; The mapping module is used to determine the mapping weights between each path point and the local grid nodes of the pre-established auxiliary rule grid; The forward computation module is used to allocate the dwell time of each path point to the corresponding grid node of the auxiliary rule grid according to the current dwell time of each path point and the mapping weight, and to perform a convolution operation on the grid dwell amount obtained after allocation and the tool influence function to obtain the predicted removal amount; The reverse calculation module is used to use the mapping weights to backpropagate the residual between the target removal amount and the predicted removal amount from the auxiliary rule grid to each of the path points to obtain the backpropagation amount of each of the path points, so as to update the dwell time of each of the path points; wherein, the convolution parameters used in the convolution operation are the same as the backpropagation parameters used in the backpropagation.

[0007] To address the aforementioned technical problems, the present invention also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the method for determining the dwell time of optical processing described above.

[0008] To address the aforementioned technical problems, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for determining the dwell time of optical processing described above.

[0009] The beneficial effects of this invention are as follows: The method for determining the dwell time of optical processing provided by this invention maps any path point on the obtained processing path to a pre-established auxiliary regular grid. This allows for efficient forward removal prediction using frequency domain convolution on the auxiliary regular grid, which was originally impossible to accelerate the dwell time calculation of irregular paths using fast Fourier transform. The method also employs paired backpropagation with convolution parameters consistent with the forward calculation, ensuring that the forward removal calculation and the backpropagation of the residual correspond to the same implicit removal relationship. This provides an accurate dwell time update direction for the iterative solver without constructing or storing an explicit removal matrix. Ultimately, this significantly reduces memory usage and computational load while retaining the dwell time variables of any real processing path, thereby improving the computational efficiency of solving the dwell time of optical components.

[0010] Furthermore, the present invention also provides a corresponding device, electronic device, and computer-readable storage medium for determining optical processing dwell time, which have the same or corresponding technical features as the aforementioned method for determining optical processing dwell time, and have the same effects. Attached Figure Description

[0011] To more clearly illustrate the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating the method for determining the dwell time of optical processing provided in an embodiment of the present invention; Figure 2 A schematic diagram of the structure of the device for determining the dwell time of optical processing provided in an embodiment of the present invention. Detailed Implementation

[0013] 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 of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0014] It should be noted that, in the description of this invention, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0015] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] The specific application environment architecture or specific hardware architecture on which the method for determining the dwell time of optical processing depends is described here.

[0017] This invention provides a method for determining the dwell time in optical processing. This method can be applied to computer-controlled optical surface forming equipment, including but not limited to CNC machine tools, ion beam shaping equipment, magnetorheological shaping equipment, airbag polishing equipment, and robotic polishing equipment. It can be used for deterministic polishing, ion beam shaping, magnetorheological shaping, airbag polishing, jet polishing, and other optical processing procedures. This method can be executed by a controller integrated into the aforementioned equipment or by a standalone computing device. The execution platform includes personal computers, industrial PCs, machine tool controllers, robot controllers, cloud servers, edge computing devices, embedded processors, graphics processors, field-programmable gate arrays, or multiprocessor systems. In this embodiment, the processor and memory are used to store input data, intermediate variables, mapping weights, frequency domain transform plans, and solver parameters, and to execute the calculation operations of the above steps.

[0018] The following section describes the method in detail, taking into account the execution flow of the method for determining the dwell time in optical processing. Figure 1 A flowchart of the method for determining the optical processing dwell time provided in the embodiments of the present invention is shown below. Figure 1 As shown, the method includes: S101, Obtain the target removal amount, tool influence function, and multiple path points on the processing path.

[0019] It should be noted that the target removal amount refers to the material thickness distribution that needs to be removed at various locations, determined based on the surface shape error of the surface to be processed. This target removal amount can be calculated using surface shape error data obtained from measuring equipment such as interferometers or profilometers. The target removal amount can be determined based on surface shape error, machining allowance, edge protection requirements, and actual process conditions.

[0020] The tool influence function refers to the distribution of material removed per unit time at a given location when a machining tool acts on the surface of an optical element. The tool influence function can be obtained through measured calibration, theoretical calculations, simulation results, or historical machining data. It can be an asymmetric influence function, an attitude-dependent influence function, a velocity-dependent influence function, or a composite influence function composed of multiple sub-influence functions. Furthermore, the tool influence function can be a multi-tool kernel, a zone kernel, a pressure-dependent kernel, a time-varying kernel, or a kernel updated by online measurement.

[0021] A machining path refers to the trajectory of a machining tool on a work surface. Multiple path points on the machining path are discrete sampling positions along that trajectory. Machining paths can be spiral paths, random paths, non-sequential paths, boundary-following paths, or locally refined paths. Machining paths can also be Lissajous paths, obstacle-avoidance paths, or paths generated by machine tool dynamic constraints.

[0022] In step S101, this invention provides a data foundation for dwell time calculation by acquiring the target removal amount, the tool influence function, and multiple path points on the processing path. The target removal amount reflects the material thickness distribution to be removed at various locations on the surface of the optical element; the tool influence function characterizes the material removal capability of the processing tool per unit time; and the multiple path points on the processing path are the locations where dwell time is applied. This invention directly uses discrete points on the actual processing path as dwell time variables, preserving the point distribution of any real path, such as spiral paths, random paths, non-sequential paths, boundary-following paths, and locally refined paths. This allows the dwell time calculation results to be directly applied to the actual processing trajectory without additional interpolation or approximation.

[0023] S102. Determine the mapping weights between each path point and the local grid nodes of the pre-established auxiliary rule grid.

[0024] It should be noted that the auxiliary regular mesh is a regular sampling mesh used to implement convolution operations. This mesh is independent of the original distribution of the processing path points. The size of the auxiliary regular mesh can be the same as the surface sampling mesh, or it can be expanded or divided into blocks according to the range of the tool influence function. The auxiliary regular mesh can be a two-dimensional planar mesh, a block mesh, a multi-resolution mesh, or a surface parametric domain mesh. The auxiliary regular mesh can also be a local mesh, a one-dimensional contour mesh, or a three-dimensional parametric domain mesh. For a given path point, a local mesh node refers to multiple mesh nodes in the auxiliary regular mesh that are spatially adjacent to that path point. The mapping weight refers to the proportional coefficient that each local mesh node bears when the dwell time at the path point is distributed to each local mesh node.

[0025] In step S102, this invention provides a regular data carrier for frequency domain convolution (such as Fast Fourier Transform) by establishing an auxiliary regular grid. Since path points are arbitrarily distributed, Fast Fourier Transform cannot be directly applied. However, this invention determines the mapping weights between each path point and the local grid nodes of the pre-established auxiliary regular grid, converting the dwell time on any path into the dwell time on the regular grid, thereby accelerating computation through frequency domain convolution. Unlike explicit matrix removal methods that require establishing a complete correspondence between all dwell points and all surface sampling points, this invention only needs to determine the mapping weights between each path point and a finite number of nearby local grid nodes, significantly reducing memory usage.

[0026] S103. Based on the current dwell time and mapping weight of each path point, allocate the dwell time of each path point to the corresponding grid node of the auxiliary rule grid, and perform convolution operation on the grid dwell amount obtained after allocation and the tool influence function to obtain the predicted removal amount.

[0027] It should be noted that dwell time refers to the length of time the processing tool stays at each path point. Mesh dwell time refers to the accumulated dwell time value at each mesh node after the dwell time at each path point is distributed to each mesh node of the auxiliary regular mesh according to the mapping weight. Predicted removal amount refers to the expected material thickness distribution to be removed at various locations on the surface of the optical element, calculated based on the current dwell time. The closer the predicted removal amount is to the target removal amount, the better the current dwell time is. Convolution operations are implemented through frequency domain transformation. The convolution parameters include at least the zero-padding size, the frequency domain transformation size, and the clipping size. Zero padding is used to avoid aliasing effects caused by frequency domain convolution and ensure the correctness of linear convolution; clipping is used to extract the effective part corresponding to the auxiliary regular mesh size from the convolution result. Frequency domain convolution can employ Fast Fourier Transform, block-based fast convolution, overlapping addition, overlapping preservation, real-number Fast Fourier Transform, Discrete Cosine Transform, Discrete Sine Transform, or other equivalent fast calculation methods.

[0028] In step S103, this invention transforms the problem of calculating dwell time on irregular paths into a convolution calculation problem on regular grids by mapping the dwell time of arbitrary path points to an auxiliary regular grid. From the operator perspective, the allocation operation in step S103 can be regarded as a local mapping operator S, and the convolution operation can be regarded as a convolution operator C. The two are sequentially combined to form a complete mapping from path dwell time to predicted removal amount, i.e., the implicit composite operator R=C·S; where S represents the local mapping operator from path dwell points to the auxiliary regular grid, and C represents the fast linear convolution and clipping operator defined by the tool influence function. On the auxiliary regular grid, the convolution of grid dwell amount and tool influence function can be efficiently completed through frequency domain transformation. For example, the grid dwell amount and tool influence function are respectively transformed to the frequency domain through fast Fourier transform and multiplied point by point, and then transformed back to the spatial domain through inverse transform. The explicit removal matrix method requires calculating the removal contribution between each dwell point and each surface sampling point; while this invention uses fast Fourier transform to significantly reduce the amount of computation and significantly improve the computational efficiency.

[0029] S104. Using the mapping weights, the residual between the target removal amount and the predicted removal amount is backpropagated from the auxiliary rule grid to each path point to obtain the backpropagation amount of each path point, so as to update the dwell time of each path point; wherein, the convolution parameters used in the convolution operation are the same as the backpropagation parameters used in the backpropagation.

[0030] It should be noted that the residual refers to the difference between the target removal amount and the predicted removal amount at various locations in the spatial domain. The residual can be represented by the difference between the target removal amount and the predicted removal amount, that is, the residual equals the target removal amount minus the predicted removal amount. Backpropagation refers to the process of transferring the residual within the aperture range from the surface sampling region back to the path point space. Backpropagation parameters include at least the zero-fill size, frequency domain transform size, and clipping size. The backpropagated value refers to the value obtained at each path point after the residual has been backpropagated. This backpropagated value is used to determine the update direction and magnitude of the dwell time at each path point.

[0031] In step S104, the present invention transmits the residual between the target removal amount and the predicted removal amount back to each path point through paired backpropagation operations, providing an accurate basis for updating the dwell time. From the operator perspective, the backpropagation operation in step S104 includes two steps: first, the residual is passed through the transposed convolution operator C. T The data is passed back to each grid node on the auxiliary regular grid, and then transmitted through the transpose mapping operator S. T The mesh backpropagation at the mesh nodes is converged back to each path point. Based on the implicit composite operator R=C·S defined in step S103, the paired transpose operator corresponding to the backpropagation operation is R. T =S T ·C T Among them, S TC represents the transpose mapping operator from the auxiliary regular mesh to the path dwell point. T This represents the transposed convolution operator paired with fast linear convolution and clipping. In this invention, the zero-padding size, frequency domain transform size, and clipping size used in the backpropagation are exactly the same as the corresponding parameters used in the forward convolution operation, and the mapping weights are also consistent. This ensures that the forward and backward computations correspond to the same implicit removal relation, and the update direction calculated by the iterative solver is accurate. If they are not consistent, the forward removal computation and the backward residual backpropagation will correspond to different removal models, leading to distorted gradient calculations and non-convergence of the iterative process. Through the above operations, the upper-level solver can complete the iterative update of the dwell time without accessing or saving the explicit removal matrix.

[0032] In the optical processing dwell time determination method provided in this embodiment of the invention, any path point on the obtained processing path is mapped to a pre-established auxiliary regular grid. This allows for efficient forward removal prediction using frequency domain convolution on the auxiliary regular grid, enabling the calculation of dwell time for irregular paths, which was originally impossible to accelerate using Fast Fourier Transform, to be performed. Backpropagation is performed using convolution parameters consistent with those used in the forward calculation, ensuring that the forward removal calculation and the backpropagation of the residual correspond to the same implicit removal relationship. This provides an accurate dwell time update direction for the iterative solver without constructing or storing an explicit removal matrix. Ultimately, this significantly reduces memory usage and computational load while retaining any real processing path dwell time variable, thereby improving the computational efficiency of solving the dwell time of optical components.

[0033] Furthermore, in a specific implementation, in the method for determining the dwell time of optical processing provided in the embodiments of the present invention, step S102, which determines the mapping weight between each path point and the local grid nodes of the pre-established auxiliary regular grid, may specifically include: selecting multiple local grid nodes adjacent to each path point in the pre-established auxiliary regular grid; and determining the mapping weight of each local grid node according to the relative positional relationship between each path point and the corresponding local grid node.

[0034] In practice, for any given path point, it typically doesn't lie exactly on a single grid node in the auxiliary regular grid, but rather between multiple grid nodes. This invention selects multiple adjacent local grid nodes in the auxiliary regular grid near the path point, and then assigns mapping weights based on the relative positional relationship between the path point and each local grid node. For example, when using bilinear mapping, four local grid nodes are selected around each path point, and the weights are determined by the distance between the path point and each grid node; the closer the distance, the greater the weight, and the sum of all weights is 1. When using spline mapping or Gaussian kernel mapping, the determination of mapping weights is smoother, reducing numerical errors introduced during the mapping process. The mapping method from path points to the auxiliary grid can also employ nearest neighbor mapping, bicubic mapping, compact support kernel mapping, path segment length-weighted mapping, or mapping generated from calibration data.

[0035] Since each path point only needs to record the indexes and mapping weights of a limited number of local grid nodes in its vicinity (far less than the total number of auxiliary regular grid nodes), the computational and storage requirements of this step are significantly reduced compared to the explicit matrix removal method.

[0036] Furthermore, in a specific implementation, in the optical processing dwell time determination method provided in the embodiments of the present invention, step S103, which allocates the dwell time of each path point to the corresponding grid node of the auxiliary regular grid, may specifically include: accumulating the dwell time of each path point to each local grid node according to the mapping weight of each local grid node corresponding to the path point; after traversing all path points, using the dwell time accumulated by each grid node on the auxiliary regular grid as the grid dwell amount.

[0037] In implementation, for each path point, after obtaining its dwell time, the dwell time is proportionally allocated to the corresponding local grid nodes according to the mapping weight determined in step S102. For example, the dwell time of a certain path point is... The mapping weights for the four local grid nodes are respectively , , , Then, the increments are added to these four grid nodes respectively. , , , This accumulation operation corresponds to the following calculation: ; in, Indicates the first The dwell time of each path point Indicates the first The path point is assigned to the first Mapping weights of auxiliary grid nodes, Indicates the first The number of nodes remaining on each auxiliary grid. After performing the above accumulation operation on each path point, each grid node on the auxiliary regular grid receives the accumulated contribution from different path points, and the accumulated result is the grid number remaining.

[0038] It should be noted that during the accumulation process, there may be nodes on the auxiliary regular mesh that have never been selected as local mesh nodes by any path point, and their mesh dwell time remains zero. In actual calculation, only the local mesh nodes with non-zero weights are summed, so the computational cost is related to the number of locally mapped nodes, and it is not necessary to establish a complete correspondence between all path points and all surface points. This step transforms the dwell time of arbitrarily distributed path points into a dwell time distribution on the regular mesh, enabling subsequent convolution calculations using frequency domain transformations on the regular mesh. Furthermore, since the accumulation operation only involves a few local mesh nodes for each path point, the computational cost is far less than constructing a complete removal matrix.

[0039] Furthermore, in a specific implementation, in the optical processing dwell time determination method provided in the embodiments of the present invention, step S103 performs a convolution operation on the allocated grid dwell time and the tool influence function to obtain the predicted removal amount. Specifically, it may include: transforming the allocated grid dwell time and the tool influence function to the frequency domain respectively, performing point-by-point multiplication in the frequency domain, transforming the point-by-point multiplication result back to the spatial domain to obtain the convolution result; and cropping the convolution result to the size corresponding to the auxiliary regular grid to obtain the predicted removal amount.

[0040] In implementation, the grid dwell size is represented in two-dimensional matrix form, and the tool influence function is also represented in two-dimensional matrix form. When performing convolution operations, a suitable frequency domain transform size can be determined first. This size is at least one less than the sum of the grid dwell size and the tool influence function size to ensure that aliasing does not occur during linear convolution.

[0041] Specifically, the grid dwell size is zero-padded to the predetermined frequency domain transform size, and the tool influence function is also zero-padded to the same size to ensure consistency. Then, a two-dimensional fast Fourier transform is performed on each of the two padded matrices to obtain their respective frequency domain representations. In the frequency domain, the elements at corresponding positions are multiplied point-by-point to obtain the frequency domain representation of the product. This product is then subjected to an inverse two-dimensional fast Fourier transform, transforming back to the spatial domain to obtain the complete convolution result. Finally, the middle portion corresponding to the auxiliary regular grid size is cropped from the convolution result as the prediction removal amount. This convolution calculation can be expressed as: ; in, Indicates the predicted removal amount. Indicates the influence function of the tool Auxiliary grid dwell time The result after linear convolution and cropping. This represents the aperture mask. This step utilizes the Fast Fourier Transform to achieve efficient convolution calculation, significantly reducing the computational load compared to explicit matrix multiplication, making it particularly suitable for applications involving high-resolution surfaces and large-size tool influence functions.

[0042] Furthermore, in a specific implementation, in the optical processing dwell time determination method provided in the embodiments of the present invention, after obtaining the target removal amount in step S101, it may further include: obtaining an aperture mask. Correspondingly, after performing a convolution operation on the allocated mesh dwell time and the tool influence function in step S103 to obtain the predicted removal amount, it may further include: extracting the effective removal amount within the processing aperture range from the obtained predicted removal amount according to the aperture mask; and using the effective removal amount as the predicted removal amount for comparison with the target removal amount.

[0043] In practice, during optical processing, the processing area is typically limited to the effective aperture range of the optical element. Areas outside this range require no processing or should not be included in the residual evaluation. The aperture mask is a binary matrix with the same size as the surface sampling area. Positions within the processing aperture range are assigned a value of 1, while positions outside the range are assigned a value of 0. After obtaining the convolution result, it is multiplied point-by-point with the aperture mask. This ensures that the predicted removal amount within the aperture range retains its original value, while the predicted removal amount outside the aperture range is set to zero, yielding the effective removal amount within the aperture range. This effective removal amount is then compared with the target removal amount to calculate the residual. Because data outside the aperture range is excluded, the residual calculation focuses only on the effective processing area, improving the accuracy and efficiency of the iterative solution while avoiding interference from non-physical removal amounts outside the aperture boundary.

[0044] Furthermore, in a specific implementation, in the optical processing dwell time determination method provided in the embodiments of the present invention, step S104 uses mapping weights to backpropagate the residual between the target removal amount and the predicted removal amount from the auxiliary regular grid to each path point, obtaining the backpropagation amount of each path point. Specifically, this may include: backfilling the residual between the target removal amount and the predicted removal amount into the extended grid region in the reverse order of the clipping in the convolution operation; performing a convolution operation between the backfilled residual and the spatial flipping result of the tool influence function to obtain the grid backpropagation amount of each grid node; clipping the part corresponding to the auxiliary regular grid from the grid backpropagation amount; and performing a weighted summation of the grid backpropagation amounts of each local grid node corresponding to each path point according to the mapping weights, as the backpropagation amount of each path point. The mapping weights used in this weighted summation operation are exactly the same as the mapping weights used when allocating the dwell time of the path point to the auxiliary regular grid, and the convergence direction is opposite to the allocation direction.

[0045] In implementation, the residual is a two-dimensional matrix defined over a spatial domain with the same size as the auxiliary regular grid. The residual can be expressed by the following formula: ; in, Represents the residual. Indicates the predicted removal amount. This represents the target removal amount. Since zero-padding and clipping operations were performed during the forward convolution, these operations need to be executed in reverse during backpropagation. Specifically, first, a zero matrix with a size equal to the expanded mesh size in the forward convolution is created. Then, the residuals are filled into the corresponding regions of this zero matrix in the reverse order of clipping in the forward convolution. Next, the tool influence function is flipped 180 degrees in the spatial domain (i.e., flipped along both the row and column directions). The backfilled residuals are convolved with the flipped tool influence function to obtain the convolution result at the expanded size. The portion corresponding to the auxiliary regular mesh is clipped from this convolution result to obtain the mesh backpropagation amount of each mesh node on the auxiliary regular mesh. Finally, for each path point, the mesh backpropagation amounts of each local mesh node corresponding to that path point are weighted and summed according to the mapping weights determined in step S102 to obtain the backpropagation amount of that path point. This weighted summation can be expressed as: ; in, Indicates the first Backtransfer at each path point Indicates the first The path point is assigned to the first Mapping weights of auxiliary grid nodes, Indicates the first Mesh backtransfer on each auxiliary mesh node.

[0046] The zero-fill size, frequency domain transform size, clipping size, and mapping weight used in the reverse propagation process are exactly the same as those used in the forward computation. This ensures that the forward removal computation and the reverse residual propagation correspond to the same implicit removal relationship, thereby guaranteeing the accuracy of the iterative solution.

[0047] In software implementation, forward-reverse consistency can be checked using random test vectors, and this consistency check is expressed as: ; in, This represents the result of the forward computation. This indicates the result of the reverse calculation. represents the dwell time at the waypoint, and 'e' represents the residual.

[0048] For any random vectors t and e, verify whether the inner product of the forward computation result and e is equal to the inner product of t and the backward computation result. If the absolute value of the difference between the two is less than a preset threshold, it indicates that the forward and backward operations satisfy the pairing relationship within the numerical precision. The specific value of this threshold can be determined according to the computational precision requirements and data type, and can also be dynamically adjusted according to the actual computational error level in practical applications. If the absolute value of the difference of the above inner products exceeds the preset threshold, it indicates that at least one of the following is inconsistent in the forward and backward computations: mapping weight, tool influence function center position, zero-fill size, frequency domain transform size, clipping size, aperture mask position, or numerical precision. It is necessary to re-check the settings of each parameter to ensure the pairing relationship.

[0049] Furthermore, in a specific implementation, in the optical processing dwell time determination method provided in the embodiments of the present invention, before executing step S104 to backpropagate the residual between the target removal amount and the predicted removal amount from the auxiliary regular grid to each path point using the mapping weight, it may further include: zero-filling the residual between the target removal amount and the predicted removal amount according to the zero-filling size used in the convolution operation, so that the size of the residual is consistent with the grid size after expansion in the convolution operation.

[0050] In implementation, during the forward convolution operation, the mesh dwell size is expanded to a specific frequency domain transform size using zero-padding. During the backward propagation, the residual also needs to be expanded to the same size before subsequent frequency domain convolution operations can be performed. This step zero-padding the residual with the same zero-padding size as the forward convolution, ensuring that the residual size matches the expanded mesh size. Because the zero-padding size is the same, it guarantees that the subsequent backward convolution is performed on the same spatial scale as the forward convolution, ensuring consistent spatial correspondence between the forward and backward operations.

[0051] Furthermore, when performing convolution operations on the spatial flip results of the backfilled residual and the tool influence function, the zero-fill size, frequency domain transformation size, and clipping size used are the same as those used when performing convolution operations on the mesh dwell amount and the tool influence function.

[0052] In implementation, the zero-padding size is the same, ensuring alignment of the forward and backward feeds in the expanded mesh size; the frequency domain transform size is the same, ensuring that the forward and backward feeds use the same size Fast Fourier Transform; the clipping size is the same, ensuring consistent extraction of the effective region from the convolution result. Through the consistency of these parameters, the forward and backward calculations correspond to the same implicit removal relationship, avoiding computational distortion caused by inconsistent parameters.

[0053] Furthermore, in a specific implementation, in the optical processing dwell time determination method provided in the embodiments of the present invention, step S104 updates the dwell time of each path point, which may specifically include: determining the dwell time update amount of each path point based on the back propagation of each path point and the preset iteration step size; superimposing the current dwell time of each path point with the corresponding update amount to obtain the updated dwell time; determining whether the current iteration meets the preset termination condition, if it meets the condition, outputting the dwell time of each path point, if it does not meet the condition, returning to the step of performing convolution operation on the mesh dwell amount and the tool influence function, and continuing to execute the next round of iteration with the updated dwell time as the current dwell time.

[0054] In implementation, backtransference reflects the direction and relative magnitude of the dwell time adjustment required for each path point to reduce the residual. This step determines the dwell time update for each path point based on the backtransference of each path point and a preset iteration step size. The iteration step size controls the magnitude of the dwell time adjustment in each iteration to avoid overshoot or oscillation. The method for determining the iteration step size depends on the specific solver type used. For example, in the conjugate gradient method, the iteration step size is determined through line search; in the steepest descent method, the iteration step size can be a fixed value or adjusted adaptively. The current dwell time is superimposed with the corresponding update amount to obtain the updated dwell time. Then, it is determined whether the preset termination condition is met. The termination condition can be one of several conditions or a combination thereof, such as the residual norm being less than a preset residual threshold, the number of iterations reaching a preset maximum number of iterations, or the change in dwell time being less than a preset change threshold. When the termination condition is met, the current dwell time of each path point is output as the final result; when it is not met, the process returns to step S103, using the updated dwell time as the current dwell time to continue the next iteration. This step achieves an iterative optimization closed loop for the residence time, gradually bringing it closer to the optimal solution. The threshold for the termination condition can be determined based on the processing accuracy requirements. For example, the residual threshold can be set to one percent of the maximum target removal amount, or it can be set based on the root mean square value of the surface shape error, or it can be flexibly set by the operator according to the actual process requirements.

[0055] It should be noted that this invention can be used in conjunction with various optimization algorithms, including closed-form solutions, non-negative least squares methods, conjugate gradient methods, least squares iterative methods, projective gradient methods, proximal gradient methods, and alternating direction multiplier methods. In closed-form or low-dimensional solutions, a low-dimensional dwell time base in the path space can be constructed using the transposed residual backpropagation result, and then the basis function coefficients can be determined through forward prediction calculation. In non-negative least squares solutions, the prediction calculation residual can be removed by forward propagation, and the update direction can be calculated by backpropagation using the transposed residual, while maintaining the non-negative dwell time during the update process. In conjugate gradient methods, least squares iterative methods, projective gradient methods, proximal gradient methods, and alternating direction multiplier methods, this invention can serve as a matrix calculation backend to provide the solver with forward removal and path space backpropagation quantities.

[0056] When updating the dwell time of each path point, a non-negative constraint is applied to the updated dwell time, setting negative values ​​to zero; and / or, a smoothness constraint is applied to the updated dwell time of each path point, limiting the variation in dwell time between adjacent path points. Process constraints may also include feed rate constraints, acceleration constraints, jerk constraints, total machining time constraints, path segment length constraints, and machine tool dynamic performance limitations. The output dwell time can be further converted into a feed rate table or machining control file. The output can also be a machining schedule, a path segment control table, or a closed-loop shaping control instruction.

[0057] This invention can be proven feasible through software simulation and actual processing verification. In software simulation, a two-dimensional surface error map, aperture mask, tool influence function, and a set of arbitrary path points can be input. An auxiliary regular mesh and mapping weights are established according to the method of this invention, completing the calculation of predicted removal amount and residual backpropagation without generating an explicit removal matrix. In a verification example of establishing a small-scale explicit removal matrix, the predicted removal amount obtained by this invention is compared with the result of explicit matrix multiplication, and the consistency between forward and backward calculations is checked. If the error between the two meets a preset verification threshold, it indicates that the calculation result of this invention is comparable to the calculation result of the explicit matrix. This verification threshold can be set according to the calculation accuracy requirements. Under the conditions of large-scale surface shapes and wide tool influence functions, explicit removal matrices occupy a large amount of storage space, while this invention only saves the local mapping weights and parameters required for fast convolution. The main computational load comes from local mapping, local convergence, and fast convolution, thus making it more suitable for dwell time calculation under conditions of high-resolution surface shapes and a large number of path points. In engineering applications, this invention can be embedded into existing deterministic machining software. The process may include importing measurement surface shape and process calibration data, reading or generating machining paths, optimizing dwell time, outputting dwell time or feed rate, and using it for machining control or machining simulation. This process does not require changes to the main structure of the machining equipment.

[0058] From the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by software plus necessary general-purpose hardware platforms, and of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0059] This invention also provides a device for determining the dwell time of optical processing. Figure 2 This is a schematic diagram of the structure of the device for determining the dwell time of optical processing provided in an embodiment of the present invention. This embodiment is based on the perspective of functional modules, such as... Figure 2 As shown, the device includes: The acquisition module 10 is used to acquire the target removal amount, the tool influence function, and multiple path points on the processing path; Mapping module 11 is used to determine the mapping weights between each path point and the local grid nodes of the pre-established auxiliary regular grid; The forward calculation module 12 is used to allocate the dwell time of each path point to the corresponding grid node of the auxiliary rule grid according to the current dwell time and mapping weight of each path point, and to perform convolution operation on the grid dwell amount obtained after allocation and the tool influence function to obtain the predicted removal amount. The reverse calculation module 13 is used to backpropagate the residual between the target removal amount and the predicted removal amount from the auxiliary rule grid to each path point using the mapping weights, so as to obtain the backpropagation amount of each path point and update the dwell time of each path point; wherein, the convolution parameters used in the convolution operation are the same as the backpropagation parameters used in the backpropagation.

[0060] In the optical processing dwell time determination device provided in this embodiment of the invention, the acquisition module 10 acquires the target removal amount, the tool influence function, and multiple path points on the processing path; the mapping module 11 maps any path point on the acquired processing path to a pre-established auxiliary regular grid and determines the mapping weights; the forward calculation module 12 maps the path dwell time to the auxiliary regular grid and performs convolution to obtain the predicted removal amount; and the backward calculation module 13 pairs the residuals and backpropagates them to the path points to update the dwell time. These four modules cooperate to enable the device to retain any real processing path dwell time variable while avoiding the construction and storage of an explicit removal matrix, reducing memory usage and computational load, and improving the computational efficiency of solving the dwell time of optical components.

[0061] Since the embodiments of the optical processing dwell time determination apparatus and the optical processing dwell time determination method correspond to each other, the descriptions of the features in the embodiment corresponding to the optical processing dwell time determination apparatus can be found in the relevant descriptions of the embodiments corresponding to the optical processing dwell time determination method, and will not be repeated here. Furthermore, it has the same beneficial effects as the optical processing dwell time determination method mentioned above.

[0062] Furthermore, in a specific implementation, the optical processing dwell time determination device provided in the embodiments of the present invention may further include: a matrix operator module, used to store parameters used in both forward convolution and backward propagation operations, including mapping weights, tool influence function sampling results, tool influence function center position, zero-fill size, frequency domain transformation size, and clipping size, so as to ensure that the forward calculation and backward calculation use the same set of parameters and the same set of rules.

[0063] Furthermore, in a specific implementation, the optical processing dwell time determination device provided in the above embodiments of the present invention may further include: a solver interface module, used to connect to the upper-level dwell time optimization algorithm; the upper-level algorithm calls the forward prediction removal result and the reverse propagation result through the interface to update the dwell time of each path point without accessing or saving the explicit removal matrix.

[0064] Furthermore, in specific implementations, the optical processing dwell time determination device provided in the embodiments of the present invention may further include: a processing control interface module, used to convert the output dwell time table, feed rate table or processing control file into control commands recognizable by specific processing equipment; the processing equipment includes CNC machine tools, ion beam shaping equipment, magnetorheological shaping equipment, airbag polishing equipment, and robotic polishing equipment; the processing control interface module can complete coordinate transformation, path format conversion, speed command generation, start-stop control, inter-segment connection control and processing parameter distribution according to the equipment type.

[0065] Furthermore, in a specific implementation, in the optical processing dwell time determination device provided in the embodiments of the present invention, the mapping module 11 can be specifically used to select multiple local grid nodes adjacent to each path point in the auxiliary regular grid, and determine the mapping weight of each local grid node according to the relative positional relationship between each path point and the corresponding local grid node.

[0066] Furthermore, in a specific implementation, in the optical processing dwell time determination device provided in the embodiments of the present invention, the forward calculation module 12 can be specifically used to accumulate the dwell time of each path point according to the mapping weight of each local grid node corresponding to the path point to each local grid node; after traversing all path points, the dwell time accumulated by each grid node on the auxiliary regular grid is used as the grid dwell amount. The forward calculation module 12 can also be used to transform the grid dwell amount and the tool influence function obtained after allocation to the frequency domain, perform point-by-point multiplication in the frequency domain, transform the point-by-point multiplication result back to the spatial domain to obtain the convolution result; and crop the convolution result to the size corresponding to the auxiliary regular grid to obtain the predicted removal amount.

[0067] Furthermore, in a specific implementation, in the optical processing dwell time determination device provided in the embodiments of the present invention, the reverse calculation module 13 can be specifically used to backfill the residual between the target removal amount and the predicted removal amount into the extended grid region in the reverse order of clipping in the convolution operation; perform a convolution operation on the backfilled residual and the spatial flipping result of the tool influence function to obtain the grid backtransfer of each grid node; clip out the part corresponding to the auxiliary regular grid from the grid backtransfer; and perform a weighted summation of the grid backtransfer of each local grid node corresponding to each path point according to the mapping weight as the backtransfer of each path point.

[0068] Furthermore, in a specific implementation, in the optical processing dwell time determination device provided in the embodiments of the present invention, before the reverse calculation module 13 performs the backpropagation of the residual between the target removal amount and the predicted removal amount from the auxiliary regular grid to each path point using the mapping weights, it can also be used to zero-padded the residual between the target removal amount and the predicted removal amount according to the zero-padding size used in the convolution operation, so that the size of the residual is consistent with the grid size after expansion in the convolution operation. When performing a convolution operation between the backfilled residual and the spatial flipping result of the tool influence function, the zero-padding size, frequency domain transformation size, and clipping size used are the same as those used when performing a convolution operation between the grid dwell amount and the tool influence function.

[0069] Furthermore, in a specific implementation, in the optical processing dwell time determination device provided in the embodiments of the present invention, the reverse calculation module 13 can be specifically used to determine the dwell time update amount of each path point based on the reverse transfer of each path point and the preset iteration step size; to superimpose the current dwell time of each path point with the corresponding update amount to obtain the updated dwell time; to determine whether the current iteration meets the preset termination condition, if it meets the condition, the dwell time of each path point is output, if it does not meet the condition, the forward calculation module is called to continue the next iteration with the updated dwell time as the current dwell time.

[0070] Furthermore, in a specific implementation, in the optical processing dwell time determination device provided in the embodiments of the present invention, the acquisition module 10 can also be used to acquire the aperture mask; the forward calculation module 12 can also be used to extract the effective removal amount within the processing aperture range from the obtained predicted removal amount based on the aperture mask, and use the effective removal amount as the predicted removal amount to compare with the target removal amount.

[0071] Embodiments of the present invention also provide an electronic device, including a memory and a processor. The memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above embodiments of the method for determining the dwell time of optical processing. The processor may be a central processing unit or a specific integrated circuit chip, etc. The memory may include read-only memory and random access memory, and provides instructions and data to the processor. The electronic device may be a standalone computing device or integrated into the control system of the optical processing equipment.

[0072] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program configured to execute the steps in any of the above embodiments of the method for determining the dwell time of optical processing.

[0073] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0074] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the method for determining the dwell time of optical processing.

[0075] Embodiments of the present invention also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above embodiments of the method for determining the dwell time of optical processing.

[0076] Any of the components, modules, units, parts, methods, and operations described herein can be implemented using software, firmware, hardware (e.g., fixed logic circuitry), manual processing, or any combination thereof. Alternatively or additionally, any functionality described herein can be executed at least in part by one or more hardware logic components, such as, but not limited to, a central processing unit (CPU), a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), an application-specific standard product (ASSP), a system-on-chip (SoC), a complex programmable logic device (CPLD), a microcontroller unit (MCU), etc. The terms "system," "computing device," or "apparatus" as used herein encompass various means, devices, and machines for processing data, including, for example, one or more programmable processors, computers, SoCs, or combinations thereof. The apparatus may also include code that creates an execution environment for the computer program in question, such as code constituting processor firmware, a protocol stack, a database management system, an operating system, a cross-platform runtime environment, a virtual machine, or a combination thereof. The aforementioned computer program (also known as a program, software, software application, app, script, or code) can be written in any form of programming language, including compiled or interpreted languages, declarative or procedural languages, and can be deployed in any form, including as a standalone program or as a module, component, subroutine, object, or other unit suitable for a computing environment.

[0077] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0078] The present invention has provided a detailed description of the method, apparatus, equipment, and medium for determining the dwell time of optical processing. Specific examples have been used to illustrate the principles and implementation methods of the invention. The descriptions of these embodiments are only intended to aid in understanding the method and core ideas of the invention. It should be noted that those skilled in the art can make various improvements and modifications to the invention without departing from its principles, and these improvements and modifications also fall within the scope of protection of the present invention.

Claims

1. A method for determining the dwell time during optical processing, characterized in that, include: Obtain the target removal amount, tool influence function, and multiple path points on the processing path; Determine the mapping weights between each path point and the local grid nodes of the pre-established auxiliary rule grid; Based on the current dwell time and the mapping weight of each path point, the dwell time of each path point is allocated to the corresponding grid node of the auxiliary rule grid, and the grid dwell time obtained after allocation is convolved with the tool influence function to obtain the predicted removal amount; The residual between the target removal amount and the predicted removal amount is backpropagated from the auxiliary rule grid to each of the path points using the mapping weights to obtain the backpropagation amount of each of the path points, so as to update the dwell time of each of the path points; wherein, the convolution parameters used in the convolution operation are the same as the backpropagation parameters used in the backpropagation.

2. The method for determining the dwell time of optical processing according to claim 1, characterized in that, Determining the mapping weights between each path point and the local grid nodes of the pre-established auxiliary regular grid includes: Select multiple local grid nodes adjacent to each of the path points in the pre-established auxiliary rule grid; The mapping weight of each local grid node is determined based on the relative positional relationship between each path point and its corresponding local grid node.

3. The method for determining the dwell time of optical processing according to claim 1, characterized in that, Assigning the dwell time of each path point to the corresponding grid node of the auxiliary rule grid includes: The dwell time of each path point is accumulated onto each local grid node according to the mapping weight of each local grid node corresponding to the path point; After traversing all path points, the dwell time obtained by accumulating the dwell time of each grid node on the auxiliary rule grid is used as the grid dwell time.

4. The method for determining the dwell time of optical processing according to claim 1, characterized in that, The grid dwell time obtained after allocation is convolved with the tool influence function to obtain the predicted removal amount, including: The grid dwell time obtained after allocation and the tool influence function are transformed to the frequency domain, and then multiplied point by point in the frequency domain. The result of the point-by-point multiplication is transformed back to the spatial domain to obtain the convolution result. The convolution result is cropped to the size corresponding to the auxiliary rule grid to obtain the predicted removal amount.

5. The method for determining the dwell time of optical processing according to claim 1, characterized in that, Using the mapping weights, the residual between the target removal amount and the predicted removal amount is backpropagated from the auxiliary rule grid to each of the path points to obtain the backpropagation amount for each of the path points, including: The residual between the target removal amount and the predicted removal amount is backfilled into the expanded grid region in the reverse order of the clipping in the convolution operation; The residual after backfilling is convolved with the spatial flipping result of the tool influence function to obtain the mesh backtransfer of each mesh node; The portion corresponding to the auxiliary rule grid is clipped from the grid inverse transfer; The mesh back propagation of each local mesh node corresponding to each path point is weighted and summed according to the mapping weight, and used as the back propagation of each path point.

6. The method for determining the dwell time of optical processing according to claim 1, characterized in that, Before using the mapping weights to backpropagate the residual between the target removal amount and the predicted removal amount from the auxiliary rule grid to each of the path points, the method further includes: The residual between the target removal amount and the predicted removal amount is zero-padded according to the zero-padding size used in the convolution operation, so that the size of the residual is consistent with the expanded grid size in the convolution operation.

7. The method for determining the dwell time of optical processing according to claim 1, characterized in that, Update the dwell time of each of the aforementioned path points, including: Based on the reverse data transfer of each path point and the preset iteration step size, determine the dwell time update amount of each path point; The current dwell time of each path point is added together with the corresponding update amount to obtain the updated dwell time; Determine whether the current iteration meets the preset termination condition. If it does, output the dwell time of each path point. If it does not, return to the step of performing convolution operation on the grid dwell time and the tool influence function, and continue to execute the next iteration with the updated dwell time as the current dwell time.

8. A device for determining the dwell time of optical processing, characterized in that, include: The acquisition module is used to acquire the target removal amount, the tool's influence function, and multiple path points on the processing path; The mapping module is used to determine the mapping weights between each path point and the local grid nodes of the pre-established auxiliary rule grid; The forward computation module is used to allocate the dwell time of each path point to the corresponding grid node of the auxiliary rule grid according to the current dwell time of each path point and the mapping weight, and to perform a convolution operation on the grid dwell amount obtained after allocation and the tool influence function to obtain the predicted removal amount; The reverse calculation module is used to use the mapping weights to backpropagate the residual between the target removal amount and the predicted removal amount from the auxiliary rule grid to each of the path points to obtain the backpropagation amount of each of the path points, so as to update the dwell time of each of the path points; wherein, the convolution parameters used in the convolution operation are the same as the backpropagation parameters used in the backpropagation.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the method for determining the optical processing dwell time as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method for determining the optical processing dwell time as described in any one of claims 1 to 7.

Images