Control method, device and storage medium for laser cutting path planning

Abstract

This application discloses a control method, device, and storage medium for laser cutting path planning, relating to the field of laser cutting technology. The method includes: discretizing the geometric contour data of the part to be processed to obtain node clusters and candidate puncture nodes; combining processing priority constraints and virtual starting nodes to construct a generalized cost matrix; performing two-segment decoding on each encoded population to obtain the target puncture point and cutting access order of the part corresponding to the encoded population; determining elite individuals based on the generalized cost matrix, the target puncture point, and the cutting access order; using the elite individuals as initial solutions, adjusting the neighborhood search intensity, and performing local path optimization on the initial solutions according to cascaded neighborhood operators to obtain the target cutting path. This application solves the problem of poor path planning by constructing a generalized cost matrix, using two-segment encoding and decoding, and adaptive neighborhood optimization, achieving globally optimal planning of the laser cutting path.

Description

Technical Field

[0001] This application relates to the field of laser cutting technology, and in particular to a control method, device and storage medium for laser cutting path planning. Background Technology

[0002] In laser cutting processing scenarios, the optimization efficiency and scenario adaptability of cutting path planning are directly related to the overall processing efficiency and energy consumption cost control. Path planning algorithms that adapt to multiple constraints and scenarios are the core foundation for improving the intelligence level of laser cutting processing.

[0003] In related technologies, models are built for different scenarios, and metaheuristic algorithms with fixed neighborhood operators are used to solve path planning problems. However, this approach is not well adapted to multi-constraint scenarios, resulting in excessively long empty blade travel in the cutting path planning, which in turn leads to poor laser cutting path planning.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this application is to provide a control method, device, and storage medium for laser cutting path planning, aiming to solve the technical problem of poor laser cutting path planning.

[0006] To achieve the above objectives, this application proposes a control method for laser cutting path planning, the method comprising: Discretize the geometric contour data of the part to be processed to obtain node clusters and candidate puncture nodes; A virtual starting node is introduced, and the node cluster, the candidate puncture node, and the processing priority constraint are used to construct a generalized cost matrix according to the processing mode. The coding population is initialized using a two-segment continuous variable encoding and a hybrid heuristic algorithm. Perform two-segment decoding on each of the coded populations to obtain the target puncture point and the cutting access order of the part corresponding to the coded population; Based on the generalized cost matrix, the target puncture point of the part, and the cutting and access order of the part, the overall fitness of the encoded population is evaluated to identify elite individuals. The elite individuals are used as the initial solution for local path optimization, and the initial solution is subjected to local path optimization according to the cascaded neighborhood operator to obtain the target cutting path. The neighborhood search intensity of each iteration in the local path optimization is determined according to the iteration progress.

[0007] In one embodiment, based on the graphic file corresponding to the layout of the parts to be processed, the entity data segments in the graphic file are located and parsed to determine the feature parameters of various geometric primitives; The feature parameters of the geometric primitives are subjected to primitive splicing operation to obtain the initial contour of the part to be processed; The initial contour is preprocessed by normalization to remove redundant points and correct minor contour deviations, thereby obtaining the geometric contour data of the part to be processed. Discretize the geometric contour data of the part to be processed, divide the contour of a single part into independent node clusters, and determine the feasible candidate puncture nodes within the node clusters.

[0008] In one embodiment, candidate puncture nodes within the node cluster are used as processing nodes, and the sequential processing associations corresponding to the processing priority constraints are embedded to obtain the augmented graph base topology. The virtual starting node is added to the node set of the augmented graph's basic topology, and the augmented graph model is obtained by uniformly modeling according to the processing mode. Based on the augmented graph model, the node configuration cost is obtained by configuring the idle travel cost between processing nodes and the return travel cost corresponding to the virtual starting node according to the processing mode. The node configuration costs between all nodes are arranged into a unified structured generalized cost matrix according to the arrangement order of the nodes in the augmented graph model.

[0009] In one embodiment, the first segment of continuous variables of individuals in the encoded population is discretely mapped to obtain the continuous node values ​​corresponding to the node cluster; The continuous values ​​of the nodes are converted into candidate puncture point indices within the node cluster to determine the target puncture point of the part to be processed. The second continuous variable of the individuals in the encoded population is sorted and mapped to obtain the access order corresponding to the node cluster; The access order of the node clusters is determined by sorting the access positions of the second continuous variable according to their numerical values, thereby determining the cutting access order of the part to be processed.

[0010] In one embodiment, an initial candidate path is generated based on the target puncture point and the cutting access order; The virtual start node in the initial candidate path is moved to the beginning of the path to obtain the standard candidate path; The generalized cost matrix is ​​invoked to extract the movement cost between nodes in the standard candidate path in turn, and the total empty trip cost corresponding to the standard candidate path is calculated. Check whether the cutting order of the standard candidate path conforms to the processing priority constraint, calculate the penalty cost for constraint violation in stages, and generate the path calculation result by combining the total cost of empty travel; Based on the path calculation results and cost weights, the overall fitness of the standard candidate path is calculated using a weighted average. After ranking the comprehensive fitness of the standard candidate paths, the optimal adaptation path is determined, and the optimal adaptation path and its path information are encapsulated to obtain the elite individual.

[0011] In one embodiment, the elite individual is used as the initial solution for local path optimization, and the neighborhood search strength of the adaptive path is calculated according to the current iteration progress to obtain the initial optimization object; Disrupt the execution order of the cascaded neighborhood operators and initialize the continuous unimproved count of the local search to obtain the neighborhood search task to be executed; According to the neighborhood search task, the initial optimization object is searched for a neighborhood, and the path is updated by exhaustive search within the operator and first improvement between operators, until the number of consecutive unimproved times reaches the neighborhood search strength threshold, and the intermediate path is output. If the intermediate path is locally optimal and the verification result is confirmed to be a non-local extremum state, then the intermediate path is output as the target cutting path.

[0012] In one embodiment, the verification result of the intermediate path is obtained by comparing the number of consecutive unimproved rounds during the neighborhood search process with the maximum number of unimproved rounds threshold. If the verification result determines that the current location is trapped in a local extremum, a perturbation mechanism is triggered. According to the preset perturbation strength parameter, non-repeating neighborhood operators are selected from the set of available neighborhood operators, and the execution order of the neighborhood operators is determined. According to the execution order of the neighborhood operators, the intermediate path is subjected to a corresponding number of neighborhood transformation operations to obtain the perturbed candidate path; The processing priority constraints of the disturbed candidate paths are checked, and invalid path segments that violate the requirement that the inner hole is processed before the outer contour are eliminated, thus determining the target cutting path.

[0013] In one embodiment, the idle travel time, total cutting time, and puncture point loss of the target cutting path during actual processing are collected and organized according to time sequence to obtain the actual data set of this processing. The measured dataset was matched with a processing performance benchmark threshold, and a quantitative deviation analysis was performed from three dimensions: path running efficiency, processing device wear, and processing rule compliance, to obtain the performance deviation results. Based on the performance deviation results, the problem points that lead to the performance deviation benchmark threshold are determined according to the deviation level, and the influence factor weights of the core parameters of each problem point in the path planning process on the performance deviation results are quantified. The core parameters corresponding to the path planning process are corrected according to the order of the influence factor weights from high to low to obtain an updated path planning parameter set, so as to optimize the generation of the cutting path for the next processing part.

[0014] In addition, to achieve the above objectives, this application also proposes a laser cutting path planning device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the laser cutting path planning control method as described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the control method for laser cutting path planning as described above.

[0016] This application provides a control method for laser cutting path planning, which includes obtaining node clusters and candidate puncture nodes by discretizing the geometric contour data of the part to be processed, introducing virtual starting nodes and constructing a generalized cost matrix by combining processing priority constraints and processing modes, using two-segment continuous variable encoding to achieve population initialization of a hybrid heuristic algorithm, synchronously outputting the target puncture point and cutting access order through two-segment decoding, completing the comprehensive fitness evaluation of the population individuals based on the generalized cost matrix to select elite individuals, and then using the elite individuals as the initial optimization solution, combining cascaded neighborhood operators and adaptively adjusting the neighborhood search intensity according to the iteration progress to carry out local path optimization, and obtaining the optimal target cutting path. This method solves the technical problems of traditional laser cutting path planning, such as the inability to coordinate the optimization of puncture point selection and contour access order, the tendency of fixed search strategies to get trapped in local optima, weak adaptability to process constraints, and low path optimization efficiency. It improves the global search capability and adaptability of processing priority constraints in path planning, reduces the idle travel loss of the laser head, and effectively improves the path rationality and overall processing efficiency of laser cutting of complex contour parts.

[0017] In summary, this application solves the technical problem of low overall efficiency in existing path optimization processes by constructing a generalized cost matrix, using two-segment encoding and decoding, and adaptive neighborhood optimization for laser cutting path planning. It significantly reduces the empty cutting stroke and improves the algorithm's convergence speed and processing efficiency. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the first embodiment of the laser cutting path planning control method of this application; Figure 2 This is a schematic diagram of the augmented graph model of this application; Figure 3 This is a flowchart illustrating the solution logic of the hybrid algorithm in this application; Figure 4 This is a flowchart illustrating the seventh embodiment of the laser cutting path planning control method of this application; Figure 5 This is a flowchart illustrating the eighth embodiment of the laser cutting path planning control method of this application; Figure 6 This is a schematic diagram of the laser cutting path planning equipment of this application.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] In related technologies, models are built for different scenarios, and metaheuristic algorithms with fixed neighborhood operators are used to solve path planning problems. However, this approach is not well adapted to multi-constraint scenarios, resulting in excessively long empty blade travel in the cutting path planning, which in turn leads to poor laser cutting path planning.

[0024] This application provides a solution: First, the geometric contour data of the part to be processed is discretized to obtain node clusters and candidate puncture nodes. Then, a virtual starting node is introduced. The node clusters, candidate puncture nodes, and processing priority constraints are used to construct a generalized cost matrix according to the processing mode. Then, a coding population is initialized through two-segment continuous variable encoding and a hybrid heuristic algorithm. Next, two-segment decoding is performed on each coding population to obtain the target puncture point and the cutting access order of the part corresponding to the coding population. Then, based on the generalized cost matrix, the target puncture point, and the cutting access order, the individual comprehensive fitness of the coding population is evaluated to determine elite individuals. Finally, the elite individuals are used as the initial solution for local path optimization, and the initial solution is used for local path optimization according to the cascaded neighborhood operator to obtain the target cutting path. The neighborhood search intensity of each iteration in the local path optimization is determined according to the iteration progress.

[0025] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or laser cutting path planning device capable of performing the above functions. The following description uses a laser cutting path planning device as an example to illustrate this embodiment and the subsequent embodiments.

[0026] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0027] This application provides a control method for laser cutting path planning, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the laser cutting path planning control method of this application.

[0028] In this embodiment, the control method for laser cutting path planning includes steps S10 to S50: Step S10: Discretize the geometric contour data of the part to be processed to obtain the node cluster and candidate puncture nodes.

[0029] The geometric contour data of the part to be processed is vector graphic data describing the inner hole, outer contour, and processing boundary of the part. It includes the coordinate information, closure attribute, hierarchical relationship, and core parameters of the processing type. For example, it includes the vector coordinate set of the inner hole contour, the closed path data of the outer contour, the nesting hierarchy of the contours, and the processing type labels of the inner hole and outer contour. A node cluster is a set of discrete nodes corresponding to contours of the same processing priority, grouped and categorized according to the nesting hierarchy and processing type of the contours. For example, all discrete nodes of the same inner hole contour form an independent node cluster, and all discrete nodes of the same outer contour form an independent node cluster. A candidate puncture node is a discrete node within a node cluster that can serve as the starting point for laser head puncture. It is a contour node that meets the requirements of laser cutting process, has no contour interference, and can achieve a smooth start to cutting.

[0030] In this embodiment, the geometric contour data of the part to be processed is read, and discrete processing is performed to obtain node clusters and candidate puncture nodes.

[0031] For example, there are three methods for discretizing the geometric contour data of the part to be processed and generating node clusters and candidate puncture nodes. The first method is a full-scale discretization process with contour layer pre-scan. Before the path planning task is officially started, the complete metadata of the geometric contour data of the part to be processed is read, the closure attributes, nesting levels and processing types of all contours are parsed, the inner holes and outer contours are classified into layers, and then global equidistant discrete sampling is performed on all contours to generate the initial discrete node set corresponding to each contour. Subsequently, the nodes are clustered according to the nesting level of the contours to obtain the node clusters corresponding to each contour. Finally, discrete nodes that meet the requirements of laser cutting process and have no path interference are selected in each node cluster to generate candidate puncture nodes for the corresponding node cluster. This method has a simple and stable processing logic, comprehensive contour coverage, and is not prone to node missing or clustering errors, providing comprehensive and reliable basic data for subsequent path planning.

[0032] The second method is feature-driven streaming incremental discretization. During the geometric contour data reading and parsing process, incremental discretization is performed simultaneously with the reading of contour vector data. It identifies straight segments, arc segments, and corner features of the contour segment by segment, performing differentiated discretization sampling for different feature types. For high-curvature corner regions, denser sampling is performed, while for low-curvature straight segments, sparser sampling is performed. This accumulates to generate discrete node sets for each contour. After all contours have been read, nodes are clustered according to the contour's processing type and nesting level, resulting in corresponding node clusters. Simultaneously, during the discretization process, nodes within each contour segment that meet the process requirements are screened, ultimately generating candidate puncture nodes corresponding to each node cluster. This method executes synchronously with the contour data parsing process, eliminating the need for additional pre-processing time. It fully utilizes idle periods during data reading to complete the processing, while simultaneously achieving differentiated sampling for contour features, balancing discretization accuracy and processing efficiency.

[0033] The third method is a priority-oriented clustered synchronous discretization process. First, based on processing priority constraints, all contours of the part to be processed are divided into a high-priority inner hole contour group and a low-priority outer contour group. Then, parallel discretization processing is initiated simultaneously for both groups. High-precision encrypted discretization is performed on the high-priority inner hole contour group to ensure the selection accuracy of candidate puncture nodes. Adaptive sparse discretization is performed on the low-priority outer contour group to control the overall data processing volume. After discretization, node clustering is completed on a single independent contour basis, resulting in node clusters corresponding to each contour. Then, for each node cluster, combined with the contour's cutting direction and laser head movement path constraints, candidate puncture nodes that are interference-free and meet process requirements are selected. This method, guided by processing priority, prioritizes the processing accuracy of high-priority contours while significantly improving overall processing efficiency through parallel processing, adapting to the rapid processing needs of complex, multi-nested contour parts.

[0034] In an exemplary scheme for determining node clusters and candidate puncture nodes, the system first loads preset contour discretization rules and clustering rules. These rules include internal hole contour discretization parameters, external contour discretization parameters, candidate puncture node process screening thresholds, and node clustering priority rules. Then, contour layering parsing is performed, reading the geometric contour data of the part to be processed, parsing the closure attributes, nesting levels, and processing types of each contour, dividing all contours into an internal hole contour set and an external contour set, and marking the processing priority level of each contour. Next, contour discretization processing is performed. For the internal hole contour set, equidistant discretization is performed according to a preset encrypted sampling step size to generate a discrete node set for the internal hole contour. For the external contour set, equidistant discretization is performed according to a preset standard sampling step size to generate a discrete node set for the external contour. Then, node clustering processing is performed, using a single independent closed contour as the smallest clustering unit, grouping all discrete nodes corresponding to a single contour into the same node cluster, and assigning a unique cluster identifier and corresponding processing priority level to each node cluster. Next, candidate puncture node screening is performed. For each node cluster, nodes that do not meet the starting cutting process and are located at contour corners or high curvature areas are first removed. Then, nodes that have the risk of interference with other contours are removed. Finally, the discrete nodes that meet the process requirements are retained as candidate puncture nodes for that node cluster. After screening all node clusters, the node clusters and corresponding candidate puncture nodes of all clusters are summarized and generated for subsequent generalized cost matrix construction.

[0035] Step S20: Introduce a virtual starting node, and construct a generalized cost matrix based on the processing mode by combining the node cluster, the candidate puncture nodes, and the processing priority constraints.

[0036] Processing priority constraints are pre-defined, mandatory rules in laser cutting processes used to regulate the order of contour processing. The core constraint is that internal hole contours are processed before their corresponding external contours, and higher-priority nested contours are processed before lower-priority contours. Virtual start nodes are virtual nodes added to the topology to unify path modeling for different processing modes. They represent the starting and resetting positions of the laser head and have no actual processing action; they are only used for closed-loop and open-loop path modeling. Processing modes refer to the laser cutting path operation modes, including closed-loop resetting processing modes and open-loop non-resetting processing modes. Closed-loop resetting processing modes require the laser head to return to the starting point after completing all processing, while open-loop non-resetting processing modes require the laser head to remain at the final processing point after completing all processing. The generalized cost matrix is ​​a standardized matrix used to quantify the comprehensive cost of the laser head moving between any two nodes. The elements in the matrix represent the combined value of the idle travel cost, return cost, and constraint violation penalty cost between the corresponding two nodes.

[0037] In this embodiment, a virtual starting node is introduced. There are three ways to construct a generalized cost matrix based on the processing mode, including the node cluster, the candidate puncture nodes, and processing priority constraints. The first is a static topology construction method with constraints pre-embedded. First, the candidate puncture nodes within the node cluster are initialized as real processing nodes. The processing priority constraints corresponding to the sequential processing relationships are pre-embedded, and traversable directed connections between nodes that meet the constraints are defined. Illegal connections that violate the priority constraints are eliminated, resulting in the augmented graph's basic topology. Then, the virtual starting node is added to the augmented graph's basic topology. The connection relationships and corresponding costs between the virtual starting node and all real processing nodes are configured according to the processing mode requirements. Finally, the connection costs between all nodes are integrated according to the node arrangement order to generate a static generalized cost matrix. This method pre-embeds constraints into the topology structure, eliminating illegal paths at the source. The matrix structure is stable, the calculation logic is simple, and it is suitable for conventional single-part batch processing scenarios.

[0038] The second approach is a dynamic cost mapping construction method adapted to the processing mode. First, the modeling rules and cost calculation rules for the virtual starting node are determined according to the processing mode type. Then, candidate puncture nodes within the node cluster are used as real processing nodes. Combined with processing priority constraints, the range of accessible predecessor and successor nodes is configured for each real processing node, generating a dynamic node connection topology. Subsequently, for the closed-loop reset processing mode, the double-ended return cost between the virtual starting node and the first processing node, and between the final processing node and the virtual starting node, is configured. For the open-loop non-reset processing mode, only the single-ended starting cost between the virtual starting node and the first processing node is configured. Finally, according to the dynamic connection relationship of the nodes, the empty travel cost, return cost, and constraint violation penalty cost between all nodes are mapped to a unified matrix structure, generating a dynamically adapted generalized cost matrix. This method can flexibly adapt to different processing modes without reconstructing the topology structure; matrix construction can be completed simply by adjusting the cost mapping rules, adapting to flexible processing scenarios with multiple mode switching.

[0039] The third approach is a multi-objective weighted closed-loop topology construction method. First, it extracts three core optimization objectives for laser cutting: idle travel length, processing efficiency, and device loss, and assigns corresponding weight coefficients to each objective. Then, candidate puncture nodes within the node cluster are used as actual processing nodes, embedding processing priority constraints to generate an augmented graph base topology. Subsequently, a virtual starting node is added to the topology, completing the unified modeling of closed-loop reset and open-loop non-reset processing modes. Based on the multi-objective weight coefficients, the weighted comprehensive value of idle travel cost, puncture loss cost, and constraint violation penalty cost between nodes is calculated. Finally, the weighted comprehensive cost between all nodes is integrated according to the node arrangement order to generate a generalized cost matrix for multi-objective optimization. This method can simultaneously consider multiple processing optimization objectives, and the cost values ​​in the matrix are more closely aligned with the comprehensive requirements of actual processing, making it suitable for high-precision, low-loss high-end parts processing scenarios.

[0040] In an exemplary scheme for constructing a generalized cost matrix, the system's preset processing mode configuration rules, processing priority constraint rules, and cost calculation standards are first loaded. Then, processing node initialization is performed, treating all candidate puncture nodes within node clusters as actual processing nodes. A unique node identifier is assigned to each processing node, marking the processing priority level of the corresponding node cluster, thus completing the initialization of the processing node set. Next, augmented graph basic topology construction is performed. Based on processing priority constraints, permissible directed connection rules are defined for each processing node: only directed connections from high-priority nodes to low-priority nodes are allowed, reverse connections from low-priority nodes to high-priority nodes are prohibited, while bidirectional connections between nodes of the same priority level are allowed. The definition of directed connections between nodes is completed according to this rule, generating the augmented graph basic topology. Then, virtual node adaptation is performed. According to the requirements of the current processing mode, virtual start nodes are added to the node set of the augmented graph basic topology. For the closed-loop reset processing mode, start connections from the virtual start node to all actual processing nodes and reset connections from all actual processing nodes to the virtual start node are configured. For the open-loop, non-reset processing mode, only virtual start nodes are configured to point to the initial connections of all real processing nodes; no reset connections are configured. Next, cost calculation is performed. For all valid directed connections between nodes, the idle travel cost between the corresponding two nodes is calculated. For illegal connections that violate processing priority constraints, a preset high penalty cost is configured. For connections corresponding to the virtual start node, the corresponding start cost and return cost are calculated according to the processing mode. Finally, matrix integration is performed. All nodes are arranged in the order of virtual start node, high-priority processing node, and low-priority processing node. The calculated costs between all nodes are filled into the corresponding matrix positions to generate a standardized generalized cost matrix for subsequent population individual fitness assessment.

[0041] Step S30: Initialize the coding population using two-segment continuous variable coding and a hybrid heuristic algorithm.

[0042] Two-segment continuous variable encoding is an encoding method that separates the two optimization objectives of puncture point selection and access order planning into two segments of continuous variables. The first segment of continuous variables corresponds to the selection of candidate puncture points within a node cluster, and the second segment of continuous variables corresponds to the cutting and access order of the node cluster. The hybrid heuristic algorithm population is a set of algorithm solutions composed of multiple independent encoding individuals, each corresponding to an independent candidate cutting path scheme.

[0043] In this embodiment, there are three ways to initialize the coding population using two-segment continuous variable encoding and a hybrid heuristic algorithm. The first is a cluster-aligned synchronous encoding and decoding method. First, the fixed length of the two-segment encoding is determined based on the total number of node clusters of the part to be processed, wherein the encoding lengths of the first and second segments are consistent with the total number of node clusters. Then, a two-segment continuous variable encoding of corresponding length within the 0-1 interval is generated for each individual coding population using the hybrid heuristic algorithm, thus completing the initialization of the coding population.

[0044] The second method is a progressive encoding and decoding method with priority constraint embedding. First, based on processing priority constraints, the node clusters are divided into high-priority inner hole clusters and low-priority outer contour clusters. The layer length of the two-segment encoding is determined according to the number of the two groups of clusters. The first segment of encoding is divided into a high-priority segment and a low-priority segment, which correspond to the puncture point selection of the two groups of clusters respectively. The second segment of encoding is also divided into a high-priority segment and a low-priority segment, which correspond to the internal access order of the two groups of clusters respectively. Then, a layered two-segment continuous variable encoding is generated for each individual in the population to complete the initialization of the encoded population.

[0045] The third method is an adaptive length dynamic encoding and decoding method. First, based on the number of candidate puncture nodes in the node cluster, an appropriate encoding length is configured for each node cluster to generate the first dynamic length encoding segment. Then, based on the total number of node clusters, the second fixed length encoding segment is determined. These are combined to form a two-segment dynamic length continuous variable encoding. Finally, a corresponding dynamic length encoding is generated for each individual in the encoding population to complete the initialization of the encoding population.

[0046] Step S40: Perform dual-segment decoding on each of the encoded populations to obtain the target puncture point and the cutting access order of the part corresponding to the encoded population.

[0047] Dual-segment decoding is the process of converting dual-segment continuous variable encoding into a mapping between the actual puncture point selection result and the cutting access sequence. The target puncture point of the part is the laser puncture starting point corresponding to the contour, ultimately selected from the candidate puncture nodes of each node cluster. The part cutting access sequence is the order in which the contours corresponding to all node clusters are processed.

[0048] In this embodiment, the implementation of initializing the coding population through two-segment continuous variable encoding and a hybrid heuristic algorithm includes three methods. The first is a cluster-aligned synchronous encoding and decoding method, which performs synchronous two-segment decoding on all individuals in the population: For the first segment of continuous variables, the continuous variables are mapped to the indices of candidate puncture nodes within the corresponding node clusters according to the order of the node clusters, thus determining the target puncture point for each node cluster. For the second segment of continuous variables, they are sorted in ascending order according to their numerical values ​​to obtain the access position of the node clusters, determining the cutting access order of all node clusters, and synchronously completing the decoding output of both dimensions. This method has a fixed encoding length, synchronously aligned decoding logic, and is less prone to mismatches between encoding and node clusters. The computational logic is simple and stable, and it is suitable for processing scenarios with a fixed number of node clusters.

[0049] The second method is a progressive encoding and decoding approach with priority constraint embedding. It performs a progressive two-segment decoding: first, it decodes the first and second high-priority segments to determine the target puncture points and internal access order of high-priority inner hole clusters; then, it decodes the first and second low-priority segments to determine the target puncture points and internal access order of low-priority outer contour clusters. Finally, it integrates these to generate a complete set of target puncture points and cutting access order. During the decoding process, it strictly guarantees that the access order of all high-priority clusters precedes that of low-priority clusters. This method directly embeds processing priority constraints into the entire encoding and decoding process, fundamentally avoiding the generation of invalid paths that violate constraints, significantly reducing the proportion of invalid solutions, improving the algorithm's solution efficiency, and adapting to complex part processing scenarios with multiple nesting and multiple priority levels.

[0050] The third method is an adaptive length dynamic encoding and decoding approach, which performs adaptive two-segment decoding: For the first segment of dynamic length encoding, each node cluster is decoded and mapped to the index of the corresponding candidate puncture node according to its corresponding encoding length, thus determining the target puncture point for each node cluster. For the second segment of fixed length encoding, the node clusters are sorted by numerical value to obtain their access order. Simultaneously, the access order is checked to ensure it conforms to processing priority constraints. Positions that violate the constraints are automatically corrected, ultimately outputting the target puncture point and cutting access order that meet the constraints. This method can adaptively adjust the encoding length according to the number of candidate nodes in each node cluster, balancing encoding accuracy and efficiency. Furthermore, it automatically performs constraint verification and correction during the decoding process, making it suitable for processing irregular parts with varying numbers of candidate puncture nodes.

[0051] In an exemplary scheme for determining the target puncture point and cutting access order, the system first loads pre-set hybrid heuristic algorithm population parameters, two-segment encoding and decoding rules, and processing priority constraint verification rules. These parameters include population size, encoding value range, decoding mapping rules, and constraint correction thresholds. Next, encoding length determination is performed by counting the total number of node clusters of the part to be processed and the number of candidate puncture nodes within each node cluster, determining the total length of the two-segment continuous variable encoding. The first segment encoding length is the same as the total number of node clusters, and the second segment encoding length is also the same as the total number of node clusters. The encoding value range is a continuous interval from 0 to 1. Then, population initialization is performed. According to the preset population size, a set of independent two-segment continuous variable encodings conforming to the value range requirements is generated for each individual in the population. The encoding sets of all individuals constitute the initial population of the hybrid heuristic algorithm. Subsequently, two-segment decoding preprocessing is performed, assigning corresponding encoding bit numbers to each node cluster, establishing a one-to-one correspondence between encoding bits and node clusters. Simultaneously, assigning corresponding index numbers to candidate puncture nodes of each node cluster, establishing a mapping rule between continuous variable values ​​and index numbers. Next, the first decoding stage is executed. For each individual in the population, the continuous variables of the first stage are extracted bit by bit and converted into index numbers of candidate puncture nodes within the corresponding node cluster according to the mapping rules. The candidate puncture nodes with the corresponding indices are selected, which are the target puncture points of the part for that node cluster. After decoding all node clusters, the target puncture points of the part for all clusters corresponding to that individual are generated. Then, the second decoding stage is executed. For each individual in the population, the continuous variables of the second stage are extracted and sorted in ascending order of value to obtain the sorting position corresponding to each encoding bit. According to the correspondence between the encoding bit and the node cluster, the access position of each node cluster is determined. The part cutting access order corresponding to that individual is generated in order of position from first to last. Then, constraint verification and correction are performed to verify whether the cutting access order obtained by decoding conforms to the processing priority constraint that inner holes are processed before outer contours. For access orders that violate the constraint, position adjustment and correction are performed to ensure that the access position of all high-priority inner hole clusters is before that of their respective low-priority outer contour clusters. After decoding all individuals in the population, the target puncture point and the order of cutting access for each encoded individual are summarized and output for subsequent comprehensive fitness evaluation of the population.

[0052] Step S50: Based on the generalized cost matrix, the target puncture point of the part, and the cutting access order of the part, evaluate the overall fitness of the encoded population and determine elite individuals.

[0053] Overall fitness is a core metric used to quantify the merits of candidate cutting paths for individual members of the population. The overall fitness value is negatively correlated with the overall processing cost of the path; that is, the lower the idle travel cost and the higher the constraint compliance of the path, the higher the overall fitness value. Elite individuals are those with the highest overall fitness and the best path performance among all individuals in the coding population. They fully retain all information regarding the corresponding part target puncture point, part cutting access order, and overall path cost, serving as the core foundation for subsequent local path optimization.

[0054] In this embodiment, there are three ways to evaluate the overall fitness of individuals in the coded population and determine elite individuals. The first is a multi-dimensional, hierarchical, weighted fitness evaluation method. First, three evaluation dimensions are extracted: path empty travel cost, processing priority constraint compliance, and puncture point process adaptability. A corresponding hierarchical weight coefficient is configured for each dimension. Then, based on the target puncture point of the part and the part cutting access order, a standard candidate path with a virtual start node is generated. The generalized cost matrix is ​​called to calculate the total empty travel cost of the path, while simultaneously verifying the constraint compliance and puncture point process adaptability, obtaining evaluation values ​​for each of the three dimensions. Then, the evaluation values ​​of the three dimensions are weighted and summed according to the hierarchical weight coefficients to obtain the overall fitness of the individual. Finally, the overall fitness of all individuals in the coded population is sorted in descending order, and the individual with the highest ranking is selected as the elite individual. This method has comprehensive evaluation dimensions, clear weight hierarchy, and can flexibly adjust the priority of different dimensions, adapting to multi-objective optimization processing scenarios.

[0055] The second approach is a constraint-prioritized progressive fitness evaluation method. First, it performs compliance checks on processing priority constraints, dividing the population into compliant and non-compliant individuals. A base fitness level is assigned to compliant individuals, while a punitive fitness level is assigned to non-compliant individuals, lower than that of compliant individuals. Then, for compliant individuals, based on the generalized cost matrix, the target puncture point of the part, and the part cutting access order, the total cost of the empty travel of the path is calculated. The fitness increment is calculated based on the total empty travel cost and added to the base fitness level to obtain the individual's overall fitness. For non-compliant individuals, the degree of constraint violation is calculated. The fitness deduction is calculated based on the degree of violation and added to the punitive fitness level to obtain the individual's overall fitness. Finally, the overall fitness of all individuals in the encoded population is ranked, and the individual with the highest overall fitness is selected as the elite individual. This method, with processing priority constraints as the core evaluation premise, prioritizes path compliance, avoiding the generation of paths that violate process requirements, and is suitable for precision parts processing scenarios with extremely high compliance requirements for processing rules.

[0056] The third method is a dynamic threshold-based elite retention evaluation approach. First, based on the generalized cost matrix, the target puncture point of the part, and the part cutting access order, the total path cost and constraint compliance of each individual in the population are calculated. The overall fitness of each individual is calculated by multiplying the inverse of the total path cost by the constraint compliance coefficient. Then, the average and standard deviation of the overall fitness of all individuals in the encoded population are calculated, and a dynamic elite selection threshold is set. Individuals with an overall fitness higher than the threshold are selected as candidate elites. A second verification is then performed on the path stability and process adaptability of the candidate elites. Finally, the individual with the highest overall fitness among the candidate elites that pass the second verification is selected as the final elite. This method uses a dynamic threshold to select elites, avoiding interference from accidental optimal solutions in a single iteration, improving the stability and reliability of elites, and adapting to path optimization scenarios involving large-scale populations and complex multi-contour parts.

[0057] In an exemplary scheme for assessing overall fitness and identifying elite individuals, the system first loads its preset fitness assessment rules, weight parameters, and elite selection rules. Then, candidate path standardization is performed. For each individual in the population, the corresponding part target puncture point and part cutting access order are extracted. The part target puncture points are arranged according to the cutting access order, and virtual start nodes are added to generate standard candidate paths with virtual nodes. Path alignment is then performed, moving the virtual start nodes to the beginning of the path, and the aligned standard candidate paths are output. Next, total path cost calculation is performed. For the aligned standard candidate paths, the generalized cost matrix is ​​sequentially called to extract the corresponding costs between adjacent nodes according to the path's node access order. The costs are accumulated to obtain the total empty travel cost for the path, and the path information with the total empty travel cost is output. Finally, constraint compliance verification is performed. For the standard candidate path, it is verified whether its cutting access order conforms to the processing priority constraint of inner holes before outer contours. The number and severity of constraint violations are counted, and the corresponding constraint penalty cost is calculated. Simultaneously, it is verified whether the path's target puncture point meets the laser cutting process requirements, and the corresponding process penalty cost is calculated. The path calculation result with penalty items is output. Next, the overall fitness calculation is performed. Based on preset weight parameters, the total cost of the empty journey, the constraint penalty cost, and the process penalty cost are weighted and summed to obtain the overall processing cost of the path. The reciprocal of the overall processing cost is taken as the overall fitness of the individual in the population. After the calculation is completed for all individuals in the population, a set of overall fitness for all individuals is generated. Then, elite individual selection is performed. All individuals in the population are sorted in descending order of overall fitness from high to low. Invalid individuals with overall fitness below a preset threshold are removed. The path validity of the top-ranked individuals is checked a second time to confirm that their paths are interference-free and meet process requirements. Finally, the individuals with the highest overall fitness that pass the second check are selected as elite individuals. At the same time, according to the preset number of elites to be retained, the top-ranked elite individuals are retained and an elite retention strategy is implemented to output the final elite individuals for subsequent local path optimization.

[0058] Step S60: The elite individual is used as the initial solution for local path optimization, and the initial solution is subjected to local path optimization according to the cascaded neighborhood operator to obtain the target cutting path. The neighborhood search intensity of each iteration in the local path optimization is determined according to the iteration progress.

[0059] Neighborhood search strength is a core parameter used to control the search depth, number of iterations, and termination conditions for local path optimization. Higher search strength results in more iterations and a deeper search depth. Cascaded neighborhood operators are a combination of multiple heterogeneous search operators used to perform local transformations on candidate paths and generate new candidate paths. Each operator corresponds to a local path adjustment method, such as node swapping, path reversal, puncture point replacement, and sequence shifting. Local path optimization is a process of performing a refined search within the neighborhood of the optimal solution corresponding to an elite individual, further optimizing path performance and avoiding getting trapped in local extrema. The target cutting path is the final laser cutting execution path generated after global optimization and local optimization, meeting processing requirements, having the shortest idle travel, and achieving the best overall performance.

[0060] In this embodiment, there are three ways to perform local path optimization and obtain the target cutting path using elite individuals as the initial solution. The first is a cascaded neighborhood search method with adaptive intensity control. First, the standard candidate path encapsulated by the elite individual is extracted as the initial solution for local optimization. Based on the current iteration progress of the hybrid heuristic algorithm and the fitness distribution of the individuals in the population, the appropriate neighborhood search intensity and the maximum number of rounds without improvement threshold are dynamically calculated. Then, the execution order of the cascaded neighborhood operators is randomly shuffled, the consecutive unimproved count of the local search is initialized, and the neighborhood search is executed sequentially according to the shuffled operator order. The path is updated using a strategy of exhaustive search within operators and first improvement between operators until the number of consecutive unimproved rounds reaches the threshold corresponding to the configured search intensity. The optimized path is then output. Finally, constraint verification and process verification are performed on the optimized path to output the final target cutting path. This method can adaptively adjust the search intensity according to the algorithm iteration state, balancing the algorithm's optimization accuracy and convergence efficiency, avoiding increased time consumption due to over-search, and is suitable for conventional high-efficiency processing scenarios.

[0061] The second approach is a perturbation-triggered cyclic neighborhood optimization method. First, the standard candidate paths corresponding to elite individuals are used as the initial solutions and the current globally optimal path. The neighborhood search strength and the count of consecutive unimproved iterations are initialized. Neighborhood search is executed in a random order according to cascaded neighborhood operators, updating the current path and the globally optimal path while accumulating the number of consecutive unimproved iterations. When the number of consecutive unimproved iterations reaches the maximum threshold, the current solution is determined to be trapped in a local optimum, triggering a perturbation mechanism. A preset strength of random neighborhood perturbation is applied to the currently recorded globally optimal path, generating a new local search starting path. The consecutive unimproved iteration count is reset, and the neighborhood search is restarted. This process is repeated until the preset maximum number of iterations is reached, finally outputting the globally optimal path as the target cutting path. This method, by using a perturbation mechanism to escape local optima, can significantly improve the algorithm's global optimization capability, avoid getting trapped in local optima, and is suitable for complex, multi-contour, and highly challenging path optimization scenarios.

[0062] The third approach is a progressive neighborhood optimization method that combines global and local optimization. First, elite individuals are used as initial solutions for local optimization, while several second-best elite individuals ranked high in the population are retained to construct an elite solution set. A high-intensity neighborhood search is performed on the initial solution in the first stage, using cascaded neighborhood operators to perform a full neighborhood transformation and explore the optimization potential of the initial solution. If the path performance does not significantly improve after the search, the process switches to a second-stage linked search. Path segments from second-best elite individuals are randomly selected from the elite solution set and replaced with the corresponding path segments of the current initial solution to generate a new optimized initial solution. Then, a low-intensity refined neighborhood search is performed, and this progressive optimization is repeated until a preset termination condition is met. Finally, the path with the best performance is output as the target cutting path. This method, through the linkage of the global elite solution set and the local refined search, balances global optimization capability with local optimization accuracy, fully exploring the optimization potential of all elite individuals, and is suitable for high-end machining scenarios with ultra-complex multi-part and large-scale nested contours.

[0063] In an exemplary scheme for performing local path optimization to obtain the target cutting path, the system first loads a set of cascaded neighborhood operators, neighborhood search intensity control rules, perturbation mechanism parameters, and iteration termination conditions. The set of cascaded neighborhood operators includes four types of heterogeneous neighborhood operators: node swapping operators, path reversal operators, puncture point replacement operators, and sequence shift operators. The neighborhood search intensity control rules include an adaptive calculation formula for search intensity, a maximum no-improvement round threshold, and perturbation intensity configuration. Then, initial solution initialization is performed, extracting the standard candidate path encapsulated by elite individuals, the corresponding target puncture point of the part, and the part cutting access order, using these as the initial solution for local path optimization. Simultaneously, the comprehensive fitness and total empty travel cost corresponding to this initial solution are recorded as benchmark parameters for the current globally optimal path. Next, adaptive neighborhood search intensity control is performed, obtaining the current iteration count, total iteration count, and fitness variance of the individual population in the hybrid heuristic algorithm. According to the control formula, the neighborhood search intensity and the maximum no-improvement round threshold corresponding to this local search are dynamically calculated, and the optimization task with intensity configuration is output. Subsequently, cascaded neighborhood operator preprocessing is performed. The execution order of the four types of heterogeneous cascaded neighborhood operators is randomly shuffled, the consecutive unimproved count of the local search is initialized, and the current optimized path is set as the initial solution. Next, cascaded neighborhood search is performed. Following the shuffled operator order, the corresponding neighborhood operators are sequentially called to perform neighborhood transformations on the current optimized path, generating multiple candidate neighborhood paths. For a single operator, an exhaustive optimization strategy is used to select the candidate neighborhood path with the best performance after the operator's transformation. If the overall fitness of this path is better than the current optimized path, the current optimized path and the globally optimal path are updated, the consecutive unimproved count is reset, and the process switches to the next operator to execute the first improvement strategy. If no better path is found for this operator, the consecutive unimproved count is accumulated, and the process switches to the next operator. Subsequently, local extremum verification is performed. When the number of consecutive unimproved iterations reaches the configured maximum threshold for no improvement, the current solution is determined to be trapped in a local extremum. A preset perturbation mechanism is immediately triggered, retrieving preset perturbation strength parameters. Unique neighborhood operators are randomly selected from the neighborhood operator set, and neighborhood transformation operations corresponding to the perturbation strength are continuously performed on the currently recorded global optimal path. Perturbed candidate paths are generated, their constraint compliance is verified, and they are used as the new starting path for the local search. The consecutive unimproved iteration count is reset, and the neighborhood search is restarted. Finally, iteration termination verification is performed. When the number of iterations in the local search reaches the preset maximum number of iterations or the path optimization magnitude is lower than the preset convergence threshold, the local optimization process is terminated. Processing priority constraint verification, process interference verification, and path validity verification are performed on the final global optimal path. After passing all verifications, the path is output as the final target cutting path for laser cutting machine tool processing.

[0064] Second Embodiment This embodiment provides an exemplary scheme for the geometric contour analysis and node cluster generation of a part to be processed. In this example, the entity data segment of the layout graphic file of the part to be processed is first parsed in response to the cutting command, the geometric primitive feature parameters are extracted and the primitives are spliced ​​to obtain the initial contour, and then the initial contour is subjected to normalization preprocessing to obtain standard geometric contour data. Finally, the node cluster division and candidate puncture node screening are completed through discretization processing, and the basic node data required for path planning is output. Step S10 includes steps A11 to A14: Step A11: Based on the graphic file corresponding to the layout of the parts to be processed, locate and parse the entity data segments in the graphic file, and determine the characteristic parameters of various geometric elements.

[0065] Step A12: Perform a primitive splicing operation on the feature parameters of the geometric primitives to obtain the initial contour of the part to be processed.

[0066] Step A13: Perform normalization preprocessing on the initial contour, remove redundant contour points and correct minor contour deviations to obtain the geometric contour data of the part to be processed.

[0067] Step A14: Discretize the geometric contour data of the part to be processed, divide the contour of a single part into independent node clusters, and determine the feasible candidate puncture nodes within the node clusters.

[0068] Geometric primitives are the basic vector graphic units that constitute the contour of a part to be processed, and are the smallest processing units for contour analysis and splicing. Examples include line segment primitives, arc segment primitives, elliptical arc segment primitives, spline curve segment primitives, circle primitives, and rectangle primitives. Feature parameters are the core set of parameters used to uniquely characterize the geometric attributes, spatial position, and topological relationships of geometric primitives, and are the core basis for primitive splicing and contour construction. Examples include the start and end coordinates of a line segment, the center coordinates, radius, and start and end angles of an arc segment, the control point coordinates of a spline curve, and the layer attributes and processing type markers of the primitives. The initial contour is the contour path of the part to be processed, formed by splicing geometric primitives and possessing a closed topological structure; it is the input object for contour normalization preprocessing. Examples include the closed contour of a single inner hole, the closed path of a single outer contour, and nested multi-contour combination structures.

[0069] In this example, when parsing the entity data segments of a graphic file to extract geometric primitive feature parameters, the method of contour layer-based pre-scanning full-volume parsing can be used. Alternatively, a streaming incremental parsing method can be employed, where the entity data segments of the graphic file are read while the feature parameters of the geometric primitives are parsed segment by segment, and the topological connections between primitives are identified simultaneously. This accumulates to obtain a complete set of feature parameters for all geometric primitives, thereby completing the acquisition of geometric primitive feature parameters.

[0070] After acquiring the geometric primitive feature parameters, the primitive stitching process is initiated. For all geometric primitives, the overlap and topological connectivity of the primitive endpoints are identified sequentially, and adjacent primitives are stitched and closed to obtain the initial contour of the part to be processed. Subsequently, a normalization preprocessing is performed on the initial contour to eliminate contour defects and redundant data, resulting in standardized geometric contour data. Next, discrete sampling and clustering processing are performed on the geometric contour data to divide node clusters and screen candidate puncture nodes. This hierarchical contour analysis and preprocessing process improves the accuracy of the contour data and the precision of node clustering, avoiding subsequent path planning deviations and processing interference caused by contour analysis errors.

[0071] For example, there are two ways to obtain the initial contour of the part to be processed by splicing geometric primitive feature parameters. The first is a topology-locked segment-by-segment serial splicing and breakpoint-based dynamic closure method. According to the original drawing sequence bound to the geometric primitive feature parameters, starting from the beginning of the feature sequence, two adjacent geometric primitives in the sequence are selected sequentially to perform endpoint overlap comparison. After each set of adjacent primitives is compared, it is simultaneously determined whether the corresponding primitives in the set meet the preset topology connection requirements. If they meet the requirements, the corresponding primitives are spliced ​​and included in the current continuous contour segment, and the node sequence of the continuous contour is continuously accumulated. If they do not meet the requirements, the splicing process of the current continuous contour segment is terminated, and the determination and splicing of the next continuous contour segment begins. After all adjacent primitive pairs have been compared and spliced, breakpoint closure verification is performed on all continuous contour segments, and the contour segments whose endpoint overlap meets the closure requirements are connected end to end to generate a closed initial contour. This method employs a time-locked, pairwise serial comparison and a breakpoint-based dynamic accumulation calculation logic. By sequentially splicing and verifying the closure in accordance with the original drawing timing, it avoids the problem of mis-splitting of non-associated primitives and restores the actual topological structure of the outline of the part to be processed.

[0072] The second method involves feature clustering for contour interval division and parallel stitching across all intervals. This method reads and analyzes the feature parameters of all geometric primitives, then performs clustering based on the overall distribution of spatial coordinates within these parameters. Geometric primitives located within the same preset interval, with consistent layer attributes and processing types are grouped into the same contour feature interval. Non-continuous boundary nodes between intervals are marked, generating multiple contour feature intervals that are computationally independent. Parallel stitching is then initiated simultaneously for all divided contour feature intervals. Within each interval, the endpoint overlap and topological connectivity of all adjacent primitives are compared, completing the stitching and closure of all primitives within the interval to generate closed contour segments. After all parallel stitching and closure of contour feature intervals are completed, all closed contour segments are aggregated, and cross-interval nested level verification and contour deduplication are performed to finally generate the initial contour of the part to be processed. This method employs a computational logic of global clustering interval partitioning and parallel verification of the entire interval. By pre-emptively clustering global spatial features, it locks the potential contour interval of the parts to be processed in advance, reducing the number of invalid adjacent primitive comparisons and improving the contour splicing efficiency of large-size, multi-part layout graphics.

[0073] Third Embodiment This embodiment provides an exemplary scheme for constructing a generalized cost matrix for laser cutting paths. In this example, candidate puncture nodes within a node cluster are first used as processing nodes, and processing priority constraints are embedded to construct an augmented graph base topology. Then, virtual starting nodes are added to complete the unified modeling of multiple processing modes to obtain an augmented graph model. Subsequently, based on the augmented graph model, the idle travel cost and return travel cost between nodes are calculated to obtain the node configuration cost. Finally, all node configuration costs are integrated into a standardized generalized cost matrix. Step S20 includes steps B11 to B14: Step B11: Take the candidate puncture nodes in the node cluster as processing nodes, embed the sequential processing association corresponding to the processing priority constraint, and obtain the augmented graph basic topology.

[0074] Step B12: Add the virtual starting node to the node set of the augmented graph base topology, and model it uniformly according to the processing mode to obtain the augmented graph model.

[0075] Step B13: Based on the augmented graph model, configure the idle travel cost between processing nodes and the return travel cost corresponding to the virtual starting node according to the processing mode to obtain the node configuration cost.

[0076] Step B14: Arrange the node configuration costs between all nodes into a unified structured generalized cost matrix according to the arrangement order of the nodes in the augmented graph model.

[0077] Actual machining nodes are candidate puncture nodes within a node cluster that can be used for actual laser head puncture and cutting. They are the core foundational nodes for augmented graph topology construction and correspond to actual machining points on the contour of the part to be processed. Examples include candidate cutting nodes on the inner hole contour, candidate puncture nodes on the outer contour, and discrete contour nodes that meet the requirements of laser processing technology.

[0078] The processing priority constraint, corresponding to the sequential processing association rule, is a mandatory rule used to limit the legality of directed connections between nodes in the augmented graph. Its core principle is that high-priority inner hole nodes can only point to low-priority outer contour nodes in a unidirectional association, while nodes of the same priority can have bidirectional associations. This is used to avoid illegal paths that violate processing rules at the topology level. Examples include legal directed connections from inner hole nodes to outer contour nodes, illegal reverse connections from outer contour nodes to inner hole nodes, and legal bidirectional connections between inner hole nodes at the same level.

[0079] The augmented graph's basic topology is a directed graph structure composed of actual machining nodes and directed connections that conform to machining priority constraints. It serves as the foundational framework for subsequently adding virtual nodes and completing multi-modal modeling. For example, a directed connection topology containing only actual machining nodes for the inner hole and outer contour, or a node connection graph conforming to the constraint that the inner hole is machined before the outer contour.

[0080] A virtual start node is a virtual node added to the augmented graph to unify the processing modes of closed-loop reset and open-loop non-reset. It represents the processing start and reset positions of the laser head, enabling unified path modeling for different processing modes. Examples include a start and reset dual-connection virtual node in closed-loop mode and a start-connection only virtual node in open-loop mode.

[0081] An augmented graph model is a standardized directed graph model obtained by adding virtual start nodes to the basic topology of an augmented graph and completing the processing mode adaptation modeling. It is the core carrier for node configuration cost calculation and generalized cost matrix construction. For example, there is a directed graph model with virtual nodes adapted to the closed-loop reset processing mode, and a directed graph model with virtual nodes adapted to the open-loop non-reset processing mode.

[0082] Idle travel cost is the comprehensive cost corresponding to the movement of the laser head between two actual processing nodes without cutting action. It is a core component of node configuration cost and is positively correlated with the length of the idle travel and the movement time. Examples include the idle travel cost between two inner hole nodes, the idle travel cost from an inner hole node to an outer contour node, and the time cost of rapid laser head movement.

[0083] The return cost is the movement cost of the laser head returning from the final machining node to the virtual starting node after completing all machining operations in closed-loop reset machining mode. It only applies in closed-loop machining mode. Examples include the movement cost of returning from the final outer contour machining node to the machine tool origin, and the travel cost of the laser head resetting after machining.

[0084] The node configuration cost is the comprehensive cost corresponding to a directed connection between any two nodes in the augmented graph model. It includes the cost of empty trips, return trips, and penalty costs for violating processing priority constraints. It is the smallest constituent unit of the generalized cost matrix.

[0085] In this example, when constructing the basic topology of the augmented graph, it can be done using a static topology construction method with constraints preceding the construction. Alternatively, it can be done through dynamic topology generation, simultaneously reading node clusters and processing priority constraint parameters while configuring valid directed connections for each node, thus simultaneously completing the construction and validity verification of the basic topology of the augmented graph, thereby completing the construction of the basic topology of the augmented graph.

[0086] After completing the construction of the augmented graph's basic topology, the augmented graph model construction process is initiated. Virtual starting nodes are added to the node set of the augmented graph's basic topology. According to the currently selected processing mode, directed connections between the virtual starting node and all real processing nodes are configured, completing the unified modeling of closed-loop and open-loop processing modes, resulting in a standardized augmented graph model. Subsequently, based on the augmented graph model, the empty journey cost, start-up cost, and return cost are calculated for each effective directed connection between nodes, obtaining the node configuration cost for all nodes. Finally, according to the preset arrangement of nodes, all node configuration costs are integrated into a structured generalized cost matrix. This hierarchical topology construction and cost calculation process improves the multi-mode adaptability and cost calculation accuracy of the generalized cost matrix, avoiding subsequent path optimization deviations caused by unreasonable topology modeling.

[0087] For example, there are two ways to calculate node configuration costs and construct a generalized cost matrix based on an augmented graph model. The first is a row-by-row serial calculation and sequential matrix filling method with node locking. According to the preset arrangement order of nodes in the augmented graph model, starting from the first row of the matrix, the starting node corresponding to the row and the target node corresponding to the column are selected sequentially, and the directed connection validity check and cost calculation are performed on each node pair. After each set of node pairs is checked and calculated, the calculated node configuration cost is simultaneously filled into the corresponding row and column positions of the matrix. If the directed connection of the node pair meets the processing priority constraint, the calculated empty trip cost or return trip cost is filled in. If the directed connection of the node pair violates the constraint, a preset high penalty cost is filled in. After all node pairs have been calculated and filled, a complete standardized generalized cost matrix is ​​generated. This method employs a time-locked, pairwise serial verification and sequential matrix filling calculation logic. By performing successive verification and filling in accordance with the node arrangement order, it avoids matrix structure errors caused by node row and column misalignment, thus ensuring the structural accuracy of the generalized cost matrix and the completeness of cost calculation.

[0088] The second approach involves priority-based interval partitioning and parallel computation across all intervals. This method reads and performs priority-based hierarchical parsing on all nodes in the augmented graph model. Based on the processing priority level of the node cluster to which a node belongs, hierarchical processing is performed, dividing nodes of the same priority into the same interval and nodes of different priorities into different intervals. The legal connection directions between intervals are marked, generating multiple independent node computation intervals. Parallel computation is then initiated synchronously for all partitioned node computation intervals. Within each interval, the bidirectional connection cost of all node pairs is calculated. Between different intervals, the unidirectional connection cost of cross-interval node pairs is calculated according to the legal connection directions, and the corresponding matrix blocks are filled synchronously. After all parallel computation and filling of all node computation intervals are completed, all matrix blocks are aggregated, and the cost of the corresponding rows and columns of the virtual nodes is filled, ultimately generating a complete standardized generalized cost matrix. This method employs a computational logic of priority-based hierarchical interval partitioning and full-interval parallel verification. By prioritizing the processing priority, the legal connection range between nodes is locked in advance, reducing the number of invalid and illegal connection verifications and improving the efficiency of constructing the generalized cost matrix for complex parts with multiple nodes and multiple contours.

[0089] Further, please refer to Figure 2 , Figure 2 This is a schematic diagram of the augmented graph model of this application. Two processing modes: Mode=Closed→c i,0 =Dist(i,0): The cost c in closed-loop mode from node i to the virtual starting point (Depot, i.e., point 0). i,0= Actual distance from node i to the virtual starting point. Meaning: In closed-loop machining mode, after the laser cutting head finishes machining, it needs to return to the machine tool origin (virtual starting point). Therefore, the cost of this return trip needs to be calculated based on the actual distance and included in the total path cost.

[0090] Mode=Open→c i,0 =0: Open-loop mode → cost c from node i to the virtual starting point i,0 =0. Meaning: In open-loop processing mode, the laser cutting head does not need to return to the machine tool origin after processing, so the cost of the return trip is directly recorded as 0, and no additional path cost is added.

[0091] The laser cutting path planning problem is modeled as a Precedence-Constrained Generalized Traveling Salesman Problem (PCGTSP). The cluster node structure of the GTSP adapts to the constraint of selecting the optimal puncture point for each part, while the priority constraint characterizes the subsequent cutting process sequence requirements. To clearly define the core elements of the model, the specific modeling rules are set as follows: to be processed The outline of each part is set as There are node clusters, where each feasible puncture point on the contour of a part is a node within the corresponding cluster. Let the set of real part clusters be denoted as . , No. The set of candidate perforation points for each part is as follows It should be noted that, to uniformly characterize different processing scenarios (such as whether the process needs to be reset to the starting position after completion, whether the path needs to be closed, etc.), this solution further introduces the concepts of virtual nodes (Depot) and augmented graphs. This setting enables standardized descriptions of various scenarios, improving the model's versatility. Virtual starting nodes constitute a set of virtual starting points. For standard GTSP scenarios without a fixed starting point, Set up all real processing nodes With the virtual starting point set Merging nodes to form the set of nodes in the augmented graph And define the directed connections between nodes as a set of arcs. The process requirement of "cutting the inner hole first, then the outer contour" is extracted as a priority constraint set. If satisfied , then it indicates the part For parts The inner hole and prior to the part Cutting. Based on the node set. Arc set and priority constraint set Constructing an augmented graph mathematical topology for the generalized traveling salesman problem with priority constraints. .

[0092] For any two real processing nodes in the augmented graph Define its generalized cost Let Euclidean distance be the idle travel distance of the laser head between two points. This applies to any real machining node in the augmented graph. Return to the virtual starting node Directed arc The return cost is dynamically configured based on the received processing mode instructions. If the processing mode command is closed-loop reset mode, set ,in For actual processing nodes The Euclidean distance to the virtual starting node. If the machining mode command is open-loop non-reset mode, force setting. The generalized cost With dynamic configuration of return cost Together, we can construct the final generalized cost matrix. This eliminates the interference of backhaul cost on algorithm optimization under a unified augmented graph topology.

[0093] The PCGTSP model is established as follows:

[0094] The constraints are as follows:

[0095]

[0096]

[0097]

[0098]

[0099]

[0100]

[0101]

[0102] The optimization objective of the PCGTSP model established above is to find an optimal path that allows the laser head to start from a preset starting position (a specific starting point or the initial node of the first node cluster), traverse all node clusters and visit each node cluster only once, while satisfying the cutting process sequence constraint of "inner hole first, then outer contour", and finally complete all processing tasks.

[0103] This invention sets the following constraints to ensure that the laser cutting path planning results meet the requirements of processing technology, path topology validity, and global traversal: Constraint 1: Cluster Selection Constraint limits each real part to a node cluster and allows only one candidate puncture point to be selected as the actual machining puncture point for that part. This ensures that each part undergoes only one laser puncture feed, matching the basic process requirements of laser cutting. Constraint 2: Flow Conservation Constraint is set for all nodes in the augmented graph model. It requires that the selected puncture point node has a unique entry and exit path. Unselected nodes do not participate in path topology construction, thus ensuring the continuity and uniqueness of the laser head's movement path and avoiding path interruptions or invalid jumps. Constraint 3: Virtual Starting Point Constraint is set for application scenarios with a fixed machining starting point. It forces the virtual starting point node corresponding to the initial dwell position of the laser head to be selected, ensuring that the machining path always starts from the preset machine tool origin, adapting to the standardized machining process of the machine tool. Constraint 4: Priority Constraint is set based on the priority constraint set generated by the nesting relationship between parts. It forces the machining access position of the part as an inner hole to be earlier than its corresponding outer contour part, thus satisfying the "inner-to-outer" laser cutting process requirement and avoiding the problem of decreased machining accuracy caused by workpiece detachment or thermal deformation during processing. Constraint 5: The sub-loop elimination constraint is constructed using the Miller-Tucker-Zemlin (MTZ) constraint. By associating path decision variables with part processing position auxiliary variables, isolated sub-loops in the path are completely eliminated, ensuring that the planned single path can completely traverse all parts to be processed and avoiding the risk of missed processing. Constraint 6: The variable domain constraint explicitly defines all path decision variables and puncture point selection variables as binary variables. The value range of the part processing position auxiliary variable matches the total number of parts to be processed, ensuring the discreteness of the model solution and the validity of the output results.

[0104] The Slime Mould Algorithm - Variable Neighborhood Search (SMA-VNS) is a hybrid heuristic algorithm.

[0105] Since the aforementioned mathematical model is a non-deterministic polynomial-hard problem, this embodiment employs an improved heuristic algorithm for solving it. The SMA-VNS algorithm combines the global search capability of the slime mold algorithm with the local optimization capability of the variable neighborhood search. During the SMA iteration process, VNS is periodically applied to the elite individuals in the population for local search, and the improved solution is fed back into the population to participate in subsequent evolution.

[0106] During the initialization phase, the specific discrete mapping logic of the two-segment continuous variable encoding scheme is as follows: Let the total number of node clusters be... The position vector of each individual in the population in the hybrid heuristic algorithm Divided into two segments, the two-segment decoding process is as follows: Dimensional variables Used to map the selection of candidate nodes within each node cluster, for the A cluster of nodes, of which It is known that it contains the following set of feasible candidate nodes: Let the total number of candidate nodes in this set be . Then the index of the actual puncture point selected for this cluster is ,in This represents the floor function. Dimensional variables The cutting and access order of node clusters is determined by sorting their numerical values ​​in ascending order and extracting the original array indices as the cutting and access order of the node clusters. .

[0107] The system generates an initial discrete path, calculates the total empty travel distance of the current individual decoding path, and before constraint checking, if a virtual starting node exists in the current augmented graph model, performs path rotation processing on the decoding path, forcing the virtual starting node to move to the beginning of the path to ensure global consistency of the circular path constraint check. This is based on the aforementioned priority constraint set. A hierarchical penalty function method is used to handle the constraints on the cutting order of the inner and outer contours of parts with holes and the legality of node cluster access.

[0108] Define the decoding path after rotation processing. The original total empty travel distance is The penalty for violating the restrictions is Then the comprehensive fitness function with penalty Defined as:

[0109] Among them, the constraint violation penalty item The grading calculation formula is as follows:

[0110] in, Represents a cluster of nodes Prior to node clusters Visited; and Representing node clusters respectively and In the path The access position in the middle, This indicates that the node cluster is missing in the path; The default penalty coefficient for missing clusters is... The predefined constraint violation penalty coefficient is used, and the following conditions are met: This is to ensure that no processing features are overlooked.

[0111] The global evolution update phase specifically employs a slime mold optimization mechanism, assuming the current iteration number is... The maximum number of iterations is Population size is , No. The current position of each individual is Generate random numbers Set the preset exploration probability And based on the current overall fitness of the individual. Calculate extreme value control parameters The formula for updating the individual position is:

[0112] in, Let i be the position of the i-th individual in the next iteration. For random locations within the search space, The current optimal individual position, and For two randomly selected individual locations; when ( To preset the exploration probability, the preferred method is... When an individual explores the search space randomly, it generates a new location. To ensure overall population diversity; when ( These are extreme value control parameters. When ), the individual moves closer to the current optimal solution; when At that time, the individual performs a contraction exploration; In this process, the overall fitness Weights calculated in ascending order for:

[0113] Adaptive oscillation parameters Shrinkage parameters Shrinkage parameters b= Using the inverse hyperbolic tangent function ( The nonlinear decay characteristics of the algorithm enable a smooth transition between large-scale exploration in the early stage of iteration and high-precision convergence in the later stage.

[0114] Variable neighborhood search is a local search metaheuristic algorithm based on systematic neighborhood transformation. In this invention, the algorithm employs an exhaustive search strategy, traversing all possible moves within each neighborhood structure to find the optimal improved solution. When no improvement can be found in any neighborhood, a shaking mechanism is used to escape the local optimum, and the search continues.

[0115] To balance exploration and development, the search depth of VNS (maximum number of rounds without improvement) It adaptively adjusts as the iteration progresses, and its calculation formula is:

[0116] in, The preset initial minimum search depth, The preset maximum search depth, This represents the current iteration number. The maximum number of iterations, This represents the floor function; By dynamically allocating local optimization computing power through the above adaptive adjustment formula, the algorithm maintains a low search intensity in the early stage of iteration to quickly explore and maintain population diversity, and allocates a higher search intensity in the later stage of iteration to refine and deeply explore the local optimal solutions of elite individuals.

[0117] Fourth embodiment This embodiment provides an exemplary scheme for layered decoding and mapping of two-segment variables in a laser cutting path. In this example, the first segment of continuous variables for the encoded population is first discretely mapped to generate continuous node values ​​corresponding to node clusters. Then, the continuous node values ​​are converted into candidate puncture point indices to determine the target puncture point. Subsequently, the second segment of continuous variables is sorted and mapped to obtain the node cluster access order. Finally, the access order is sorted according to the variable values ​​to accurately determine the overall cutting access order of the part. Step S40 includes steps C11~C14: Step C11: Discretize the first segment of continuous variables of the individuals in the encoded population to obtain the continuous values ​​of the nodes corresponding to the node cluster.

[0118] Step C12: Convert the continuous values ​​of the nodes into candidate puncture point indices within the node cluster to determine the target puncture point of the part to be processed.

[0119] Step C13: Sort and map the second continuous variable of the individuals in the encoded population to obtain the access order corresponding to the node cluster.

[0120] Step C14: Sort the access order of the node cluster according to the numerical value of the second continuous variable to determine the cutting access order of the part to be processed.

[0121] Discrete mapping is the process of converting the continuous values ​​in the 0-1 interval of the first continuous variable encoding into integer indices of candidate puncture nodes within the corresponding node cluster. It is the core decoding step that transforms continuous encoding into actual puncture point selection results. For example, it involves linearly mapping the continuous values ​​in the 0-1 interval to the index of the candidate puncture node within the node cluster, matching the mapping rules of the corresponding candidate puncture nodes based on the magnitude of the continuous values, and using a dynamic interval mapping method that adapts to the number of candidate nodes within the node cluster.

[0122] Sorting mapping is the process of sorting the continuous values ​​in the 0-1 interval of the second continuous variable encoding according to their numerical value, and converting them into the access order of the corresponding node clusters. It is the core decoding step that transforms continuous encoding into the actual cutting and access order. For example, there are mapping rules for sorting continuous values ​​in ascending order to obtain the access order of node clusters, sorting them in descending order to obtain the processing order of node clusters, and hierarchical sorting mapping methods that adapt to processing priority constraints.

[0123] In this example, when matching two-segment continuous variable codes to individuals in the population to generate the coded population, a cluster-aligned synchronous coding method can be used. Alternatively, a priority-based hierarchical dynamic coding method can be employed. First, the node clusters are divided into different priority groups based on processing priority constraints. Corresponding coding segments of different lengths are configured for each priority group. Then, hierarchical two-segment continuous variable codes are generated for each individual in the population, and the constraint compliance verification of the coding is completed synchronously, thereby completing the initial generation of the coded population.

[0124] After initializing and generating the encoded population, a two-segment decoding process is initiated. For each individual in the encoded population, the first segment of continuous variables is extracted and subjected to discrete mapping processing to obtain the candidate puncture node indices within each node cluster, thus determining the target puncture point corresponding to each node cluster. Then, the second segment of continuous variables is extracted and subjected to sorting mapping processing to determine the access order of each node cluster based on its numerical values, obtaining the cutting access order of the part to be processed. This decoupled two-segment encoding and hierarchical decoding process simultaneously achieves the coordinated optimization of the two core objectives: puncture point selection and access order planning. This avoids the limitation of single-segment encoding in failing to consider both optimization objectives, improving the global optimization efficiency and solution accuracy of the hybrid heuristic algorithm.

[0125] For example, there are two ways to obtain the target puncture point and cutting access order through two-segment continuous variable encoding and decoding. The first is a cluster-aligned segment-by-segment serial decoding and constraint verification method. Based on the one-to-one correspondence time sequence of the node clusters bound by the two-segment continuous variable encoding, starting from the beginning of the encoding sequence, the continuous variables corresponding to the node clusters in the first segment of continuous variables are extracted sequentially. Discrete mapping processing is performed on each cluster to convert the continuous values ​​into candidate puncture node indices within the corresponding node cluster, thus determining the target puncture point of that node cluster. After decoding the puncture points of all node clusters, the continuous variables in the second segment of continuous variables are extracted sequentially, and sorting mapping is performed in ascending order of values ​​to obtain the access position corresponding to each node cluster. The initial cutting access order is generated according to the position order. Then, it is verified whether the initial cutting access order conforms to the processing priority constraint. Positions that violate the constraint are corrected, and finally, the compliant target puncture point and cutting access order are output. This method employs a segmented serial decoding and post-constraint verification processing logic with cluster alignment. By mapping the encoding bits to the node clusters in a one-to-one correspondence, it avoids the problem of mismatch between encoding and node clusters, ensuring the accuracy of puncture point selection and access order decoding. The decoding logic is stable and controllable, and it is suitable for the processing scenarios of parts with a fixed number of node clusters.

[0126] The second approach involves priority-based interval partitioning and parallel decoding across all intervals. This method performs priority-based parsing of the two-segment continuous variable encoding sequence. Based on the processing priority level of node clusters, both the first and second segments of the encoding sequence are divided into independent encoding intervals corresponding to priority groups. The mapping relationship between each encoding interval and its corresponding node cluster group is marked, generating multiple decoding intervals that are computationally independent of each other. Parallel decoding is then initiated synchronously for all partitioned decoding intervals. Within each decoding interval, the discrete mapping of the first segment of continuous variables is completed simultaneously to determine the target puncture points of all node clusters within the interval. The sorting mapping of the second segment of continuous variables is also completed synchronously to determine the internal access order of all node clusters within the interval. After all parallel decoding of all decoding intervals is completed, the node cluster access order of each interval is concatenated according to the processing priority level from high to low to generate a complete cutting access order. Finally, the target puncture points of all clusters and the compliant cutting access order are output. This method employs a priority-based hierarchical interval partitioning and full-interval parallel decoding processing logic. By pre-leveling the processing priority, processing priority constraints are directly embedded into the decoding process, thereby preventing the generation of invalid access sequences that violate constraints from the source and reducing subsequent constraint correction operations. At the same time, parallel decoding improves the decoding efficiency of large-scale node clusters and adapts to complex part processing scenarios with multiple nesting and multiple priority groups.

[0127] Further, please refer to Figure 3 , Figure 3 The flowchart below shows the solution logic of the hybrid algorithm in this application. Algorithm 1: SMA-VNS hybrid algorithm, input: population size N, maximum number of iterations T_max, elite ratio ρ, VNS execution interval interval, search depth range [L_min, L_max].

[0128] Output: Optimal solution s*, optimal fitness f* 1: Initialize the population P = {X_1, X_2, ..., X_N} 2: for i = 1 to N do 3: Decode X_i to obtain the path tour_i 4: Evaluate fitness f_i ← f(tour_i) 5:end for 6: Record the globally optimal s*, f* 7: for t = 1 to T_max do 8: / / SMA Location Update 9: Calculate the weights W and the adaptive parameters a, b 10: for i = 1 to N do 11: Update position X_i according to SMA rules 12: Decode the path tour_i 13: Evaluate fitness f_i 14:if f_i <f* then 15:s* ← tour_i, f* ← f_i 16:end if 17:end for 18: / / Populations sorted by fitness 19: Sort the population in ascending order by f_i 20: / / VNS local search (periodic execution) 21: if t mod interval = 0 then 22:n_elite ← max(1, N × ρ ) 23:L ← L_min + (t / T_max) × (L_max - L_min)

[0129] 24:for i = 1 to n_elite do 25:tour ← The path of the i-th elite individual 26:(tour', f') ← VNS(tour, L) 27:if f' <f_i then 28: Update the path and fitness of the i-th individual. 29: Encode tour' back into a position vector 30:if f' <f* then 31:s* ← tour', f* ← f' 32:end if 33:end if 34:end for 35:end if 36: Record the convergence curve 37: end for 38: return s*, f* VNS is applied only to the top ρ proportion of elite individuals in the population based on fitness, to balance computational efficiency and optimization effectiveness. The number of elite individuals is:

[0130] A typical parameter is set to ρ=0.1, which means performing a local search on the top 10% of individuals.

[0131] In this method, four complementary neighborhood structures—node replacement, swapping, 2-opt, and insertion—are designed to assist VNS search based on the characteristics of the PCGTSP model. The details are as follows: The intra-cluster node replacement operator is used to traverse candidate puncture points within a part cluster to optimize intra-cluster node selection; The swap operator is used to swap the access order of two part clusters in a path; The 2-opt operator is used to reverse the subpaths between two part clusters in a path to eliminate intersections of physical cut paths; The insertion operator is used to move a single cluster of parts in a path to another access position; A local search is performed on the current cutting path corresponding to the elite individual. In each round of local search, the access order of the four neighborhood operators is randomly shuffled. A hybrid search strategy of exhaustive search within operators and first-time improvement between operators is adopted: a global exhaustive traversal is performed in the currently selected neighborhood operators to search for the best improved solution that reduces the individual's overall fitness the most within the neighborhood; if the operator successfully outputs the best improved solution, the solution is immediately accepted, the current round of search for other operators is terminated, and the operator access order is shuffled again to start a new round of local search. The local search algorithm flow is as follows: Algorithm 2: Local Search Input: Current solution s, current fitness f Output: Local optimal solution s*, local optimal fitness f* 1:s* ← s 2:f* ← f 3:improved ← true 4:while improved do 5:improved ← false 6: / / Randomly shuffle the order of neighboring domains 7:order ← RandomPermutation({1, 2, 3, 4}) 8: for k = 1 to 4 do 9:op ← order[k] 10: / / Exhaustive search in the neighborhood N_op 11:(s', f', found) ← ExhaustiveSearch(s*, f*, op) 12: if found then 13:s* ← s' 14:f* ← f' 15:improved ← true、 16: break / / Find improvements and start over 17:end if 18:end for 19: end while 20: return s*, f* If the number of consecutive unimproved rounds in the local search reaches the maximum number of unimproved rounds L(t) and no improved solution is found, then the current solution is determined to be trapped in a local optimum, and a perturbation mechanism is triggered: the operation of randomly selecting a neighborhood operator is performed σ times on the currently recorded global optimal path to generate a new local search starting path, where σ is a preset perturbation strength. The perturbation algorithm flow is as follows: Algorithm 3: Shake Input: Current solution s, disturbance intensity σ Output: The perturbed solution s' 1:s' ← s 2:for i = 1 to σ do 3:op ← RandomInteger(1, 4) 4:s' ← ApplyRandomMove(s', op) 5:end for 6: return s' Repeat the above local search and perturbation operations until the cascaded heterogeneous neighborhood search at the current tolerance depth is completed, and output the optimal cutting path.

[0132] During the CNC machining code generation stage, the algorithm first outputs the optimal path sequence `path_nodes`, which includes the access order of part clusters and the corresponding selected puncture point indices. To ensure that the physical cutting trajectory strictly follows the "optimal entry position" planned by the algorithm, the system needs to dynamically reconstruct the original geometric data. Specifically, the system traverses `path_nodes`. For each part to be machined, it first extracts the corresponding original absolute coordinates from the pre-stored `problem_data` based on the current node ID as the physical entry point. Subsequently, the system calls the vector norm calculation function to calculate the Euclidean distance from each point in the original discrete contour point set `raw_contour` of the cluster to which the part belongs to to the entry point, thereby accurately locating the point index `start_idx` with the smallest distance. Based on this, the system performs a circular shift rearrangement operation. A new index sequence `indices` is constructed using modulo operation, and its calculation logic is `mod((start_idx-1:start_idx-1+num_pts-1),num_pts) +1`. This operation reorganizes the connection order of contour points without changing the contour geometry, generating an ordered contour sequence `ordered_contour` that closes in a counterclockwise or clockwise direction with the selected puncture point as the first node. This step ensures that the laser head can seamlessly switch from idle travel to cutting travel during physical machining, avoiding any unnecessary path corrections.

[0133] To address the issue that general CNC software often automatically performs "center alignment" or "compact rearrangement" when importing external G-code, thereby disrupting the original coordinate system, this embodiment employs two mandatory protection measures in the code generation logic: During the coordinate analysis phase, the system explicitly locks the global translation offset (offset_x, offset_y) to zero, refusing to perform any relative translation calculations based on the center of the bounding box, thus strictly preserving the original absolute coordinates of all geometric points in the algorithm's solution space. In the header of the generated G-code file, immediately following the unit setting (G21) and absolute programming mode (G90) instructions, a special origin anchoring instruction G00 X0 Y0 (Anchor Origin) is pre-defined. This instruction does not perform actual machining but instead drives the laser head (or logical cursor) to explicitly access the machine tool's physical origin. The engineering intent is to force the CNC system's parsing engine to recognize a maximum graphic bounding box containing the physical origin (0,0), thereby locking the workpiece coordinate system at the software level and ensuring that the layout drawing and the machine tool bed achieve "what you see is what you get" precise material placement.

[0134] Following the restructured cluster access order, the system sequentially generates the standard G-code stream that drives the laser cutting machine's actions: Positioning action: First, a rapid traverse command G00 is generated, which drives the laser head to move at a preset rapid feed rate (such as the feed_rate_rapid parameter setting value) to the infeed starting point of the current part; Activation action: Immediately generate M03 command to activate the laser, and call S parameters to set the processing power according to the material thickness; Cutting action: Traverse the ordered_contour sequence, continuously generate linear interpolation instructions G01, and drive the laser head to complete material separation along the solid contour in conjunction with the set cutting feed rate (such as feed_rate_cut); Reset action: After a single piece is processed, an M05 command is generated to turn off the laser, and a G00 Z command is generated to raise the cutting head to a safe height (safe_height) to avoid the risk of collision during the empty stroke.

[0135] After all parts have been machined, the system writes the M30 instruction to end the program and reset the machine tool, completing the complete closed loop from algorithm optimization to physical machining.

[0136] Fifth Embodiment This embodiment provides an exemplary scheme for evaluating the overall fitness of individuals in a laser cutting path population and determining elite individuals. In this example, an initial candidate path is first generated based on the decoded target puncture point and cutting access order. After rotation and alignment processing, a standard candidate path with a virtual starting node is obtained. Then, the total cost of the empty journey of the path is calculated by calling the generalized cost matrix. Simultaneously, the compliance of processing priority constraints is verified and the penalty cost for constraint violation is calculated hierarchically to generate the path calculation result. Subsequently, the overall fitness of the standard candidate path is calculated by weighting the path calculation result. Finally, the optimal adapted path is locked after ranking the overall fitness of all individuals, and the elite individuals are encapsulated. Step S50 includes steps D11~D16: Step D11: Generate an initial candidate path based on the target puncture point and the cutting access order.

[0137] Step D12: Move the virtual starting node in the initial candidate path to the beginning of the path to obtain the standard candidate path.

[0138] Step D13: Call the generalized cost matrix to extract the movement costs between nodes in the standard candidate path in turn, and calculate the total empty journey cost corresponding to the standard candidate path.

[0139] Step D14: Check whether the cutting order of the standard candidate path conforms to the processing priority constraint, calculate the penalty cost for constraint violation in stages, and generate the path calculation result by combining the total cost of the empty journey.

[0140] Step D15: Based on the path calculation results, the comprehensive fitness of the standard candidate path is calculated by weighting according to the cost weight.

[0141] Step D16: After ranking the comprehensive fitness of the standard candidate paths, determine the optimal adaptation path, and encapsulate the optimal adaptation path and its path information to obtain the elite individual.

[0142] The initial candidate path is an initial traversal path generated by concatenating the virtual starting node and all real processing nodes based on the decoded target puncture points and cutting access order. It is the foundational object for standardization processing and fitness evaluation. For example, it includes the initial traversal sequence containing the virtual starting node, all internal hole and external contour target puncture points, the original path sequence without order alignment, and the candidate path structure with node connection relationships.

[0143] A standard candidate path is a standardized path sequence that has been rotated and aligned from the initial candidate paths, with the virtual start node fixed at the beginning and end of the path. It serves as a unified input for empty journey cost calculation and constraint verification, eliminating calculation biases caused by differences in path start points. Examples include a standardized node sequence with the virtual start node at the beginning and arranged in the order of cutting access, a path structure adapted to the generalized cost matrix calculation format, and a path traversal sequence conforming to the processing flow.

[0144] The total cost of idle travel is the cumulative cost of the laser head's movement between all adjacent nodes without cutting in a standard candidate path. It is a core indicator for evaluating the overall performance of the path and is positively correlated with the total length of the laser head's idle travel and the total travel time. For example, it is the cumulative sum of the initial movement cost from the virtual starting node to the first processing node, the idle travel cost between adjacent processing nodes, and the return cost from the final node to the virtual starting node in closed-loop mode.

[0145] The penalty cost is a tiered, high-value penalty applied to standard candidate paths that violate processing priority constraints. It is used to suppress the generation of invalid paths that violate process requirements from a fitness perspective. The penalty cost is positively correlated with the severity and frequency of constraint violations. For example, there is a first-level penalty cost for processing the outer contour before its corresponding inner hole, a second-level penalty cost for disordered processing order of contours at the same level, and a third-level penalty cost for paths containing invalid sub-loops.

[0146] The path accounting result is a comprehensive path cost accounting set that integrates the total cost of empty trips, constraint violation penalty cost, and process adaptability penalty cost of standard candidate paths. It serves as the core input for comprehensive fitness calculation. Examples include multi-dimensional cost datasets containing total cost of empty trips, constraint penalty cost, and process penalty cost; weighted path cost accounting reports; and standardized path cost sequences for fitness calculation.

[0147] Overall fitness is a core evaluation metric used to quantify the merits of standard candidate paths for individuals within a population. It is negatively correlated with the total processing cost of a path; that is, the shorter the empty journey and the higher the constraint compliance, the higher the overall fitness value. It is a core criterion for selecting elite individuals. Examples include the reciprocal of the total processing cost of a path, the normalized fitness value after weighted summation of multi-dimensional costs, and the path performance score with constraint compliance correction.

[0148] In this example, when generating standard candidate paths, a multi-dimensional, hierarchically weighted fitness evaluation method can be used. Alternatively, a streaming incremental processing method can be adopted, simultaneously receiving the target puncture point and cut access order decoded from individual populations, while simultaneously performing initial candidate path generation, rotation alignment processing, and standardization verification, accumulating a set of standard candidate paths corresponding to all individuals in the population, thereby completing the preprocessing for fitness evaluation.

[0149] After generating all standard candidate paths, the comprehensive fitness evaluation process is initiated. For each standard candidate path, the total cost of the empty journey is first calculated by accumulating the generalized cost matrix. Then, the compliance of processing priority constraints is verified, and penalty costs are calculated at different levels to generate complete path calculation results. Subsequently, the comprehensive fitness corresponding to the path is calculated based on preset weight coefficients. After the comprehensive fitness calculation of all individuals in the population is completed, the comprehensive fitness is sorted in descending order to lock the optimal fit path and encapsulate all information to obtain elite individuals. In this way, through standardized calculation and multi-dimensional evaluation throughout the entire process, the accuracy of comprehensive fitness calculation and the reliability of elite individual selection are improved, avoiding the problem of missing high-quality solutions due to a single evaluation dimension.

[0150] For example, there are two ways to calculate the overall fitness and select elite individuals through standard candidate path verification. The first is a row-by-row serial verification and constraint pre-verification method with individual locking. According to the preset iteration sequence of the population individuals, starting from the first individual of the population, the target puncture point and cutting access order corresponding to each individual are extracted sequentially, and the full process evaluation is performed for each individual. After the initial candidate path generation and rotation alignment processing of each individual is completed, the processing priority constraint pre-verification is performed simultaneously. If the path has serious violations, it is directly assigned a very low overall fitness and the subsequent cost calculation is skipped. If the path is compliant, the generalized cost matrix is ​​called to calculate the total cost of the empty journey, the graded penalty cost is calculated to generate the path verification result, and the weighted calculation is used to obtain the overall fitness of the individual. After all individuals in the entire population have completed the evaluation and fitness calculation, the overall fitness of all individuals is sorted in descending order, and the individual with the first place in the sort and passing the full verification is selected. Its full path information is encapsulated to obtain the final elite individual. This method employs a processing logic of individual-locked row-by-row serial accounting and constraint pre-verification. By pre-verifying serious violations, invalid individuals are eliminated in advance, reducing unnecessary cost accounting computation. The evaluation logic is stable and controllable, and the individual evaluation process is independent and interference-free, making it suitable for conventional processing scenarios of small and medium-sized populations.

[0151] The second approach involves parallel computation and hierarchical selection across the entire population using dimensional decomposition. This method reads and decomposes the path information of all individuals in the population, extracting four independent processing dimensions: path standardization, empty-journey cost calculation, constraint verification, and fitness calculation. Simultaneously, all individuals are aligned according to these dimensions, initiating a parallel processing flow across all dimensions. In the path standardization dimension, initial candidate paths are generated and rotated for all individuals, resulting in a batch of standard candidate paths. In the empty-journey cost calculation dimension, the total empty-journey cost of all standard candidate paths is calculated in batches based on the generalized cost matrix. In the constraint verification dimension, compliance verification of processing priority constraints for all standard candidate paths is performed simultaneously, and penalty costs for all individuals are calculated hierarchically. After all parallel processing across dimensions is complete, a weighted summation is performed on all individuals to obtain the overall fitness of the entire population in batches. Then, a hierarchical selection process is performed: invalid individuals with overall fitness below a preset threshold are first removed, and the remaining valid individuals are sorted in descending order to lock in the optimal fit path and encapsulate it into elite individuals. This method employs a multi-dimensional, parallel computational approach across the entire population and a hierarchical selection process. By fully utilizing computing resources through multi-dimensional parallel processing, it improves the efficiency of fitness assessment for large-scale populations. At the same time, hierarchical selection ensures the compliance and optimality of elite individuals, making it suitable for high-precision path optimization scenarios involving large-scale populations and complex multi-contour parts.

[0152] Sixth Embodiment This embodiment provides an exemplary scheme for adaptive neighborhood search optimization of laser cutting path and output of target cutting path. In this example, elite individuals are first used as the initial solution for local path optimization. The neighborhood search intensity of the optimal path is calculated based on the current iteration progress of the algorithm to obtain the initial optimization object. Then, the execution order of cascaded neighborhood operators is randomly shuffled and the continuous unimproved count of local search is initialized to generate a standardized neighborhood search task to be executed. Subsequently, neighborhood optimization is performed on the initial optimization object according to the neighborhood search task. The path is iteratively updated by using the strategy of exhaustive search for the optimal within the operator and the first improvement between operators until the number of consecutive unimproved times reaches the neighborhood search intensity threshold, at which point an intermediate path is output. Finally, local optimum verification is performed on the intermediate path. After confirming that the verification result is a non-local extremum state, the intermediate path is output as the target cutting path. Step S60 includes steps E11~E14: Step E11: Use the elite individual as the initial solution for local path optimization, and calculate the neighborhood search strength of the adaptive path according to the current iteration progress to obtain the initial optimization object.

[0153] Step E12: Shuffle the execution order of the cascaded neighborhood operators and initialize the continuous unimproved count of the local search to obtain the neighborhood search task to be executed.

[0154] Step E13: Perform a neighborhood search on the initial optimization object according to the neighborhood search task, and update the path by exhaustively searching for the optimal within the operator and the first improvement strategy between operators, until the number of consecutive unimproved times reaches the neighborhood search strength threshold, and output the intermediate path.

[0155] Step E14: Perform local optimum verification on the intermediate path. If the verification result is confirmed to be a non-local extremum state, then output the intermediate path as the target cutting path.

[0156] Neighborhood search intensity is a core control parameter used to regulate the search depth, iteration termination condition, and maximum number of rounds without improvement in local path optimization. Search intensity is positively correlated with the algorithm's iteration progress; lower intensity in the early stages of iteration ensures faster convergence, while higher intensity in the later stages ensures higher optimization accuracy. Examples include low-intensity fast search parameters in the early stages of iteration, high-intensity deep search parameters in the later stages, and dynamic search intensity thresholds that adapt to the population fitness distribution.

[0157] Cascaded neighborhood operators are a combination of multiple neighborhood search operators with heterogeneous path transformation capabilities. Each operator corresponds to an independent path local optimization transformation method and is the core execution unit for generating new candidate paths during the neighborhood search process. Examples include a four-level cascaded neighborhood operator set consisting of node swapping operators, path reversal operators, puncture point replacement operators, and sequence shift operators, as well as a combination of heterogeneous neighborhood operators adapted for laser cutting path optimization.

[0158] The consecutive unimproved count is a core statistical indicator used to count the number of iterations in the neighborhood search process where a better path is not obtained after continuous operator transformations. It is the core basis for determining the termination condition of the neighborhood search. For example, it includes the number of consecutive unimproved iterations within a single operator, the cumulative number of consecutive unimproved iterations after cascading execution of all operators, and the consecutive unimproved count for the entire local search process.

[0159] The neighborhood search task is a standardized local optimization task package that integrates the initial optimization object, randomly sorted cascaded neighborhood operators, initialized consecutive unimproved counts, and a neighborhood search intensity threshold. It serves as the entire process control carrier for neighborhood search execution. Examples include local optimization tasks with operator execution order, neighborhood search instructions with matching termination conditions, and local optimization tasks with initial solutions and parameter configurations.

[0160] The intra-operator exhaustive optimization strategy is an execution strategy that, for a single cascaded neighborhood operator, exhaustively searches all executable neighborhood transformations for that operator under the current path, and selects the candidate path with the best performance after the transformation. This strategy is used to ensure the accuracy of local optimization within a single operator. Examples include the exhaustive search of all interchangeable node pairs under a node swapping operator, and the full traversal of all candidate puncture points under a puncture point replacement operator.

[0161] The first-improvement strategy among operators is an execution strategy that, after exhaustive optimization of a single operator, if a better solution is found, the current path is immediately updated and the consecutive unimproved count is reset. The system then directly switches to the next operator for optimization, without needing to complete a full traversal of the remaining operators. This strategy is used to balance optimization accuracy and search efficiency. Examples include a fast iteration strategy that immediately switches operators after obtaining an improved solution from a single operator optimization, and an efficient optimization mechanism that updates the path upon the first improvement.

[0162] Intermediate paths are optimized paths obtained through iterative operator optimization after the neighborhood search process is completed, satisfying the termination condition of the neighborhood search strength. They serve as input objects for local optimum verification and as candidate output objects for the target cutting path. Examples include the optimized path sequence after completing the full-process neighborhood search, candidate cutting paths updated iteratively by cascaded operators, and intermediate optimized paths with empty travel cost calculation results.

[0163] Non-local extremum states are used to characterize the path state where the current intermediate path has not fallen into a local optimum trap. The criteria for determining this state are that the path still has room for optimization after neighborhood search, or the path performance is better than the global historical best solution. This is the core condition for determining whether a path can be used as the final target cutting path output. For example, effective path states that have not fallen into local optima, healthy path states with continuous optimization potential, and non-extremum path states that meet the requirements of global optimization.

[0164] In this example, when calculating the neighborhood search strength to obtain the initial optimization object, an iterative dynamic adaptation method can be used to obtain the current iteration progress of the hybrid heuristic algorithm, the population fitness variance, and the improvement of the global optimal solution, while dynamically calculating the neighborhood search strength and the maximum number of rounds without improvement threshold for this local search. Simultaneously, the encapsulation and parameter configuration of the initial optimization object are completed, thereby completing the pre-initialization of the neighborhood search.

[0165] After initializing the neighborhood search task, the neighborhood search execution process is initiated. Following a randomly shuffled order of cascaded neighborhood operators, operators are sequentially invoked to perform neighborhood transformations on the initial optimization object. The optimized path is iteratively updated using a strategy of exhaustive search for the optimal within each operator and initial improvement between operators, while simultaneously accumulating the count of consecutive unimproved steps until the number of consecutive unimproved steps reaches the neighborhood search intensity threshold. The optimized intermediate path is then output. Subsequently, a local optimum check is performed on the intermediate path. Once it is confirmed to be in a non-local extremum state, it is output as the final target cutting path. This approach, through adaptive intensity control and dual-strategy collaborative neighborhood search, balances the accuracy and convergence efficiency of path optimization while avoiding local extrema through local optimum check, ensuring the global optimality of the final output path.

[0166] For example, there are two ways to obtain the target cutting path through neighborhood search optimization. The first is an iterative locking operator-by-operator serial search and breakpoint accumulation method. According to the execution sequence of the cascaded neighborhood operators after random shuffling, starting from the first operator, each neighborhood operator is extracted sequentially and a full neighborhood transformation is performed. After exhaustively searching the full transformation of each operator, the optimal candidate path under that operator is simultaneously selected. If the comprehensive fitness of the candidate path is better than the current optimized path, the current optimized path is immediately updated, the consecutive unimproved count is reset, and the process directly switches to the next operator for optimization. If there is no better candidate path for the operator, the consecutive unimproved count is accumulated, and the process switches to the next operator to continue execution. When the number of consecutive unimproved times reaches the threshold corresponding to the neighborhood search intensity, the neighborhood search process is terminated, the current optimized path is output as an intermediate path, and then a local optimum check is performed on the intermediate path. After confirming that it is in a non-local extreme state, the final target cutting path is output. This method employs an iterative locking serial search by operator and a breakpoint-based dynamic accumulation execution logic. It ensures optimization accuracy through exhaustive search within operators and improves search efficiency through first-time improvement between operators. The execution logic is stable and controllable, and the path optimization process is traceable, making it suitable for small-to-medium-scale parts processing scenarios with conventional accuracy requirements.

[0167] The second approach involves multi-branch parallel search and dynamic optimization using operator grouping. This method performs functional grouping on randomly sorted cascaded neighborhood operators, dividing path structure transformation operators and puncture point optimization operators into two computationally independent parallel operator branches. Simultaneously, the initial optimization object is copied into two parallel optimization copies, corresponding to the two operator branches respectively, initiating the multi-branch parallel neighborhood search process. Within each operator branch, neighborhood transformations are executed sequentially according to the operator order. An exhaustive optimization strategy within the operator is used to filter the optimal candidate path within the branch, while simultaneously accumulating the consecutive unimproved count within the branch. After the parallel search of both branches is completed, the optimal candidate paths of the two branches are aggregated, and the globally optimal path is selected to update the current optimization path. If the path is improved, the consecutive unimproved count is reset; otherwise, the consecutive unimproved count is accumulated. This parallel search process is repeated until the number of consecutive unimproved steps reaches the neighborhood search strength threshold, at which point the optimized intermediate path is output. Subsequently, a local optimum check and constraint compliance secondary check are performed on the intermediate path. After confirming that it is in a non-local extreme state, the final target cutting path is output. This method employs multi-branch parallel search with operator grouping and global dynamic optimization execution logic. It improves local search efficiency through parallel processing with decoupled operator functions, while expanding the neighborhood search range through independent dual-branch optimization, effectively avoiding getting trapped in local extrema, and balancing optimization efficiency and global optimality. It is suitable for high-end machining scenarios of high-precision, complex multi-contour parts.

[0168] Seventh Embodiment This embodiment provides an exemplary scheme for determining local extrema and optimizing perturbations in laser cutting paths. In this example, the accumulated number of consecutive unimproved iterations during the neighborhood search process is compared with a preset maximum unimproved iterations threshold to obtain the local extrema verification result of the intermediate path. If the verification result determines that the current path is trapped in a local extrema, a perturbation mechanism is immediately triggered. According to preset perturbation intensity parameters, non-repeating neighborhood operators are selected from the available neighborhood operator set, and their execution order is determined. Then, according to the determined execution order, the corresponding number of neighborhood transformation operations are performed on the intermediate path to generate perturbed candidate paths. Finally, the processing priority constraint compliance verification is performed on the perturbed candidate paths, and invalid path segments that violate the requirement that inner holes are processed before outer contours are removed, thus determining the final target cutting path. Please refer to... Figure 4 , Figure 4 This is a flowchart illustrating the seventh embodiment of the laser cutting path planning control method of this application. Following step E13, steps F11-F14 are also included: Step F11: Compare the number of consecutive unimproved cycles during the neighborhood search process with the maximum number of unimproved cycles threshold to obtain the verification result of the intermediate path.

[0169] Step F12: If the verification result determines that the current location is trapped in a local extremum, a perturbation mechanism is triggered. According to the preset perturbation strength parameter, non-repeating neighborhood operators are selected from the set of available neighborhood operators, and the execution order of the neighborhood operators is determined.

[0170] Step F13: Perform the neighborhood transformation operation on the intermediate path a corresponding number of times according to the execution order of the neighborhood operator to obtain the perturbed candidate path.

[0171] Step F14: Verify the processing priority constraint of the disturbed candidate path, eliminate invalid path segments that violate the requirement that the inner hole is processed before the outer contour, and determine the target cutting path.

[0172] The maximum number of iterations without improvement threshold is a preset threshold used to determine whether the neighborhood search process has stalled or whether the path has fallen into a local extremum. It is the core criterion for local extremum verification, and its value is positively correlated with the adaptively adjusted neighborhood search intensity. For example, a low threshold is set in the early stage of iteration for fast judgment, a high threshold is set in the later stage of iteration for deep verification, and a dynamic threshold for the number of iterations without improvement is adapted to the current search intensity.

[0173] The local extremum verification result is a binary judgment result obtained by comparing the number of consecutive unimproved cycles with the maximum number of unimproved cycles threshold. It is used to characterize whether the current path has fallen into a local extremum. It is divided into a positive result of falling into a local extremum and a negative result of not falling into a local extremum. It is the core control basis for whether the disturbance mechanism is triggered. For example, the judgment result of falling into a local extremum when the number of consecutive unimproved cycles reaches the threshold, the judgment result of not falling into a local extremum when the number of consecutive unimproved cycles does not reach the threshold, and the local extremum verification report with stagnation level marking.

[0174] The perturbation mechanism is a path random transformation mechanism that is automatically triggered when the path determination gets stuck in a local extremum. It is used to escape the local optimum trap. By performing multiple rounds of heterogeneous neighborhood transformation on the current path, it breaks the local stagnation state of the path and generates a search starting path with new optimization potential. It is the core mechanism to ensure the global optimization capability of the algorithm. Examples include fixed-strength random neighborhood perturbation mechanism, adaptive-strength multi-level perturbation mechanism, and directional perturbation mechanism with constraint verification.

[0175] The perturbation intensity parameter is a core parameter used to control the execution strength of the perturbation mechanism. It represents the number of operators and iterations used to perform neighborhood transformations on the current path. The higher the perturbation intensity, the greater the transformation amplitude of the path and the higher the probability of escaping local extrema. For example, there are preset fixed perturbation number parameters, dynamic perturbation intensity adapted to the level of local extrema stagnation, and adaptive perturbation number threshold matched with the total number of node clusters.

[0176] The available neighborhood operator set is a predefined set of all heterogeneous neighborhood operators that can be used for path perturbation transformation. It includes all neighborhood transformation operators suitable for laser cutting path optimization and represents the range of operators selected during the perturbation process. For example, it includes the complete set of neighborhood operators such as node swapping operators, path reversal operators, puncture point replacement operators, and sequence shift operators, as well as a set of unconstrained heterogeneous operators adapted to path perturbation.

[0177] The perturbated candidate path is a candidate cutting path with a new path structure and optimization potential generated after multiple rounds of neighborhood transformation through the perturbation mechanism. It is the core input object for subsequent constraint verification and target path determination. For example, the perturbed path after three heterogeneous neighborhood transformations, the new search starting path that escapes local extrema, and the candidate cutting path with a complete node sequence.

[0178] In this example, when comparing the number of consecutive unimproved iterations with the maximum number of rounds without improvement threshold to obtain the local extremum verification result, it can be performed using an iteratively locked operator-by-operator serial search method. Alternatively, a streaming incremental comparison method can be used, accumulating the number of consecutive unimproved iterations while performing neighborhood search. After each operator iteration, a comparison is performed synchronously with the maximum number of rounds without improvement threshold, and the local extremum verification result is output in real time, thereby completing the dynamic verification of local extrema.

[0179] After completing the local extremum verification and determining that the current path is trapped in a local extremum, a perturbation mechanism is immediately triggered. According to preset perturbation strength parameters, a corresponding number of unique neighborhood operators are randomly selected from the available set of neighborhood operators, and the execution order of the operators is randomly generated. Then, the neighborhood transformation operation is continuously performed on the intermediate path a corresponding number of times according to the execution order, generating perturbed candidate paths. Subsequently, the processing priority constraint compliance verification is performed on the perturbed candidate paths, eliminating invalid path segments that violate the requirement that inner holes be processed before outer contours. Finally, a compliant target cutting path is output. This perturbation mechanism breaks the local stagnation state of the path, effectively escaping the local optimum trap, improving the algorithm's global optimization capability, and avoiding the problem of insufficient path optimization accuracy caused by neighborhood search falling into local extrema.

[0180] For example, there are two ways to generate the target cutting path through the perturbation mechanism. The first method is based on the cumulative time sequence of the number of consecutive unimproved iterations. Starting from the initial iteration of the neighborhood search, the number of consecutive unimproved iterations is accumulated synchronously after each operator iteration. The accumulated number of consecutive unimproved iterations is then compared sequentially with a preset maximum threshold for the number of rounds without improvement. After each comparison, it is simultaneously determined whether the current path is trapped in a local extremum. If the number of consecutive unimproved iterations has not reached the threshold, the neighborhood search process continues, and the number of consecutive unimproved iterations is continuously accumulated. If the number of consecutive unimproved iterations reaches the threshold, the current neighborhood search is immediately terminated, triggering the perturbation mechanism. According to the preset perturbation strength parameters, neighborhood operators that match the perturbation strength and are mutually exclusive are randomly selected from the set of available neighborhood operators, generating a random execution order. The neighborhood transformation operation is then performed on the intermediate path a corresponding number of times according to this order, generating a perturbed candidate path. Subsequently, the processing priority constraint is checked on the perturbed candidate path, invalid path segments that violate the requirement that the inner hole is processed before the outer contour are removed, the path access order is corrected, and finally, a compliant target cutting path is output. This method employs a timing-locked successive comparison and a single-step serial perturbation execution logic. By verifying the iteration timing of neighborhood search in real time, it accurately triggers the perturbation mechanism. The perturbation execution process is controllable and traceable, effectively avoiding the loss of high-quality path features caused by excessive perturbation, and is suitable for conventional small and medium-sized parts processing scenarios.

[0181] The second approach involves multi-operator batch perturbation and parallel verification. This method monitors the cumulative number of consecutive unimproved iterations during the neighborhood search process in real time and compares them against a global threshold. When the number of consecutive unimproved iterations reaches the maximum threshold for unimproved iterations, the current neighborhood search is terminated, triggering a perturbation mechanism. Based on a preset perturbation strength parameter, multiple sets of non-overlapping neighborhood operator combinations are randomly selected from the available set of neighborhood operators. The number of operators in each combination matches the perturbation strength, generating multiple independent operator execution sequences. Parallel perturbation processing is then initiated synchronously for each operator execution sequence, performing the corresponding number of neighborhood transformation operations on intermediate paths to generate multiple perturbed candidate paths in batches. After all parallel perturbation processing is completed, compliance verification of processing priority constraints is performed synchronously on all perturbed candidate paths. Invalid paths that violate the constraints are eliminated, and the path with the highest overall fitness is selected from the remaining compliant candidate paths as the final target cutting path. This method employs a multi-operator combination batch perturbation and full-path parallel verification execution logic. By processing multiple sets of heterogeneous perturbations in parallel, it broadens the path search range and improves the success rate of escaping local extrema. At the same time, parallel verification ensures the compliance of the perturbated path, making it suitable for high-end processing scenarios with complex contours and high-difficulty path optimization.

[0182] Eighth embodiment This embodiment provides an exemplary scheme for closed-loop self-optimization iteration of laser cutting path planning parameters. In this example, the idle travel time, total cutting time, and puncture point loss of the target cutting path during actual processing are first collected. A standardized measured dataset for this processing is then obtained by organizing the data according to the processing sequence. This dataset is then compared with a preset processing performance benchmark threshold. Quantitative deviation analysis is performed from three core dimensions: path running efficiency, processing device loss, and processing rule compliance, to obtain performance deviation results. Next, based on the performance deviation results, the core problem points causing the processing performance to deviate from the benchmark threshold are located. The influence factors of each core parameter in the entire path planning process on the performance deviation results are quantified and decomposed. Finally, the core parameters of the path planning are corrected according to the order of influence factor weight from high to low, generating an updated path planning parameter set, thus achieving continuous iterative optimization of the cutting path generation for the next part to be processed. Please refer to... Figure 5 , Figure 5 This is a flowchart illustrating the eighth embodiment of the laser cutting path planning control method of this application. Following step S60, steps G11-G14 are also included: Step G11: Collect the idle travel time, total cutting time, and puncture point loss of the target cutting path during actual processing, and organize them according to the time sequence to obtain the actual data set of this processing.

[0183] Step G12: Match the measured dataset with the processing performance benchmark threshold, and perform quantitative deviation analysis from three dimensions: path running efficiency, processing device loss, and processing rule compliance to obtain the performance deviation results.

[0184] Step G13: Based on the performance deviation results, determine the problem points that lead to the performance deviation benchmark threshold according to the deviation level, and quantify the influence factor weights of the core parameters of each problem point in the path planning process on the performance deviation results.

[0185] Step G14: Correct the core parameters corresponding to the path planning process according to the order of the influence factor weights from high to low, and obtain the updated path planning parameter set to optimize the generation of the cutting path for the next processing part.

[0186] Idle travel time is the cumulative time the laser head spends moving between adjacent processing nodes without cutting action. It is a core indicator of path operation efficiency and is positively correlated with the total idle travel length. For example, it includes the time it takes for the laser head to move from the virtual starting node to the first puncture point, the rapid movement time between adjacent contour puncture points, and the return travel time to the reset point after processing.

[0187] Total cutting time is the cumulative processing time for the laser head to complete all contour cutting actions on the part to be processed. It includes the total processing time of the entire process, including the puncture start-up time, contour cutting time, and idle travel time. It is a core comprehensive indicator for measuring processing efficiency. For example, the total time for full contour cutting of a single part, the processing cycle time of a single part in batch processing, and the total processing time including auxiliary actions.

[0188] Puncture point loss is the quantified cumulative loss caused to the laser nozzle, focusing lens, and material surface by the laser head when performing the initiation cut at the puncture point. It is positively correlated with the number of puncture points and the duration of a single puncture, and is a core indicator for measuring the wear and tear of processed components and processing costs. Examples include the quantified nozzle loss for a single puncture, the total cumulative loss of all puncture points on a single part, and the cumulative puncture loss for batch processing.

[0189] The measured dataset is a standardized collection of full-dimensional performance measured data collected during the current processing, organized according to the processing time sequence. It is the core input object for performance deviation analysis. For example, it includes time-series datasets containing idle travel time, total cutting time, puncture point loss, and processing compliance rate; a set of measured performance data for the entire process of single-part processing; and a single-batch average performance dataset for batch processing.

[0190] Machining performance benchmark thresholds are preset critical values ​​used to measure whether machining performance meets standards. They are pre-set based on historical best machining data, machine tool rated performance parameters, and process requirements, and serve as the judgment benchmark for performance deviation analysis. Examples include the maximum allowable threshold for idle travel time, the benchmark target value for total cutting time, the upper limit threshold for puncture point loss, and the minimum requirement threshold for machining rule compliance rate.

[0191] The path operation efficiency dimension is a core evaluation dimension used to measure the proportion of idle travel in the cutting path and the utilization rate of processing time. The core evaluation indicators are the proportion of idle travel time and the target completion rate of total cutting time. For example, the ratio of idle travel time to total cutting time, the deviation rate between actual processing time and the benchmark target time, and the proportion of effective cutting time of the laser head.

[0192] The wear dimension of processed components is a core evaluation dimension used to measure component wear and processing costs during laser processing. The core evaluation indicators are the total wear at puncture points and the average wear value per puncture. For example, the cumulative wear at puncture points of a single part, the percentage of nozzle lifespan wear, and the quantified value of ablation wear at puncture points of the sheet metal.

[0193] The compliance dimension of processing rules is a core evaluation dimension used to measure whether the cutting path meets the processing priority constraints and process requirements. The core evaluation indicators are the compliance rate of internal hole processing before external contour processing, the path non-interference rate, and the puncture point process compliance rate. For example, the percentage of contours whose processing sequence meets the priority constraints, the percentage of non-interference path segments, and the percentage of puncture points that meet the starting cutting process requirements.

[0194] Performance deviation results are obtained by comparing and analyzing measured datasets with processing performance benchmark thresholds. These results provide a comprehensive quantitative assessment of performance deviation across all dimensions, including deviation values, deviation rates, and deviation levels. This data serves as the core basis for identifying problem areas and analyzing influencing factors. Examples include a path operation efficiency deviation quantification report, device loss deviation values, compliance deviation rates, and comprehensive performance deviation analysis results with deviation level markings.

[0195] Impact factor weights are numerical values ​​used to quantify the proportion of influence of each core parameter on the performance deviation results throughout the path planning process. The higher the weight, the greater the impact of that parameter on the performance deviation, and it is the core basis for prioritizing parameter adjustments. Examples include the influence proportion of the generalized cost matrix weight coefficient, the influence weight of the neighborhood search intensity, the influence proportion of the perturbation intensity parameter, and the influence weight of the processing priority penalty coefficient.

[0196] The path planning parameter set is a standardized set of parameters that integrates all modifiable core parameters in the entire path planning process. It serves as the core configuration basis for generating the next cutting path. For example, it includes a parameter set containing generalized cost matrix weight coefficients, neighborhood search strength, perturbation strength, processing priority constraint penalty coefficients, contour discrete sampling step size, and a path planning configuration parameter package adapted to machine tool performance.

[0197] In this example, when collecting actual processing data of the target cutting path to obtain a standardized test dataset, the global full-volume synchronous acquisition method can be used. Alternatively, a streaming incremental real-time acquisition method can be used, where the idle travel time, cutting time, and puncture point loss data of each stage are collected in real time during the laser cutting process. Data processing and standardization are completed synchronously according to the processing sequence, and the full standardized test dataset is directly output after processing, thus completing the acquisition of processing performance data.

[0198] After collecting the standardized measured dataset, the performance deviation analysis process was initiated. The measured dataset was compared with the preset processing performance benchmark thresholds. Quantitative deviation analysis was performed from three core dimensions: path operation efficiency, processing component wear, and processing rule compliance, to obtain the performance deviation results for this processing. Subsequently, based on the performance deviation results, and according to the deviation levels of the three dimensions of path operation efficiency, processing component wear, and processing rule compliance, the core deviation dimensions were identified. The corresponding path planning stages were traced back to investigate the compatibility of core parameters with processing conditions, machine tool performance, and material characteristics. Combined with parameter sensitivity analysis, the core problem points that caused the performance to deviate from the benchmark thresholds were located, providing a basis for subsequent decomposition of influencing factor weights. The impact factors of the core parameters of each problem point in the path planning on the performance deviation results are quantitatively decomposed. Finally, the core parameters are corrected in order of the impact factor weight from high to low, and an updated path planning parameter set is generated for the next cutting path generation of the parts to be processed. In this way, through closed-loop feedback and iterative optimization of processing test data, the machine tool adaptability and processing performance of the path planning scheme are continuously improved, avoiding the problem of decreased processing performance caused by fixed parameters being unable to adapt to machine tool performance degradation, changes in sheet material characteristics, and fluctuations in batch processing conditions.

[0199] For example, there are two ways to generate an updated path planning parameter set through performance deviation analysis and parameter correction. The first method involves extracting measured data from the three core dimensions sequentially, starting with the time-series data from the initial processing stage, based on the processing time sequence bound to the measured dataset. Each dimension is then compared against its corresponding processing performance benchmark threshold. After each dimension's deviation comparison, the deviation value and deviation rate are calculated simultaneously to determine if a performance deviation exists. If a performance deviation exists, the dimension is added to the core problem point set, and the corresponding path planning core parameters are continuously broken down, calculating the influence factor weights of each parameter. If no performance deviation exists, the in-depth breakdown analysis of that dimension is skipped, and the comparison analysis proceeds to the next dimension. After all three dimensions have been compared and analyzed, all core problem points and their corresponding parameter influence factor weights are summarized, sorted in descending order of influence factor weights, and the values ​​of the corresponding core parameters are corrected sequentially to generate the updated path planning parameter set. This method employs a time-locked, dimension-by-dimensional serial comparison and breakpoint-based in-depth analysis logic. By analyzing the process timeline in a dimension-by-dimensional manner, it accurately locates the core source of performance deviations. The parameter correction process has clear priorities, is controllable and traceable, effectively avoiding excessive correction of non-core parameters and is suitable for conventional single-part batch processing scenarios.

[0200] The second approach involves multi-dimensional clustering interval partitioning and parallel optimization of all parameters. This method reads and analyzes the standardized measured dataset across all dimensions. Based on the performance data features of three core dimensions, clustering is performed, dividing performance data within the same preset deviation range and those that are sequentially continuous into the same performance feature range. Deviation level boundaries between ranges are marked, generating multiple performance analysis ranges that are computationally independent. Parallel analysis is then initiated simultaneously for all partitioned performance analysis ranges. Within each range, the deviation between the measured data and the baseline threshold is compared to pinpoint the core problem areas. Corresponding path planning core parameters are simultaneously decomposed, and the influence factor weights of each parameter are quantified. After the parallel analysis and weight calculation for all performance analysis ranges are completed, the influence factor weights of all core parameters are aggregated and sorted from highest to lowest weight. Parallel correction and verification are then performed on all core parameters simultaneously to eliminate coupling conflicts between parameters, ultimately generating a standardized updated path planning parameter set. This method employs an analysis logic of multi-dimensional clustering interval partitioning and parallel optimization of all parameters. By performing pre-processing clustering of all-dimensional features, the core intervals of performance deviations are identified in advance, reducing the number of invalid dimensional comparisons. At the same time, parallel analysis improves the efficiency of multi-dimensional deviation analysis and parameter optimization, making it suitable for flexible processing scenarios with multiple varieties and variable batches.

[0201] This application provides a laser cutting path planning device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the laser cutting path planning control method in the above embodiment 1.

[0202] The following is for reference. Figure 6 The diagram illustrates a structure suitable for implementing the laser cutting path planning device of the embodiments of this application. The laser cutting path planning device in the embodiments of this application may include, but is not limited to, mobile terminals such as fiber laser cutting machines, sheet metal laser cutting blanking machines, and three-dimensional five-axis laser cutting machine tools, as well as fixed terminals such as laser cutting CNC systems and laser cutting automatic nesting machines. Figure 6 The laser cutting path planning device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0203] like Figure 6As shown, the laser cutting path planning device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the laser cutting path planning device. The processing unit 1001, the ROM 1002, and the RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the laser cutting path planning device to communicate wirelessly or wiredly with other devices to exchange data. Although laser cutting path planning devices with various systems are shown in the figures, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0204] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0205] The laser cutting path planning device provided in this application, employing the laser cutting path planning control method in the above embodiments, can solve the technical problem of poor laser cutting path planning. Compared with the prior art, the beneficial effects of the laser cutting path planning device provided in this application are the same as those of the laser cutting path planning control method provided in the above embodiments, and other technical features of this laser cutting path planning device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0206] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0207] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0208] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the control method for laser cutting path planning in the above embodiments.

[0209] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, radio frequency (RF), etc., or any suitable combination thereof.

[0210] The aforementioned computer-readable storage medium may be included in the laser cutting path planning equipment; or it may exist independently and not be assembled into the laser cutting path planning equipment.

[0211] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by a laser cutting path planning device, the laser cutting path planning device performs the following actions: discretizes the geometric contour data of the part to be processed to obtain a node cluster and candidate puncture nodes; introduces a virtual starting node and constructs a generalized cost matrix based on the processing mode using the node cluster, the candidate puncture nodes, and processing priority constraints; initializes a coding population using a two-segment continuous variable encoding and a hybrid heuristic algorithm; performs two-segment decoding on each coding population to obtain the target puncture point and cutting access order of the part corresponding to the coding population; evaluates the individual comprehensive fitness of the coding population based on the generalized cost matrix, the target puncture point, and the cutting access order, and determines elite individuals; uses the elite individuals as the initial solution for local path optimization and performs local path optimization on the initial solution according to the cascaded neighborhood operator to obtain the target cutting path, wherein the neighborhood search intensity of each iteration in the local path optimization is determined according to the iteration progress.

[0212] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0213] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation that may be implemented in systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0214] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0215] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described laser cutting path planning control method, thereby solving the technical problem of poor laser cutting path planning. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the laser cutting path planning control method provided in the above embodiments, and will not be repeated here.

[0216] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A control method for laser cutting path planning, characterized in that, The method includes: Discretize the geometric contour data of the part to be processed to obtain node clusters and candidate puncture nodes; A virtual starting node is introduced, and the node cluster, the candidate puncture node, and the processing priority constraint are used to construct a generalized cost matrix according to the processing mode. The coding population is initialized using a two-segment continuous variable encoding and a hybrid heuristic algorithm. Perform two-segment decoding on each of the coded populations to obtain the target puncture point and the cutting access order of the part corresponding to the coded population; Based on the generalized cost matrix, the target puncture point of the part, and the cutting and access order of the part, the overall fitness of the encoded population is evaluated to identify elite individuals. The elite individuals are used as the initial solution for local path optimization, and the initial solution is subjected to local path optimization according to the cascaded neighborhood operator to obtain the target cutting path. The neighborhood search intensity of each iteration in the local path optimization is determined according to the iteration progress.

2. The control method for laser cutting path planning as described in claim 1, characterized in that, The step of discretizing the geometric contour data of the part to be processed to obtain the node cluster and candidate puncture nodes includes: Based on the graphic file corresponding to the layout of the parts to be processed, locate and parse the entity data segments in the graphic file, and determine the characteristic parameters of various geometric elements; The feature parameters of the geometric primitives are subjected to primitive splicing operation to obtain the initial contour of the part to be processed; The initial contour is preprocessed by normalization to remove redundant points and correct minor contour deviations, thereby obtaining the geometric contour data of the part to be processed. Discretize the geometric contour data of the part to be processed, divide the contour of a single part into independent node clusters, and determine the feasible candidate puncture nodes within the node clusters.

3. The control method for laser cutting path planning as described in claim 1, characterized in that, The step of introducing a virtual starting node, and constructing a generalized cost matrix based on the node cluster, the candidate puncture nodes, and processing priority constraints according to the processing mode includes: By taking the candidate puncture nodes in the node cluster as processing nodes and embedding the sequential processing associations corresponding to the processing priority constraints, the basic topology of the augmented graph is obtained. The virtual starting node is added to the node set of the augmented graph's basic topology, and the augmented graph model is obtained by uniformly modeling according to the processing mode. Based on the augmented graph model, the node configuration cost is obtained by configuring the idle travel cost between processing nodes and the return travel cost corresponding to the virtual starting node according to the processing mode. The node configuration costs between all nodes are arranged into a unified structured generalized cost matrix according to the arrangement order of the nodes in the augmented graph model.

4. The control method for laser cutting path planning as described in claim 1, characterized in that, The step of performing dual-segment decoding on each of the encoded populations to obtain the target puncture point and the cutting access sequence of the part corresponding to the encoded population includes: Discrete mapping is performed on the first segment of continuous variables of individuals in the encoded population to obtain the continuous node values ​​corresponding to the node cluster; The continuous values ​​of the nodes are converted into candidate puncture point indices within the node cluster to determine the target puncture point of the part to be processed. The second continuous variable of the individuals in the encoded population is sorted and mapped to obtain the access order corresponding to the node cluster; The access order of the node clusters is determined by sorting the access positions of the second continuous variable according to their numerical values, thereby determining the cutting access order of the part to be processed.

5. The control method for laser cutting path planning as described in claim 1, characterized in that, The step of evaluating the overall fitness of the encoded population and determining elite individuals based on the generalized cost matrix, the target puncture point of the part, and the cutting and access order of the part includes: Based on the target puncture point and the cutting access order, an initial candidate path is generated; The virtual start node in the initial candidate path is moved to the beginning of the path to obtain the standard candidate path; The generalized cost matrix is ​​invoked to extract the movement cost between nodes in the standard candidate path in turn, and the total empty trip cost corresponding to the standard candidate path is calculated. Check whether the cutting order of the standard candidate path conforms to the processing priority constraint, calculate the penalty cost for constraint violation in stages, and generate the path calculation result by combining the total cost of empty travel; Based on the path calculation results and cost weights, the overall fitness of the standard candidate path is calculated using a weighted average. After ranking the comprehensive fitness of the standard candidate paths, the optimal adaptation path is determined, and the optimal adaptation path and its path information are encapsulated to obtain the elite individual.

6. The control method for laser cutting path planning as described in claim 1, characterized in that, The step of using the elite individual as the initial solution for local path optimization, and performing local path optimization on the initial solution according to the cascaded neighborhood operator to obtain the target cutting path, wherein the neighborhood search intensity of each iteration in the local path optimization is determined according to the iteration progress includes: The elite individuals are used as the initial solutions for local path optimization, and the neighborhood search strength of the adaptive path is calculated according to the current iteration progress to obtain the initial optimization objects. Disrupt the execution order of the cascaded neighborhood operators and initialize the continuous unimproved count of the local search to obtain the neighborhood search task to be executed; According to the neighborhood search task, the initial optimization object is searched for a neighborhood, and the path is updated by exhaustive search within the operator and first improvement between operators, until the number of consecutive no improvements reaches the neighborhood search strength threshold, and the intermediate path is output. If the intermediate path is locally optimal and the verification result is confirmed to be a non-local extremum state, then the intermediate path is output as the target cutting path.

7. The control method for laser cutting path planning as described in claim 6, characterized in that, The control method for laser cutting path planning further includes the steps of performing a neighborhood search on the initial optimization object according to the neighborhood search task, updating the path through a strategy of exhaustive search within operators and first improvement between operators, until the number of consecutive unimproved steps reaches the neighborhood search intensity threshold, and then outputting the intermediate path. The verification result of the intermediate path is obtained by comparing the number of consecutive unimproved rounds during the neighborhood search process with the maximum number of unimproved rounds threshold. If the verification result determines that the current location is trapped in a local extremum, a perturbation mechanism is triggered. According to the preset perturbation strength parameter, non-repeating neighborhood operators are selected from the set of available neighborhood operators, and the execution order of the neighborhood operators is determined. According to the execution order of the neighborhood operators, the intermediate path is subjected to a corresponding number of neighborhood transformation operations to obtain the perturbed candidate path; The processing priority constraints of the disturbed candidate paths are checked, and invalid path segments that violate the requirement that the inner hole is processed before the outer contour are eliminated, thus determining the target cutting path.

8. The control method for laser cutting path planning as described in claim 1, characterized in that, The method for controlling laser cutting path planning further includes the step of using the elite individual as the initial solution for local path optimization and performing local path optimization on the initial solution according to the cascaded neighborhood operator to obtain the target cutting path. Following the step of determining the neighborhood search intensity of each iteration in the local path optimization process based on the iteration progress, the control method further includes: The idle travel time, total cutting time, and puncture point wear of the target cutting path during actual processing are collected and organized according to time sequence to obtain the actual data set of this processing. The measured dataset was matched with a processing performance benchmark threshold, and a quantitative deviation analysis was performed from three dimensions: path running efficiency, processing device wear, and processing rule compliance, to obtain the performance deviation results. Based on the performance deviation results, the problem points that lead to the performance deviation benchmark threshold are determined according to the deviation level, and the influence factor weights of the core parameters of each problem point in the path planning process on the performance deviation results are quantified. The core parameters corresponding to the path planning process are corrected according to the order of the influence factor weights from high to low to obtain an updated path planning parameter set, so as to optimize the generation of the cutting path for the next processing part.

9. A laser cutting path planning device, characterized in that, The laser cutting path planning device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the control method for laser cutting path planning as described in any one of claims 1 to 8.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the control method for laser cutting path planning as described in any one of claims 1 to 8.

Images