Automatic typesetting method and device for cutting machine based on heuristic nested algorithm

Abstract

The invention discloses a heuristic nested algorithm-based automatic typesetting method and system for a cutting machine, and relates to the technical field of automatic typesetting. The method comprises the following steps: extracting geometric interlocking features of a part by using deep learning, and generating a feature description matrix; calculating feature matching degrees among the parts, and constructing a feature complementation degree matrix; carrying out topological mapping on the plate space based on the complementation degree, and screening candidate positions; constructing a dynamic evaluation function in combination with the distance and the interlocking coefficient; detecting a local sealing risk through time sequence analysis and dynamically adjusting a search weight; and utilizing Monte Carlo tree search to execute iteration, and outputting a typesetting scheme and a process instruction with an occlusion identifier. According to the method, the geometric features are deeply analyzed, the meshing structure is preferentially guided to be formed, the problem that special-shaped part typesetting is prone to falling into local optimum is effectively solved, and the material utilization rate and typesetting efficiency are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of automatic typesetting technology, and in particular to an automatic typesetting method and system for a cutting machine based on a heuristic nested algorithm. Background Technology

[0002] In the field of existing cutting machine technology, when automatically laying out irregularly shaped parts with deep recesses, such as C-shaped, U-shaped, or L-shaped parts, existing technologies mainly rely on heuristic algorithms based on the bottom-left corner filling strategy. When dealing with a large number of irregularly shaped parts, these algorithms tend to place two parts back-to-back, forming locally compact rectangular block structures. Although this method achieves efficient space utilization in a small area, its inherently greedy strategy leads to premature segmentation of the layout space, forming closed areas that cannot be inserted by other parts, thus significantly reducing the overall material utilization rate. More importantly, when encountering combinations of parts with strong interlocking characteristics, the algorithm is prone to falling into a geometric gravity trap, that is, the algorithm is bound by local optima and finds it difficult to break out of the current layout pattern to explore a better global layout. This topological closure problem stems from the fact that traditional methods only focus on positional compactness and ignore the geometric interlocking characteristics between parts, making it impossible for the layout process to actively identify and utilize the potential interlocking opportunities of recessed and raised structures, ultimately resulting in wasted material space and decreased layout efficiency. Therefore, existing technologies urgently need a layout mechanism that can deeply analyze the geometric features of parts and preferentially guide the formation of interlocking structures in order to overcome the limitations of simple stacking strategies.

[0003] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0004] This application aims to at least partially address one of the technical problems in the related art. To this end, one objective of this application is to propose an automatic layout method, apparatus, electronic device, and readable storage medium for a cutting machine based on a heuristic nested algorithm, thereby improving the security of the target path.

[0005] One aspect of this application provides an automatic layout method for a cropping machine based on a heuristic nested algorithm. Includes the following steps: S1. Establish a geometric feature extraction network based on deep learning to detect the depth of the recess, the direction of the opening, and the extreme points of curvature of the input C-shaped, U-shaped, or L-shaped parts, and generate a feature description matrix containing geometric interlocking feature vectors. S2. The feature description matrix is ​​used as input to calculate the feature matching degree between any two parts. By combining the opening direction matching degree and the concave-convex coupling degree, a feature complementarity matrix describing the potential bonding ability between parts is generated. S3. Based on the feature complementarity matrix, perform topological mapping on the remaining space of the current plate, use the boundary propagation algorithm to filter out the spatial regions that are compatible with the high complementarity parts, and generate a candidate position list with topological constraints. S4. For each position in the candidate position list, construct a dynamic evaluation function by combining the distance factor and the shape interlocking coefficient, calculate and output a candidate position evaluation dataset containing the current state scores of all candidate points; S5. Perform time-series trend analysis on the candidate location evaluation dataset, detect local closed topology risks through backpropagation mechanism, dynamically adjust the weight factors in the evaluation function, and generate an optimized search strategy parameter set. S6. Load the search strategy parameter set into the policy network of the Monte Carlo tree search framework, perform iterative search in the reinforcement learning environment, and output the typesetting action sequence with the highest biting priority; S7. Parse the layout action sequence, output the final layout scheme that meets the preset material utilization threshold, and generate a process instruction file with interlocking identifiers.

[0006] Preferably, the specific process of generating the feature complementarity matrix includes: S21. Extract the unit vector of the opening direction of part i from the feature description matrix. The convex direction vector of part j Calculate cosine similarity ;in, It is normalized, has a modulus of 1, and only indicates direction. The unit vector representing the direction of the convexity. This represents the direction cosine similarity between part i and part j, with a value range of [-1, 1].

[0007] S22. Extract the maximum curvature of the concave portion of part i from the feature description matrix. and the maximum curvature of the protruding part of part j Calculate the curvature coupling factor; S23. Based on the above calculation results, use the formula... Generate matrix elements and construct a feature complementarity matrix. This represents the maximum curvature value on the contour line of the recessed area of ​​part i. This represents the maximum curvature value on the contour line of the raised area of ​​part j, where This is a preset curvature tolerance parameter. : A preset constant. This represents the feature complementarity between parts i and j.

[0008] Preferably, the specific process of generating the candidate location list in step S3 includes: S31. Read the part pairs in the feature complementarity matrix whose complementarity is greater than a preset threshold; S32. Discretize the current sheet metal into a grid map, and propagate directionally along the concave direction only from the free boundary grid corresponding to the high complementarity parts. The propagation distance d and the width of the part opening w satisfy d=λw; d represents the propagation distance in the boundary propagation algorithm, w represents the physical width at the part opening, and λ represents the propagation coefficient.

[0009] S33. Record the nodes that satisfy the spatial constraints in the propagation path and generate a list of candidate locations with topological constraints that include adjacency relationships.

[0010] Preferably, the specific process of generating the candidate location evaluation dataset in step S4 includes: S41. Iterate through each position k in the candidate position list; S42. Calculate the distance Dk from position k to the plate boundary, and calculate the shape interlocking coefficient based on the feature complementarity of adjacent parts around position k. ; Values ​​range from [0, 1].

[0011] S43. Using the formula Calculate the comprehensive score for the current candidate position, and obtain the F-values ​​and their corresponding state components for all positions. Combine the results to output a candidate location evaluation dataset. Indicates the dynamic decay factor. To control the weights for complementary features; This indicates the weight for controlling placement near the edge; This indicates the weight that controls the local interlocking.

[0012] Preferred, the specific process of generating the search strategy parameter set in step S5 includes: S51. Establish a historical decision queue and continuously record the score changes in the candidate location evaluation dataset; S52. When a candidate location is detected, evaluate the score variance of the dataset for M consecutive iterations. And the average complementarity is below the threshold. When the condition is met, it is determined that the algorithm has entered a geometric gravity trap; M is an integer used to determine whether the algorithm has stalled. This is the average threshold for complementarity; when the historical average score is lower than this value, the typesetting quality is considered poor. S53. Update the weights based on the judgment result. ,in The weight is increased to the original value. The updated search strategy parameter set is then output.

[0013] Preferably, the specific process of generating the typesetting action sequence in step S6 includes: S61. Receive the search strategy parameter set as state input and predict the priority probability of each node during the expansion phase of the Monte Carlo tree; S62. During the simulation phase, the reward function is calculated based on the weight configuration in the parameter set. ; Based on the current utilization rate of the board material, This is the utilization rate weighting coefficient.

[0014] S63. When the tree search reaches the preset depth L or time threshold, extract the node sequence on the optimal path and generate the typesetting action sequence.

[0015] Preferably, the processing of the typesetting action sequence in step S7 includes: S71. Perform topological restoration on the typesetting action sequence and identify all part groups that form interlocking structures in the sequence; S72. Assign a unique identification code to each interlocking structure, indicating the required machining accuracy; S73. The final output is a process instruction file containing geometric coordinates and engagement symbols.

[0016] This invention also provides an automatic layout system for a cutting machine based on a heuristic nested algorithm, comprising: The feature description matrix generation module is used to build a geometric feature extraction network based on deep learning, and to detect the depth of the recess, the direction of the opening, and the extreme point of curvature of the input C-shaped, U-shaped or L-shaped parts, and generate a feature description matrix containing geometric interlocking feature vectors. The feature complementarity matrix generation module is used to take the feature description matrix as input, calculate the feature matching degree between any two parts, and generate a feature complementarity matrix describing the potential bonding ability between parts by combining the opening direction matching degree and the concave-convex coupling degree. The candidate location list generation module is used to perform topological mapping on the remaining space of the current sheet metal based on the feature complementarity matrix, and use the boundary propagation algorithm to filter out the spatial regions that are compatible with the high complementarity parts, and generate a candidate location list with topological constraints. The candidate location evaluation dataset generation module is used to construct a dynamic evaluation function for each location in the candidate location list by combining the distance factor and the shape interlocking coefficient, and to calculate and output a candidate location evaluation dataset containing the current state scores of all candidate points. The search strategy parameter set generation module is used to perform time-series trend analysis on the candidate location evaluation dataset, detect local closed topology risks through backpropagation mechanism, dynamically adjust the weight factors in the evaluation function, and generate an optimized search strategy parameter set. The typesetting action sequence generation module is used to load the search strategy parameter set into the policy network of the Monte Carlo tree search framework, perform iterative search in the reinforcement learning environment, and output the typesetting action sequence with the highest biting priority. The process instruction output module is used to parse the layout action sequence, output the final layout scheme that meets the preset material utilization threshold, and generate a process instruction file with interlocking identifiers.

[0017] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the steps in the described automatic layout method for a cutting machine based on a heuristic nested algorithm.

[0018] The present invention also provides a readable storage medium storing a computer program adapted for loading by a processor to perform the steps of the automatic layout method for a cutting machine based on a heuristic nested algorithm as described in any one of claims 1-7.

[0019] This application effectively overcomes the "geometric gravity trap" of traditional typesetting algorithms by constructing a feature complementarity matrix and an adaptive Monte Carlo tree search mechanism, significantly improving the material utilization rate of irregularly shaped parts. Existing technologies often employ greedy strategies, which can easily lead to premature spatial segmentation. This application first utilizes deep learning to extract geometric interlocking features such as concavity and curvature of parts, and calculates the feature complementarity matrix, fundamentally quantifying the "interlocking" potential between parts. Then, combining temporal trend analysis and dynamic weight adjustment, when the typesetting is detected to be trapped in local closure (i.e., low score variance and poor complementarity), the algorithm is forced to escape the current local optimum by increasing the complementarity weight. This mechanism can proactively guide irregularly shaped parts such as C-shaped and U-shaped parts to form a tight interlocking structure, rather than simply stacking them back-to-back, thereby maximizing the effective utilization area of ​​the sheet metal globally and significantly reducing the scrap rate.

[0020] This application introduces a boundary propagation-based topology mapping and multi-dimensional dynamic evaluation system, achieving dual optimization of layout computation efficiency and processing safety. Unlike blind full-map search, this application utilizes a boundary propagation algorithm to filter out topological regions suitable for highly complementary parts, generating constrained candidate positions. This achieves intelligent pruning of the search space, significantly reducing invalid calculations and improving layout speed. Simultaneously, by generating process instruction files with "interlocking identifiers," the system can clearly distinguish between ordinary cutting paths and high-precision interlocking paths. This not only ensures the feasibility of the tight layout scheme on actual cutting machines but also effectively avoids the risk of cutting interference or thermal deformation caused by excessively close part spacing, ensuring the final product's processing quality. Attached Figure Description

[0021] Figure 1 This is a schematic flowchart of an automatic layout method for a cutting machine based on a heuristic nested algorithm according to an exemplary embodiment of this application; Figure 2 This is a block diagram of an automatic layout system for a cutting machine based on a heuristic nested algorithm, according to an exemplary embodiment of this application. Figure 3 This is a schematic diagram of an electronic device according to an exemplary embodiment of this application; Figure 4 This is a schematic diagram of a readable storage medium according to an exemplary embodiment of this application. Detailed Implementation

[0022] To better understand this application, various aspects of this application will be described in more detail with reference to the accompanying drawings. It should be understood that these detailed descriptions are merely illustrative of exemplary embodiments of this application and are not intended to limit the scope of this application in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.

[0023] In the accompanying drawings, the size, dimensions, and shapes of the elements have been slightly adjusted for ease of illustration. The drawings are for illustrative purposes only and are not strictly to scale. As used herein, the terms “approximately,” “about,” and similar terms are used to indicate approximation, not degree, and are intended to illustrate inherent deviations in measured or calculated values ​​that will be recognized by one of ordinary skill in the art. Furthermore, the order in which the steps are described in this application does not necessarily indicate the order in which these steps occur in actual operation, unless otherwise expressly defined or deduced from the context.

[0024] It should also be understood that expressions such as "comprising," "including," "having," "containing," and / or "comprising" are open-ended rather than closed-ended expressions in this specification, indicating the presence of the stated features, elements, and / or components, but not excluding the presence of one or more other features, elements, components, and / or combinations thereof. Furthermore, when expressions such as "at least one of..." appear after a list of listed features, they modify the entire list of features, not just individual elements in the list. Additionally, when describing embodiments of this application, the word "may" is used to mean "one or more embodiments of this application." And the term "exemplary" is intended to refer to examples or illustrations.

[0025] Unless otherwise specified, all terms used herein (including engineering and technical terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that, unless expressly stated herein, terms defined in common dictionaries shall be interpreted as having the meaning consistent with their meaning in the context of the relevant art, and not as having an idealized or overly formalized meaning.

[0026] Example 1 It should be noted that, where there is no conflict, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] When traditional heuristic nested algorithms process C-shaped, U-shaped, or L-shaped parts with deep recesses, the greedy nature of the algorithm, based on the bottom-left corner filling strategy, causes two parts with strong interlocking features to be stacked back-to-back, forming a locally compact but topologically closed spatial structure. This geometric gravity trap leads the algorithm to focus only on positional compactness, failing to quantify the complementary features between parts, making it impossible for a third part to be inserted into the gaps, thus hindering overall material utilization. The essence of this problem lies in the evaluation operator's lack of comprehensive consideration of recess depth, opening direction, and curvature extrema, thus failing to identify potential interlocking structures.

[0028] For example, in a metal sheet cutting production scenario, when processing multiple C-shaped parts (with the same opening direction) and L-shaped parts, the heuristic algorithm places two C-shaped parts back-to-back, forming a rectangular closed area. The internal space of this area is completely blocked, preventing the protruding parts of the L-shaped parts from physically engaging with the recessed parts of the C-shaped parts. Furthermore, the remaining space of the sheet metal is divided into isolated gaps, forcing the layout process to adopt a suboptimal layout strategy, leading to increased raw material waste and a lack of adaptability in the layout scheme.

[0029] If the aforementioned problems are not addressed, the typesetting algorithm will remain confined to local optima, unable to overcome the cumulative effects of topologically closed spaces. Consequently, material utilization will remain low, leading to increased production costs. Furthermore, the algorithm's lack of feature complementarity will limit its applicability in complex part assemblies, impacting the stability of the overall production process and resource utilization efficiency.

[0030] Therefore, such as Figure 1 As shown, an automatic layout method for a cutting machine based on a heuristic nested algorithm is provided, including the following steps: S1. Establish a geometric feature extraction network based on deep learning to detect the depth of the recess, the direction of the opening, and the extreme points of curvature of the input C-shaped, U-shaped, or L-shaped parts, and generate a feature description matrix containing geometric interlocking feature vectors. S2. Using the feature description matrix as input, calculate the feature matching degree between any two parts. By combining the opening direction matching degree and the concave-convex coupling degree, generate a feature complementarity matrix describing the potential bonding ability between parts. S3. Based on the feature complementarity matrix, perform topological mapping on the remaining space of the current sheet metal, use the boundary propagation algorithm to filter out the spatial regions that are compatible with the high complementarity parts, and generate a list of candidate locations with topological constraints. S4. For each position in the candidate position list, construct a dynamic evaluation function by combining the distance factor and the shape interlocking coefficient, calculate and output a candidate position evaluation dataset containing the current state scores of all candidate points; S5. Perform time-series trend analysis on the candidate location evaluation dataset, detect local closed topological risks through backpropagation mechanism, dynamically adjust the weight factors in the evaluation function, and generate an optimized search strategy parameter set. S6. Load the search policy parameter set into the policy network of the Monte Carlo tree search framework, perform iterative search in the reinforcement learning environment, and output the layout action sequence with the highest biting priority; S7. Parse the layout action sequence, output the final layout scheme that meets the preset material utilization threshold, and generate a process instruction file with interlocking identifiers.

[0031] This embodiment provides a technical solution to the geometric attraction trap problem existing in the layout of C-shaped, U-shaped, or L-shaped parts with strong interlocking characteristics. The geometric feature extraction network refers to a deep learning model used to identify key geometric attributes from the original part data. It can use convolutional neural networks to encode features in the part contour image, such as extracting multi-scale concave region information through a feature pyramid network, or using an attention mechanism to focus on key nodes in the opening direction. Its main purpose is to achieve automated detection and vectorized representation of the geometric interlocking features of the parts. Furthermore, the feature complementarity matrix refers to a data structure that quantifies the potential bonding ability between any two parts. It can use shape matching-based similarity calculation methods, such as evaluating the geometric fit of the edges of two parts through contour matching algorithms, or calculating the spatial alignment error of concave and convex regions using point cloud registration technology. Its main purpose is to achieve an objective measurement of the interlocking probability between parts. Specifically, the dynamic evaluation function refers to a mathematical model used to comprehensively evaluate the merits of candidate positions. It can employ a multi-objective optimization-based scoring mechanism, such as calculating interlocking strength based on the contact boundary length between the candidate position and adjacent parts, or determining position priority through a weighted combination of space filling rate and boundary distance. Its main purpose is to achieve multi-dimensional performance evaluation of layout position selection. Furthermore, local closed topology risk detection refers to a technical means of identifying when the layout process falls into a local optimum. It can employ time-series data analysis methods, such as monitoring the entropy change trend of the remaining space distribution in continuous iterations, or analyzing the periodic fluctuation characteristics of the candidate position scoring sequence. Its main purpose is to achieve dynamic early warning of geometric gravity traps. As a preferred implementation, the Monte Carlo tree search framework refers to an algorithmic structure for optimizing layout decisions. It can employ a node expansion strategy based on simulated annealing, such as introducing random perturbations during the search process to avoid local convergence, or dynamically adjusting the exploration depth of the simulation phase using historical decision data. Its main purpose is to achieve efficient generation of high-interlocking-priority action sequences. Therefore, this application transforms the geometric interlocking features of parts into quantifiable evaluation criteria, forcing the algorithm to prioritize exploring interlocking structures rather than simple stacking. This avoids the topological closure problem caused by local compact stacking in traditional heuristic algorithms, ultimately achieving the technical effects of improved material utilization and optimized typesetting quality.

[0032] This method achieves efficient automatic layout of C-shaped, U-shaped, or L-shaped parts by constructing a feature complementarity-driven layout mechanism. First, a deep learning-based geometric feature extraction network detects the indentation depth, opening direction, and curvature extrema of the input parts, generating a feature description matrix containing geometrically interlocking feature vectors. This matrix provides accurate geometric basis for subsequent interlocking judgment. Further, using the feature description matrix as input, the feature matching degree between any two parts is calculated. A feature complementarity matrix is ​​generated by combining the opening direction matching degree and the indentation-protrusion coupling degree. This matrix quantifies the potential bonding ability between parts. Based on this, the feature complementarity matrix is ​​used to perform topological mapping on the remaining space of the current sheet metal. A boundary propagation algorithm is used to filter out spatial regions suitable for highly complementary parts, thus generating a candidate position list with topological constraints. For each position in the candidate position list, a dynamic evaluation function is constructed by combining a distance factor and a shape interlocking coefficient. A candidate position evaluation dataset containing the current state scores of all candidate points is calculated and output. Furthermore, a time-series trend analysis is performed on the candidate location evaluation dataset. Locally closed topological risks are detected through backpropagation, and the weighting factors in the evaluation function are dynamically adjusted to generate an optimized search strategy parameter set. Subsequently, the search strategy parameter set is loaded into the policy network of a Monte Carlo tree search framework, and iterative search is performed in a reinforcement learning environment to output the layout action sequence with the highest biting priority. Finally, the layout action sequence is parsed, outputting the final layout scheme that meets the preset material utilization threshold, and generating a process instruction file with biting identifiers.

[0033] In practical implementation, when processing parts with C-shaped openings facing left, the geometric feature extraction network can specifically employ a convolutional neural network architecture to detect the maximum curvature value of its concave region and the unit vector of the opening direction, while simultaneously identifying the convex features of U-shaped parts. Based on this, the matching degree of the opening direction between the two parts and the degree of concavity-convex coupling are calculated, generating corresponding feature complementarity evaluation values. For example, in the candidate position screening stage, the boundary propagation algorithm starts from the free boundary grid and propagates directionally along the concavity direction, with the propagation range dynamically adjusted according to the opening width of the part. In the evaluation stage, the dynamic evaluation function comprehensively considers the distance from the candidate position to the plate boundary, the shape interlocking coefficient of surrounding parts, and feature complementarity, forming a multi-dimensional scoring system. Thus, by monitoring score changes in real time through time-series trend analysis, when it is detected that the variance of continuous iteration scores is too small and the average complementarity is too low, the weight priority of feature complementarity is automatically increased, guiding the search process to break out of local stacking loops.

[0034] Therefore, this method effectively avoids the topological closure problem caused by local compact stacking in traditional heuristic algorithms. By transforming geometric interlocking features into quantifiable evaluation criteria, it forces the priority exploration of interlocking structures rather than simple stacking. Specifically, the construction of the feature complementarity matrix enables the algorithm to accurately identify the potential bonding capabilities between parts, the topological mapping of the boundary propagation algorithm ensures that high complementarity regions are prioritized for detection, the multi-dimensional coupling of the dynamic evaluation function breaks through the limitation of only focusing on spatial compactness, and the temporal trend analysis and dynamic weight adjustment mechanism actively identify and avoid geometric gravity traps. Finally, the generation and parsing of the typesetting action sequence improves material utilization while ensuring that the process instruction file clearly indicates the interlocking structure, fundamentally solving the technical problem of heuristic algorithms falling into geometric gravity traps when dealing with strongly interlocking feature parts.

[0035] In some of the embodiments described above in this application, a feature complementarity matrix is ​​proposed to quantify the potential bonding capability between parts. However, in its implementation, the feature complementarity calculation does not fully consider the sensitivity control of curvature differences, resulting in a disconnect between the orientation matching degree and the shape interlocking capability. When evaluating the interlocking possibility of parts, the algorithm cannot distinguish between cases where the orientation is consistent but the curvature is incompatible, thus erroneously overestimating the complementarity. This causes the layout decision to fall into the geometric gravity trap of local optima, resulting in the ineffective utilization of the plate gaps and difficulty in improving the material utilization rate.

[0036] In this regard, this application further proposes a specific process for generating the feature complementarity matrix, including: S21. Extract the unit vector of the opening direction of part i from the feature description matrix. The convex direction vector of part j Calculate cosine similarity ;in, It is normalized, with a modulus of 1, representing direction. The unit vector representing the direction of the convexity. This represents the direction cosine similarity between part i and part j, with a value range of [-1, 1].

[0037] S22. Extract the maximum curvature of the concave portion of part i from the feature description matrix. and the maximum curvature of the protruding part of part j Calculate the curvature coupling factor; S23. Based on the above calculation results, use the formula... Generate matrix elements and construct a feature complementarity matrix, where This is the preset curvature tolerance parameter. This represents the maximum curvature value on the contour line of the recessed area of ​​part i. Physically, it describes the place where the curve is "the sharpest" in the recess. This represents the maximum curvature value on the contour line of the raised area of ​​part j. Curvature tolerance parameter: a preset constant used to control sensitivity to curvature differences. The larger the value, the higher the tolerance for shape mismatch. This represents the feature complementarity between parts i and j.

[0038] In practical applications, the unit vector of the opening direction This refers to the normalized direction vector of the opening orientation of a part, determined by a geometric feature extraction network. It can be implemented using principal component analysis or contour principal axis fitting algorithms, aiming to accurately characterize the opening orientation and avoid orientation misjudgment caused by geometric position dependence; (The text also mentions a protrusion direction vector.) This refers to the unit vector extending along the contour of the protruding part of a component. It can be obtained by fitting the convex hull boundary or curvature extrema, aiming to provide a reliable geometric basis for direction matching. Among these, the maximum curvature of the recessed part... This refers to the quantitative index of the peak absolute value of curvature on the contour line of a part. It can be calculated using curvature estimation from discrete point clouds or differential geometric operators. Its purpose is to capture the point of most dramatic geometric abrupt change in contour curvature to reflect the actual interlocking capability; the maximum curvature of the protruding part... This refers to the extreme value of the curvature of the corresponding convex region, which can be determined based on the second derivative or polynomial fitting method of the contour sampling points. Its purpose is to establish a matching evaluation benchmark with the concave curvature. Specifically, the curvature tolerance parameter... This refers to the preset threshold for adjusting the sensitivity to curvature differences. It can be set as an empirical constant or obtained through training and optimization using historical layout data. The purpose is to dynamically adjust the tolerance of the exponential decay function to shape mismatch, ensuring that the complementarity calculation is consistent with the actual interlocking feasibility.

[0039] Specifically, the solution in this application first extracts the unit vector of the opening direction from the feature description matrix. with the convex direction vector Cosine similarity is calculated to ensure accurate representation of orientation matching; then the maximum curvature of the concave portion is extracted. With the maximum curvature of the protruding part The key physical features for quantifying the curvature of the contour are identified; finally, the directional similarity and curvature difference are combined using an exponential decay function to generate feature complementarity. Among them, the curvature tolerance parameter Dynamically control the decay rate. This process forms a cooperative quantization mechanism that is both direction-matched and curvature-compatible: when the curvature difference is within... When within the tolerance range, Maintain a high level to support interlocked decision-making; when discrepancies exceed tolerance. Rapid decay suppresses the chance of false interlocking. This design forces the algorithm to prioritize the identification of part pairs with highly complementary geometric features, effectively filtering invalid pairs that match in orientation but have incompatible curvature. This provides an accurate basis for interlocking decisions in subsequent topology mapping, avoiding the geometric gravity trap in the typesetting process.

[0040] As a specific implementation method, the solution of this application is implemented as follows: Considering the pairing scenario of C-shaped parts and U-shaped parts, the unit vector of the opening direction... The convexity direction vector is determined by the direction of the principal axis of the contour. Obtained by contour fitting of the U-shaped convex portion; maximum curvature of the concave portion. and the maximum curvature of the protruding part Curvature tolerance parameters are extracted using the curvature calculation module of the contour point cloud. Set to a moderate constant value to balance sensitivity. The calculated feature complementarity... It can accurately distinguish between cases where the orientation matches but the curvature is incompatible, such as when the curvature of a C-shaped concave shape differs significantly from that of a U-shaped convex shape. The value is significantly reduced, thus preventing the algorithm from incorrectly selecting the pairing scheme.

[0041] Through the above scheme, this application can truly quantify the feasibility of shape interlocking between parts, avoid the disconnect between direction matching degree and curvature compatibility in feature complementarity calculation, thereby effectively filtering out false interlocking opportunities, guiding layout decisions to avoid geometric gravity traps, and improving the utilization rate of plate gaps.

[0042] In some of the embodiments described above in this application, a topological mapping based on the feature complementarity matrix is ​​proposed to generate a candidate location list. However, in its implementation, the general boundary propagation algorithm fails to focus specifically on the free boundary corresponding to the high feature complementarity parts, resulting in a generalized search range and easy to fall into the geometric gravity trap. It cannot effectively identify potential locations that can form interlocking engagement, thus continuing the back-to-back placement mode and hindering the improvement of material utilization.

[0043] In this regard, this application further proposes a specific process for generating the candidate location list, including: S31. Read the part pairs in the feature complementarity matrix whose complementarity is greater than a preset threshold; S32. Discretize the current sheet metal into a grid map, and propagate directionally along the concave direction only from the free boundary grid corresponding to the high complementarity parts. The propagation distance d and the width of the part opening w satisfy d=λw; d represents the propagation distance in the boundary propagation algorithm, w represents the physical width at the part opening, and λ represents the propagation coefficient, which takes the value [0.5, 1.2] and is used to dynamically adjust the search range according to the size of the part opening.

[0044] S33. Record the nodes that satisfy the spatial constraints in the propagation path and generate a list of candidate locations with topological constraints that include adjacency relationships.

[0045] Among them, part pairs with a complementarity greater than a preset threshold refer to those in the feature complementarity matrix. For component combinations whose values ​​exceed a set threshold, a fixed threshold or a dynamic threshold mechanism based on historical data can be used. The purpose is to filter component pairs with high interlocking potential and avoid invalid searches in low-complementarity areas. A grid map refers to a spatial representation method that discretizes the sheet surface into a regular grid. It can use square or hexagonal grids to simplify spatial calculations and boundary representation. Free boundary grids refer to free grid points on the edge of the sheet or the boundary of already arranged components in the current layout state. They can be identified based on geometric scanning algorithms, aiming to provide effective search starting points. The recess direction refers to the direction in which the opening of a C-shaped, U-shaped, or L-shaped component faces. It can be determined based on the direction vector in the feature description matrix, aiming to guide the propagation direction to match the component's geometric features. The relationship between propagation distance d and opening width w, d=λw, refers to the ratio of propagation distance to the component opening width in the boundary propagation algorithm. An adaptive adjustment strategy can be used to dynamically adjust the search range according to the component size, ensuring that the search range adapts to changes in component size. The propagation coefficient λ is a parameter that controls the ratio of propagation distance to opening width. It can take values ​​within the range [0.5, 1.2]. Its purpose is to balance the matching degree between the search range and the part size, avoiding redundancy due to excessive value or omission of feasible positions due to insufficient value. Nodes satisfying spatial constraints are positions in the propagation path that can accommodate the part without overlapping with existing parts. These can be verified using collision detection algorithms to ensure the physical feasibility of candidate positions. The candidate position list with topological constraints is a set of candidate positions containing their coordinates and adjacency relationships with surrounding parts. It records the connection information between positions, providing structured input for subsequent evaluation functions and supporting accurate assessment of interlocking capabilities.

[0046] Specifically, the proposed solution first filters part pairs with complementarity greater than a preset threshold from the feature complementarity matrix in step S31. This step focuses on part combinations with high interlocking potential based on quantification results, providing a precise initial basis for subsequent propagation. Building on this, step S32 discretizes the current sheet metal into a grid map and propagates directionally only from the free boundary grid corresponding to the high-complementarity parts along the concave direction. Simultaneously, the propagation distance is dynamically adjusted based on the opening width of the parts. By constraining the propagation start point and direction, the search process strictly matches the concave geometric features of the parts. Finally, step S33 records the nodes in the propagation path that satisfy the spatial constraints and generates a candidate position list containing adjacency relationships. While verifying spatial availability, it retains topological adjacency information, ensuring that candidate positions not only meet physical placement conditions but also implicitly contain the possibility of forming an interlocking structure. Overall, this solution effectively guides the algorithm to prioritize exploring spatial regions that can form an interlocking structure by tightly anchoring the search process to the geometric characteristics of high-feature-complementarity parts, thereby breaking through the constraints of locally closed topologies.

[0047] As a specific implementation method, the solution of this application is implemented as follows: When the system identifies a C-shaped part and a protruding part with high feature complementarity, it starts from the free grid point of the opening boundary of the C-shaped part and propagates along the concave direction. The propagation distance is dynamically determined according to the opening width of the C-shaped part. For example, when the opening is wider, the propagation distance increases accordingly, thereby efficiently locating the concave space that can accommodate the protruding part and generating a candidate position list containing the position and its adjacency relationship with surrounding parts.

[0048] Through the above technical solution, this application effectively solves the problem of generalization of the search range, can accurately identify potential locations where interlocking can be formed, avoids the back-to-back placement mode, and thus improves the material utilization rate.

[0049] Specifically, in some of the solutions mentioned above in this application, a method for generating a list of candidate positions with topological constraints is proposed. However, in this process, the evaluation of candidate positions only depends on spatial constraints and boundary propagation results, and lacks a quantitative evaluation mechanism for the geometric interlocking features between parts. This leads to the algorithm over-prioritizing placement near the edge or local compactness when selecting positions, while ignoring feature complementarity and shape interlocking potential, thereby triggering geometric gravity traps, causing the layout to fall into a topologically closed state, and making it difficult to break through the bottleneck of material utilization.

[0050] In this regard, this application further proposes the specific process for generating the candidate location evaluation dataset in step S4, which includes: S41. Iterate through each position k in the candidate position list; S42. Calculate the distance Dk from position k to the plate boundary, and calculate the shape interlocking coefficient based on the feature complementarity of adjacent parts around position k. Dk represents the Euclidean distance from the geometric center of the current candidate position k to the nearest boundary of the plate. The shape interlock coefficient represents the probability that the position will form a tight interlock with the surrounding parts, and its value is [0,1].

[0051] S43. Using the formula Calculate the overall score for the current candidate position; a higher score indicates a better position. Then, calculate the F-values ​​and corresponding state components for all positions. Combine and output the candidate position evaluation dataset. α, β, and γ represent dynamic decay factors, where α controls the importance of "feature complementarity"; β controls the importance of "edge placement" (using 1 / Dk); and γ controls the importance of "local interlocking." represents the number of adjacent parts around the current candidate position.

[0052] Wherein, position k refers to a spatial node in the candidate position list, which can be a cell in the discretized grid of the plate material, used to represent the geometric region where the part may be placed; Dk refers to the Euclidean distance from the geometric center of the current candidate position k to the nearest boundary of the plate material, which can be calculated by the coordinate difference between position k and the boundary point in the coordinate system; α refers to the shape interlocking coefficient, which can be achieved by weighting the feature complementarity values ​​of adjacent parts in the feature complementarity matrix or by selecting the maximum value, in order to quantify the possibility of position k forming an interlocking structure with surrounding parts; F refers to the comprehensive score, which can be calculated by linearly combining the feature complementarity, the inverse distance, and the shape interlocking coefficient; α, β, and γ refer to the dynamic attenuation factors, which can be preset constants used to adjust the relative importance of different evaluation dimensions.

[0053] Specifically, the solution in this application ensures that all potential spatial nodes are systematically evaluated by traversing each position k in the candidate position list, avoiding the omission of high interlocking potential areas in local searches; calculating the distance Dk from position k to the plate boundary provides an objective measure of edge tendency, but the key lies in calculating the shape interlocking coefficient Sl based on the feature complementarity of adjacent parts around position k. This coefficient directly utilizes the curvature matching and opening direction coordination information between parts in the feature complementarity matrix to transform the concave-convex coupling capability into a quantifiable interlocking probability; using the formula Feature complementarity when calculating the overall score Emphasizing the degree of matching of geometric interlocking features between parts, distance factor Under the control of weight β, the requirements for placing the object near the edge are appropriately balanced to avoid wasting space. The shape interlocking coefficient Sl is enhanced under the guidance of weight γ to strengthen the probability of forming local interlocking structures. The final output candidate position evaluation dataset provides a high-precision decision basis for subsequent search strategies, enabling the algorithm to prioritize positions that can form geometric interlocking, thereby effectively avoiding the risk of topological closure.

[0054] As a preferred embodiment, the solution of this application is specifically implemented as follows: Multiple candidate positions are determined in the discretized grid of the board material. For position k, the Euclidean distance from its geometric center to the boundary of the board material is obtained through coordinate calculation. Based on the feature complementarity matrix, the feature complementarity values ​​of adjacent parts around position k are extracted, and the shape interlocking coefficient Sl is calculated using the weighted average method; Substitute into the formula Calculate the overall score, where α, β, and γ are configured with preset constants; combine the scores of all positions and their corresponding state components to generate a candidate position evaluation dataset, which is used to guide subsequent layout decisions.

[0055] Through the above scheme, this application can perform multi-dimensional quantitative evaluation of candidate positions, comprehensively consider feature complementarity, boundary distance and shape interlocking coefficient, avoid the algorithm from relying too much on edge placement or local compactness, thereby effectively avoiding geometric gravity traps, enabling the typesetting process to break through the topological closure state and achieve continuous optimization of material utilization.

[0056] Specifically, in the above method, a candidate position scoring mechanism based on a dynamic evaluation function is proposed to optimize the layout search process. However, in its implementation, when the layout algorithm falls into the geometric gravity trap (i.e., the variance of continuous iteration scoring is too small and the average complementarity is consistently low), the weight factor in the evaluation function fails to adjust adaptively, causing the algorithm to be unable to effectively identify and escape the local optimum, and the material utilization rate stagnates.

[0057] In this regard, this application further proposes the specific process of generating the search strategy parameter set in step S5, which includes: S51. Establish a historical decision queue and continuously record the score changes in the candidate location evaluation dataset; S52. When a candidate location is detected, evaluate the score variance of the dataset for M consecutive iterations. And the average complementarity is below the threshold. When the condition is met, it is determined that the algorithm has entered a geometric gravity trap; M is an integer used to determine whether the algorithm has stalled. This is the mean threshold for complementarity. When the historical average score is below this value, the typesetting quality is considered poor. The variance is calculated as the variance of the values ​​in the historical rating queue Q. A small variance indicates that the algorithm is in an "infinite loop" and is not exploring new solutions.

[0058] S53. Update the weights α, β, and γ based on the judgment result, where the weight of α is increased to 1+lnM times the original value, and output the updated search strategy parameter set.

[0059] The historical decision queue refers to a time-series data structure used to store historical scoring data. It can be implemented using a first-in, first-out queue or a circular buffer, aiming to accumulate iterative scoring changes to support trend analysis. Scoring variance refers to the statistical dispersion of values ​​in the historical scoring queue, which can be obtained by calculating the squared standard deviation of scores within the queue. Its purpose is to quantify the volatility of the algorithm's search process and avoid misjudgments caused by fluctuations in a single score. Average complementarity refers to the arithmetic mean of feature complementarity in the historical scoring queue, which can be based on the sum of the scores from each iteration within the queue. The calculated value aims to directly reflect the geometric feature matching level of the typesetting quality; the geometric gravity trap refers to the state in which the algorithm gets stuck in a local optimum and cannot adaptively adjust. Its judgment is based on the dual threshold conditions of the scoring variance and the average complementarity, and its purpose is to accurately distinguish between normal search fluctuations and substantial stagnation; the weight update mechanism refers to the process of dynamically adjusting the α, β, and γ parameters in the evaluation function. It can be implemented by non-linear scaling, and its purpose is to adaptively enhance the importance of feature complementarity according to the depth of the trap.

[0060] Specifically, this application establishes a historical decision queue to continuously record the score changes in the candidate position evaluation dataset, providing a continuous observation basis for time series trend analysis. When the score variance of M consecutive iterations is less than σvar2 and the average complementarity is lower than the threshold θ, a dual judgment mechanism triggers the identification of geometric gravity traps. The small variance indicates that the algorithm is trapped in repeated searches and lacks new solution exploration, while the low average complementarity directly indicates the deterioration of layout quality. Subsequently, the weights α, β, and γ are updated according to the judgment results, and the feature complementarity weight α is increased to 1+lnM times the original value. This nonlinear enhancement mechanism adaptively enhances the priority of interlocking features of parts based on the stagnation time length reflected by the M value, forcing the algorithm to turn to the layout position selection with a high probability of interlocking, thereby releasing the closed spatial topology and guiding the search to jump out of the local optimum.

[0061] As a preferred embodiment, the solution of this application is implemented as follows: During the layout process, the historical decision queue continuously records the score changes of the candidate position evaluation dataset. When the system detects that the score variance of multiple consecutive iterations is significantly lower than the preset threshold and the average complementarity remains low, the algorithm is determined to enter the geometric gravity trap state. Then, a weight update operation is performed to increase the feature complementarity weight α by logarithmic proportion, so that the subsequent search strategy prioritizes the geometric interlocking features between parts rather than simple position compactness, thereby effectively identifying and utilizing potential spatial topology.

[0062] Through the above scheme, this application enables the typesetting algorithm to respond to changes in the search state in real time, accurately identify geometric gravity traps and adaptively adjust the weight of the evaluation function, effectively avoiding the problem of stagnant material utilization caused by local optima, and significantly improving the overall quality and space utilization of the typesetting scheme.

[0063] In some of the embodiments described above in this application, a search strategy parameter set is proposed to dynamically adjust the weight of the evaluation function in order to deal with the geometric gravity trap. However, in its implementation, there is a lack of a specific mechanism to effectively integrate the parameter set into the search framework. This means that when the algorithm generates the typesetting action sequence, it may still prioritize locally compact but topologically closed stacking methods due to the greedy strategy. It cannot make full use of feature complementarity to guide the search process to prioritize the exploration of interlocking structures, resulting in limited overall material utilization.

[0064] In this regard, this application further proposes the specific process of generating the typesetting action sequence in step S6, which includes: S61. Receive the search strategy parameter set as state input and predict the priority probability of each node during the expansion phase of the Monte Carlo tree; S62. During the simulation phase, the reward function is calculated based on the weight configuration in parameter set E. U represents the current sheet metal utilization rate, calculated as U = (total area of ​​arranged parts / total sheet metal area) × 100%. η is the utilization rate weighting coefficient, used to balance the weighting of geometric interlocks and material saving in the reward calculation. S63. When the tree search reaches the preset depth L or time threshold, extract the node sequence on the optimal path and generate the typesetting action sequence.

[0065] The search strategy parameter set refers to a set of dynamically adjusted weight factors, which can be implemented using a numerical array stored in a memory cache. Its purpose is to guide the search direction in real time based on changes in the layout state. The policy network of the Monte Carlo tree search framework can be understood as a neural network component used for decision-making in reinforcement learning. It can be implemented using a fully connected network structure, aiming to output the probability distribution of node selection. The expansion phase predicts the priority probability of each node, specifically referring to the evaluation of candidate actions based on the policy network. This can be implemented using a softmax normalization function, aiming to focus on regions with high feature complementarity to avoid blind exploration. The reward function... It refers to the indicators that comprehensively evaluate the quality of typesetting actions, among which The degree of interlocking between parts is quantified, and the material utilization target ηU is balanced. This can be achieved by weighted summation, which aims to force the algorithm to prioritize the identification of interlocking opportunities rather than local compact stacking. The preset depth L or time threshold can be understood as the termination condition constraint of the search process. This can be implemented by an integer counter or system clock, which aims to control the consumption of computing resources and prevent infinite iteration. Extracting the node sequence on the optimal path specifically refers to backtracking the decision chain from the Monte Carlo tree. This can be achieved by a depth-first search algorithm, which aims to output the layout action sequence with the highest interlocking priority.

[0066] Specifically, the scheme in this application injects the search strategy parameter set as state input into the Monte Carlo tree search framework. During the expansion phase, it predicts the node priority probability based on dynamic weights, enabling the search process to respond in real time to changes in the layout state and focus on regions with high feature complementarity, effectively avoiding the geometric gravity trap. During the simulation phase, it configures the reward function based on the parameter set E. By jointly optimizing the sum of feature complementarity and board utilization, the algorithm is forced to prioritize layout actions that can form a tight fit when making decisions, thus breaking through the limitation of traditional heuristic algorithms that only rely on position compactness evaluation. When the tree search reaches the preset depth L or time threshold, the optimal path node sequence is extracted. Through the dual constraint mechanism of depth and time, a high-quality layout action sequence is output under limited computing resources, forming a closed-loop decision-making process from state perception to action generation.

[0067] As a specific implementation method, the solution of this application is implemented as follows: In the cutting machine control system, the search strategy parameter set is passed to the reinforcement learning module through shared memory. The policy network adopts a lightweight convolutional neural network structure to receive the parameter set as input, and generates a node priority probability distribution during the expansion phase; during the simulation phase, the reward function is calculated in real time. Based on the dynamic accumulation of the feature complementarity matrix, ηU dynamically adjusts the weight coefficient according to the current board utilization rate; when the tree search depth reaches the preset threshold or the system clock times out, the backtracking algorithm extracts the optimal path node sequence and converts it into typesetting action instructions. This process is executed in a GPU-accelerated environment to improve real-time performance.

[0068] Through the above technical solutions, the algorithm can effectively avoid the topological closure problem caused by the greedy strategy, and prioritize the exploration of the interlocking structure between parts rather than local compact stacking, thereby significantly improving the material utilization rate while ensuring computational efficiency, and ensuring that the layout process converges stably to the global optimal solution.

[0069] In some of the embodiments described above in this application, a layout action sequence is proposed to guide the cutting and layout process. However, in this process, the parts group forming the interlocking structure is not explicitly identified and marked with precision in the sequence, which makes it impossible for the process instructions to reflect the interlocking relationship between the parts, which may cause the interlocking structure to fail or the material utilization rate to be less than optimal during processing.

[0070] In response, this application further proposes that the processing of the typesetting action sequence in step S7 includes: S71. Perform topological restoration on the typesetting action sequence and identify all part groups that form interlocking structures in the sequence; S72. Assign a unique identification code to each interlocking structure, indicating the required machining accuracy; S73. The final output is a process instruction file containing geometric coordinates and engagement symbols.

[0071] Specifically, topology restoration refers to recovering the implicit spatial topological relationships in the sequence of typesetting actions. This can be achieved using graph theory algorithms, such as treating parts as nodes and interlocking relationships as edges to construct a topological graph. The purpose is to accurately capture the interlocking patterns between parts and avoid missing interlocking structures due to local compact arrangement. Identifying interlocking structures refers to determining the interlocking part groups based on geometric feature matching. This can be achieved using curvature extremum point coupling analysis or concave-convex contour matching, ensuring the complete extraction of interlocking patterns. Specifically, a unique identifier is a distinguishable code assigned to each interlocking structure. This can be achieved using a unique string generated by a hash function or a serial number, establishing independent tracking capabilities. Machining accuracy requirements refer to the technical parameters that constrain manufacturing tolerances. This can be achieved by dynamically setting the tolerance range based on the curvature matching degree in the feature complementarity matrix, ensuring the reliability of physical interlocking between concave and convex areas. Specifically, the process instruction file is a data carrier carrying typesetting guidance information. This can be implemented using structured data formats such as XML or JSON, achieving seamless integration between algorithm output and processing equipment.

[0072] Specifically, the solution in this application extracts the implicit spatial topological relationships between parts through topological reconstruction of the layout action sequence, thereby accurately identifying all part groups forming interlocking structures. Based on the identification results, a unique identifier is assigned to each interlocking structure and the machining accuracy requirements are marked, enabling the interlocking group to have independent tracking capabilities and ensuring that the tolerance constraints meet the curvature matching requirements. Finally, the geometric coordinates and interlocking identifiers are embedded in the process instruction file, allowing the cutting equipment to prioritize the processing of interlocking part groups based on the identifiers. Since topological reconstruction utilizes the implicit topological relationships in the sequence rather than simple geometric coordinates, it avoids the omission of interlocking structures. At the same time, the accuracy marking enforces the constraint of machining tolerances to ensure the physical requirements of interlocking, and the process instruction file directly transmits the interlocking information, thus realizing closed-loop control from algorithm output to actual processing and effectively avoiding the risk of topological closure.

[0073] As a preferred embodiment, the solution of this application is implemented as follows: In the cutting machine control system, the processing unit can be a ARM Cortex-M7 microcontroller, which executes a topology restoration algorithm to identify the meshing groups of C-shaped and U-shaped parts; assigns a unique identifier code based on SHA-256 hash to each meshing structure, and dynamically marks the micron-level machining accuracy requirements according to the feature complementarity matrix; finally outputs a JSON format process instruction file containing geometric coordinates and meshing identifiers for CNC cutting equipment to parse and execute.

[0074] Through the above scheme, this application ensures the traceability and precision control of the processing of interlocking parts, so that the process instructions accurately reflect the interlocking relationship between parts, effectively prevent the failure of the interlocking structure due to manufacturing errors, optimize the use of material space and avoid the risk of topological closure, thereby improving the actual material utilization rate of cutting and layout.

[0075] Example 2 This embodiment is based on Embodiment 1, and the content repeated here will not be described again.

[0076] like Figure 2 As shown, an automatic layout system for a cutting machine based on a heuristic nested algorithm includes: The feature description matrix generation module is used to build a geometric feature extraction network based on deep learning, and to detect the depth of the recess, the direction of the opening, and the extreme point of curvature of the input C-shaped, U-shaped or L-shaped parts, and generate a feature description matrix containing geometric interlocking feature vectors. The feature complementarity matrix generation module is used to take the feature description matrix as input, calculate the feature matching degree between any two parts, and generate a feature complementarity matrix describing the potential bonding ability between parts by combining the opening direction matching degree and the concave-convex coupling degree. The candidate location list generation module is used to perform topological mapping on the remaining space of the current sheet metal based on the feature complementarity matrix, and use the boundary propagation algorithm to filter out the spatial regions that are compatible with the high complementarity parts, and generate a candidate location list with topological constraints. The candidate location evaluation dataset generation module is used to construct a dynamic evaluation function for each location in the candidate location list by combining the distance factor and the shape interlocking coefficient, and to calculate and output a candidate location evaluation dataset containing the current state scores of all candidate points. The search strategy parameter set generation module is used to perform time-series trend analysis on the candidate location evaluation dataset, detect local closed topology risks through backpropagation mechanism, dynamically adjust the weight factors in the evaluation function, and generate an optimized search strategy parameter set. The typesetting action sequence generation module is used to load the search strategy parameter set into the policy network of the Monte Carlo tree search framework, perform iterative search in the reinforcement learning environment, and output the typesetting action sequence with the highest biting priority. The process instruction output module is used to parse the layout action sequence, output the final layout scheme that meets the preset material utilization threshold, and generate a process instruction file with interlocking identifiers.

[0077] Example 3 Figure 3 This is a schematic diagram of an electronic device structure provided in one embodiment of this application. Figure 3 As shown, according to another aspect of this application, an electronic device 500 is also provided. The electronic device 500 may include one or more processors and one or more memories. The memories store computer-readable code that, when executed by the one or more processors, can perform the automatic layout method for a cutting machine based on a heuristic nested algorithm as described above.

[0078] The method or system according to the embodiments of this application can also be used by means of Figure 3 The architecture of the electronic device shown is used to implement this. For example... Figure 3As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, a read-only memory (ROM) 503, a random access memory (RAM) 504, a communication port 505 connected to a network, an input / output component 506, a hard disk 507, etc. The storage device in the electronic device 500, such as the ROM 503 or the hard disk 507, may store the automatic layout method for the cutting machine based on the heuristic nested algorithm provided in this application. The automatic layout method for cutting machines based on heuristic nested algorithms may include, for example, the following steps: S1. Establishing a geometric feature extraction network based on deep learning to detect the concavity depth, opening direction, and curvature extreme points of the input C-shaped, U-shaped, or L-shaped parts, generating a feature description matrix containing geometric interlocking feature vectors; S2. Using the feature description matrix as input, calculating the feature matching degree between any two parts, and generating a feature complementarity matrix describing the potential bonding ability between parts by combining the opening direction matching degree and the concavity-convex coupling degree; S3. Performing topological mapping on the remaining space of the current sheet metal based on the feature complementarity matrix, using a boundary propagation algorithm to filter out spatial regions that are compatible with highly complementary parts, and generating a candidate position list with topological constraints; S4. For each position in the candidate position list, constructing a dynamic evaluation function by combining a distance factor and a shape interlocking coefficient, calculating and outputting a candidate position evaluation dataset containing the current state scores of all candidate points; S5. Performing time-series trend analysis on the candidate position evaluation dataset, detecting local closed topological risks through a backpropagation mechanism, dynamically adjusting the weight factors in the evaluation function, and generating an optimized search strategy parameter set; S6. Load the search strategy parameter set into the policy network of the Monte Carlo tree search framework, perform iterative search in a reinforcement learning environment, and output the layout action sequence with the highest biting priority; S7. Parse the layout action sequence, output the final layout scheme that meets the preset material utilization threshold, and generate a process instruction file with biting identifiers. Further, the electronic device 500 may also include a user interface 508. Of course, Figure 3 The architecture shown is merely exemplary and can be omitted as needed when implementing different devices. Figure 3 One or more components in the illustrated electronic device.

[0079] Example 4 Figure 4 This is a schematic diagram of a computer-readable storage medium structure provided in one embodiment of this application. Figure 4The diagram illustrates a computer-readable storage medium 600 according to one embodiment of this application. The computer-readable storage medium 600 stores computer-readable instructions. When executed by a processor, the computer-readable instructions can perform the automatic layout method for a cutting machine based on a heuristic nested algorithm according to an embodiment of this application, as described with reference to the above figures. The storage medium 600 includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0080] Furthermore, according to the embodiments of this application, the process described in the above-referenced flowchart can be implemented as a computer software program. For example, this application provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be executed by a processor to perform instructions corresponding to the method steps provided in this application, such as: S1. Establishing a geometric feature extraction network based on deep learning, detecting the concave depth, opening direction, and curvature extreme points of the input C-shaped, U-shaped, or L-shaped parts, and generating a feature description matrix containing geometric interlocking feature vectors; S2. Using the feature description matrix as input, calculating the feature matching degree between any two parts, and generating a feature complementarity matrix describing the potential bonding ability between parts by combining the opening direction matching degree and the concave-convex coupling degree; S3. Performing topological mapping on the remaining space of the current sheet metal based on the feature complementarity matrix, and using a boundary propagation algorithm to filter out parts with high complementarity. S4. For each position in the candidate position list, a dynamic evaluation function is constructed by combining the distance factor and the shape interlocking coefficient, and a candidate position evaluation dataset containing the current state scores of all candidate points is calculated and output; S5. The candidate position evaluation dataset is subjected to time-series trend analysis, and the risk of local closed topology is detected by backpropagation mechanism. The weight factors in the evaluation function are dynamically adjusted to generate an optimized search strategy parameter set; S6. The search strategy parameter set is loaded into the policy network of the Monte Carlo tree search framework, and iterative search is performed in the reinforcement learning environment to output the layout action sequence with the highest biting priority; S7. The layout action sequence is parsed, and the final layout scheme that meets the preset material utilization threshold is output, and a process instruction file with biting identifier is generated. When the computer program is executed by the central processing unit (CPU), the above-described functions defined in the method of this application are performed.

[0081] The methods, apparatus, and devices of this application may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the method is for illustrative purposes only, and the steps of the method of this application are not limited to the order specifically described above, unless otherwise specifically stated. Furthermore, in some embodiments, this application may also be implemented as a program recorded on a recording medium, the program including machine-readable instructions for implementing the method according to this application. Thus, this application also covers recording media storing programs for performing the method according to this application.

[0082] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0083] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An automatic layout method for a cutting machine based on a heuristic nested algorithm, characterized in that, Includes the following steps: S1. Establish a geometric feature extraction network based on deep learning to detect the depth of the recess, the direction of the opening, and the extreme points of curvature of the input C-shaped, U-shaped, or L-shaped parts, and generate a feature description matrix containing geometric interlocking feature vectors. S2. The feature description matrix is ​​used as input to calculate the feature matching degree between any two parts. By combining the opening direction matching degree and the concave-convex coupling degree, a feature complementarity matrix describing the potential bonding ability between parts is generated. S3. Based on the feature complementarity matrix, perform topological mapping on the remaining space of the current plate, use the boundary propagation algorithm to filter out the spatial regions that are compatible with the high complementarity parts, and generate a candidate position list with topological constraints. S4. For each position in the candidate position list, construct a dynamic evaluation function by combining the distance factor and the shape interlocking coefficient, calculate and output a candidate position evaluation dataset containing the current state scores of all candidate points; S5. Perform time-series trend analysis on the candidate location evaluation dataset, detect local closed topology risks through backpropagation mechanism, dynamically adjust the weight factors in the evaluation function, and generate an optimized search strategy parameter set. S6. Load the search strategy parameter set into the policy network of the Monte Carlo tree search framework, perform iterative search in the reinforcement learning environment, and output the typesetting action sequence with the highest biting priority; S7. Parse the layout action sequence, output the final layout scheme that meets the preset material utilization threshold, and generate a process instruction file with interlocking identifiers.

2. The method according to claim 1, characterized in that, The specific process for generating the feature complementarity matrix includes: S21. Extract the unit vector of the opening direction of part i from the feature description matrix. The convex direction vector of part j Calculate cosine similarity ;in, It is normalized, has a modulus of 1, and only indicates direction. The unit vector representing the direction of the convexity. This represents the direction cosine similarity between part i and part j, with a value range of [-1, 1]. S22. Extract the maximum curvature of the concave portion of part i from the feature description matrix. and the maximum curvature of the protruding part of part j Calculate the curvature coupling factor; S23. Based on the above calculation results, use the formula... Generate matrix elements and construct a feature complementarity matrix. This represents the maximum curvature value on the contour line of the recessed area of ​​part i. This represents the maximum curvature value on the contour line of the raised area of ​​part j, where This is a preset curvature tolerance parameter. It is a preset constant. This represents the feature complementarity between parts i and j.

3. The method according to claim 2, characterized in that, The specific process of generating the candidate location list in step S3 includes: S31. Read the part pairs in the feature complementarity matrix whose complementarity is greater than a preset threshold; S32. Discretize the current sheet metal into a grid map, and propagate directionally along the concave direction only from the free boundary grid corresponding to the high complementarity parts. The propagation distance d and the width of the part opening w satisfy d=λw; d represents the propagation distance in the boundary propagation algorithm, w represents the physical width at the part opening, and λ represents the propagation coefficient. S33. Record the nodes that satisfy the spatial constraints in the propagation path and generate a list of candidate locations with topological constraints that include adjacency relationships.

4. The method according to claim 3, characterized in that, The specific process of generating the candidate location evaluation dataset in step S4 includes: S41. Iterate through each position k in the candidate position list; S42. Calculate the distance Dk from position k to the plate boundary, and calculate the shape interlocking coefficient based on the feature complementarity of adjacent parts around position k. ; Values ​​range from [0, 1]. S43. Using the formula Calculate the comprehensive score for the current candidate position, and obtain the F-values ​​and their corresponding state components for all positions. Combine the results to output a candidate location evaluation dataset. Indicates the dynamic decay factor. To control the weights for complementary features; This indicates the weight for controlling placement near the edge; This indicates the weight that controls the local interlocking.

5. The method according to claim 4, characterized in that, The specific process of generating the search strategy parameter set in step S5 includes: S51. Establish a historical decision queue and continuously record the score changes in the candidate location evaluation dataset; S52. When a candidate location is detected, evaluate the score variance of the dataset for M consecutive iterations. Furthermore, if the average complementarity is below the threshold θ, it is determined that the algorithm has entered a geometric gravity trap; M is an integer used to determine whether the algorithm has stalled. This is the average threshold for complementarity; when the historical average score is lower than this value, the typesetting quality is considered poor. S53. Update the weights based on the judgment result. ,in The weight is increased to the original value. The updated search strategy parameter set is then output.

6. The method according to claim 5, characterized in that, The specific process of generating the typesetting action sequence in step S6 includes: S61. Receive the search strategy parameter set as state input and predict the priority probability of each node during the expansion phase of the Monte Carlo tree; S62. During the simulation phase, the reward function is calculated based on the weight configuration in the parameter set. ; Based on the current utilization rate of the board material, This is the utilization rate weighting coefficient; S63. When the tree search reaches the preset depth L or time threshold, extract the node sequence on the optimal path and generate the typesetting action sequence.

7. The method according to claim 6, characterized in that, The processing of the typesetting action sequence in step S7 includes: S71. Perform topological restoration on the typesetting action sequence and identify all part groups that form interlocking structures in the sequence; S72. Assign a unique identification code to each interlocking structure, indicating the required machining accuracy; S73. The final output is a process instruction file containing geometric coordinates and engagement symbols.

8. An automatic layout system for a cutting machine based on a heuristic nested algorithm, the system being used to execute the method described in any one of claims 1 to 8, characterized in that, include: The feature description matrix generation module is used to build a geometric feature extraction network based on deep learning, and to detect the depth of the recess, the direction of the opening, and the extreme point of curvature of the input C-shaped, U-shaped or L-shaped parts, and generate a feature description matrix containing geometric interlocking feature vectors. The feature complementarity matrix generation module is used to take the feature description matrix as input, calculate the feature matching degree between any two parts, and generate a feature complementarity matrix describing the potential bonding ability between parts by combining the opening direction matching degree and the concave-convex coupling degree. The candidate location list generation module is used to perform topological mapping on the remaining space of the current sheet metal based on the feature complementarity matrix, and use the boundary propagation algorithm to filter out the spatial regions that are compatible with the high complementarity parts, and generate a candidate location list with topological constraints. The candidate location evaluation dataset generation module is used to construct a dynamic evaluation function for each location in the candidate location list by combining the distance factor and the shape interlocking coefficient, and to calculate and output a candidate location evaluation dataset containing the current state scores of all candidate points. The search strategy parameter set generation module is used to perform time-series trend analysis on the candidate location evaluation dataset, detect local closed topology risks through backpropagation mechanism, dynamically adjust the weight factors in the evaluation function, and generate an optimized search strategy parameter set. The typesetting action sequence generation module is used to load the search strategy parameter set into the policy network of the Monte Carlo tree search framework, perform iterative search in the reinforcement learning environment, and output the typesetting action sequence with the highest biting priority. The process instruction output module is used to parse the layout action sequence, output the final layout scheme that meets the preset material utilization threshold, and generate a process instruction file with interlocking identifiers.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the automatic layout method for a cutting machine based on a heuristic nested algorithm as described in any one of claims 1-7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program adapted for loading by a processor to perform the steps in the automatic layout method for a cutting machine based on a heuristic nested algorithm as described in any one of claims 1-7.

Images