Industrial cutting path self-learning generation system and method

Abstract

The invention discloses an industrial cutting path self-learning generation system and method, relates to the technical field of industrial cutting paths, and is used for solving the problems that traditional cutting path planning mostly depends on a simple heuristic rule or a greedy algorithm, complex multi-laser-machine collaborative operation cannot be effectively processed, and in the multi-laser-machine collaborative cutting process, the multi-laser-machine collaborative operation cannot be effectively processed. As a result, some machines work excessively while other machines are idle, the diversity of populations is prone to loss, and a globally optimal solution cannot be found; cutting path planning is analogous to a multi-traveling salesman problem, a path is optimized through a heuristic algorithm, working loads of all laser machines are ensured to be relatively balanced through a penalty mechanism for load balancing, a new solution is generated through crossover and mutation operation, the quality of a final solution is improved, and introduction of a simulated annealing algorithm can help jump out of a local optimal solution. And the convergence speed and the solution quality are further improved, especially in the problem of relatively large search space.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial cutting paths, and more particularly, to a system and method for self-learning generation of industrial cutting paths. Background Art

[0002] It mainly generates more efficient and accurate cutting paths through optimization algorithms, machine learning technology, and multi-machine collaborative optimization, which can solve the problems of low efficiency, uneven load, and inflexible paths in traditional path planning methods. With the development of artificial intelligence, deep learning, and intelligent manufacturing technology, cutting path generation in the future will become more intelligent and automated, and can be optimized in real time according to actual conditions.

[0003] The existing technology has the following deficiencies: Traditional cutting path planning relies heavily on simple heuristics or greedy algorithms, which are ineffective in handling the complex collaborative operation of multiple laser machines. This can lead to some machines being overworked while others are idle, reducing overall efficiency. The diversity of the population is often lost, causing the algorithm to become trapped in a local optimum and unable to find a global optimal solution. Optimization methods such as crossover and mutation operations help maintain population diversity and prevent premature convergence.

[0004] In view of the above problems, the present invention proposes a solution. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an industrial cutting path self-learning generation system and method to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions: Industrial cutting path self-learning generation system and method, including the steps of: Step S1: analogize cutting path planning to the traveling salesman problem, and analogize parallel multi-machine collaborative cutting to the multiple traveling salesman problem. Use a heuristic algorithm to solve the multiple traveling salesman problem, establish a fitness function, combine parent individuals into offspring individuals through crossover operations, perform mutation operations on the newly generated offspring, generate new offspring through selection, crossover and mutation, and merge the offspring with the current population; Step S2: By monitoring the frequency of illegal solutions, load balancing penalties, and population diversity, determine whether a crossover elimination mechanism needs to be added to optimize the solution quality and algorithm; Step S3: Determine whether to use the simulated annealing algorithm to assist in exploring the search space based on the fitness value improvement rate and the convergence speed; Step S4: Through multi-generation optimization, set the iterative termination judgment condition, select the individual with the best fitness as the next generation population to obtain the optimal path for multi-laser collaborative cutting.

[0007] In a preferred embodiment, the method comprises the following steps: Cutting path planning is analogized to the traveling salesman problem. The path nodes of the workpiece to be cut, including the cutting starting point, end point or intermediate point on the workpiece, are analogized to cities. The moving trajectory of the cutting head is analogized to the path of the traveling salesman. Multiple cutting heads are analogized to multiple traveling salesmen. The goal is to minimize the total moving distance of the cutting head, that is, to solve the shortest path. For a weighted complete graph ;in, is the set of all nodes, n is the total number of nodes, and the edge set for The set of lines connecting any two points in , represents the connection between node i and node j, Yes Arrive distance, satisfy and , at the same time , r represents the weight between nodes, then the mathematical model is established as: ;in, Yes Arrive distance, yes Corresponding point Arrive The distance weight of The constraints for establishing the mathematical model are: 1. Each node is visited exactly once, for each target node j: ; 2. Path continuity: ; Among them, the graph formed by randomly selecting several vertices from the vertex set V of the graph G and the corresponding edges between these vertices in the edge set E of the original graph G is called an arbitrary subgraph S; 3. All nodes must be covered: .

[0008] Determine the path of each cutting head. In parallel multi-machine collaborative cutting, for each cutting head, determine the movement from one node to the next node in turn. The path nodes are ;in, It is a cutting head The number of path nodes, cutting head Path length It is calculated by accumulating the distances between adjacent nodes, that is, ;in, Here is the node To Node The distance weight, is the actual distance between the two nodes; Calculate the total path length D: the total path length of all cutting heads It is the sum of the path lengths of each cutting head, that is ;in, is the number of cutting heads.

[0009] The process of using heuristic algorithms to solve the multi-traveling salesman problem is as follows: Step A1: Randomly generate a certain number of individuals, denoted as PopSize, each of which represents a path allocation scheme for a cutting head. For each cutting head, randomly select a node from its unvisited node set as the starting point. According to the nearest neighbor strategy rule, that is, each time select the unvisited node closest to the current node, and select subsequent nodes in turn until all nodes are assigned to a cutting head; Step A2: For each individual in the population, calculate its fitness value according to the fitness function to evaluate the individual's performance in optimizing the objective, taking into account the minimization of the total path length and the load balancing penalty: Indicates the The load of the cutting head is the average of all cutting head loads. , then the load balancing penalty is expressed as: ; Where m is the number of cutting heads, is the penalty coefficient, and the fitness function is expressed as: ; Where D is the total path length; Step A3: Select a certain number of individuals (denoted as SelSize) from the current population as parents based on their fitness values. Use the roulette wheel selection method. The probability of an individual being selected is proportional to its fitness value. Calculate the sum of the fitness values ​​of all individuals in the population, SumFitness. For each individual i, calculate its probability of being selected. , generate a random number Ram in the range [0,1], if , then the individual is selected as the parent; Step A4: Perform a crossover operation on the selected parent individuals to generate offspring individuals. Randomly select a crossover point and exchange the parts of the two parent individuals after the point to obtain a new offspring individual. Step A5: The newly generated offspring individuals are recorded as Perform mutation operation, randomly select a node and change its position in the path; Step A6: Merge the newly generated offspring individuals with the current population to form a new population. Sort the merged population according to the fitness value, and retain the individuals with the highest fitness to form the next generation population. Step A7: Check whether the current number of iterations reaches the preset upper limit MaxIter. If so, the algorithm terminates and outputs the optimal individual in the current population. If not, return to step A2 to calculate the fitness and continue operations such as selection, crossover, and mutation until the termination condition is met.

[0010] In a preferred embodiment, the method comprises the following steps: Illegal solution frequency monitoring and analysis: each generation of population is counted to determine the number of illegal solutions And calculate the frequency of illegal solutions: ; Among them, PopSize is the population size, and the changing trend of the frequency of illegal solutions is recorded through multiple iterations. If it exceeds a certain pre-set threshold in several consecutive generations , it means that the solution quality generated by the current algorithm is poor; Analysis of load balancing penalties: Exceeding a pre-set threshold When , it indicates that the algorithm does not consider load balancing well during path planning; Population diversity monitoring and analysis: It was found that population diversity continued to decrease for several generations. When the drop exceeds a certain pre-set ratio threshold β, the algorithm converges prematurely and falls into a local optimal solution; When any one or more indicators of illegal solution frequency, load balancing penalty value or population diversity reach a pre-set threshold, a cross-elimination mechanism is added to optimize the solution quality and algorithm; The offspring is generated by sequential crossover operation. According to the load balancing penalty, the path of the offspring individual is partially adjusted. It is checked whether the adjusted individual meets the constraint conditions. If not, further adjustments are made until a legal individual is obtained.

[0011] In a preferred embodiment, the method comprises the following steps: Set up the first The fitness value of the best individual in the generation population is , No. The fitness value of the best individual in the generation population is , then the fitness value improvement rate The calculation is expressed as: ;in, is the number of iterations for continuous statistics, and cs is the starting number of iterations for starting to count the fitness value improvement rate; right Itness setting threshold ,when itness≥ When , it means that the fitness value increases rapidly: when itness When it is even close to 0, it means that the fitness value increases slowly; The convergence speed is measured by the number of iterations required. The number of iterations when the fitness value of the best individual in the population is less than or equal to the set fitness threshold for the first time is recorded as the convergence speed. A threshold is set for the convergence speed. When it is greater than or equal to the threshold, it means the convergence speed is fast. On the contrary, when it is less than the threshold, it means the convergence speed is slow. The rules for determining whether to use the simulated annealing algorithm to assist in exploring the search space are as follows: Rule 1: When the fitness value improvement rate is low and the convergence speed is fast, use the simulated annealing algorithm to assist in exploring the search space; Rule 2: When the fitness value improvement rate is high but the convergence speed is slow, use the simulated annealing algorithm to assist in exploring the search space; Rule 3: When the fitness value improvement rate and convergence speed are both ideal, simulated annealing algorithm is not used to assist in exploring the search space; Rule 4: When the fitness value improvement rate and convergence speed are not ideal, use the simulated annealing algorithm to assist in exploring the search space.

[0012] The steps for using the simulated annealing algorithm to assist in exploring the search space are as follows: Step C1: Encode the moving path of the cutting head into a node sequence: ;in, is a node on the path; Step C2: Initialize parameters, including initial temperature , termination temperature , cooling rate and the initial solution ; Step C3: For the path planning problem, the objective function is the total length of the path. Calculate the objective function value of the current solution Y. ; Step C4: Generate a new solution by randomly selecting two nodes in the path and exchanging their positions or randomly selecting a node and inserting it into the neighborhood of another random position in the path, and calculate the objective function value of the new solution ; According to the Metropolis criterion, decide whether to accept the new solution and calculate ,if , then directly accept the new solution ; if , then according to the probability Accept the new interpretation; among them, is the current temperature, generated using a random number generator A random number θ between Accept the new solution when , otherwise keep the current solution Y; Step C5: After each iteration, follow the cooling rate Update temperature, i.e. ; Step C6: Check the current temperature Is it less than the termination temperature? If yes, the algorithm stops and outputs the current optimal solution; otherwise, it returns to step C4 to continue the iterative search.

[0013] In a preferred embodiment, the method comprises the following steps: The optimization algorithm is terminated by controlling the timing of termination through dual judgment criteria, balancing computational efficiency and solution quality. The algorithm terminates when either the hard condition or the soft condition is met: Hard condition (upper limit of iterations): When the number of iterations reaches the preset maximum value MaxIter, it is forced to terminate; Flexible condition (convergence stability): The threshold value of the fitness improvement rate of the optimal solution for 30 consecutive generations is ε = 0.1%. If the fitness improvement rate of the optimal solution for 30 consecutive generations is less than 0.1% of this threshold, the algorithm will be terminated early. The best chromosome encoding structured path set after multiple generations of optimization , ] indicates the The node access sequence of the laser machine, The number of cutting nodes assigned to each laser machine.

[0014] The industrial cutting path self-learning generation system includes: analogy collaboration module, mechanism optimization module, auxiliary exploration module and iteration termination module, and the signal connections between each module; Analogy collaboration module: The cutting path planning is analogized to the traveling salesman problem. Parallel multi-machine collaborative cutting is analogized to the multiple traveling salesman problem. A heuristic algorithm is used to solve the multiple traveling salesman problem. A fitness function is established to combine parent individuals into offspring individuals through crossover operations. The newly generated offspring are mutated. New offspring are generated through selection, crossover, and mutation, and the offspring are merged with the current population. Mechanism Optimization Module: By monitoring the frequency of illegal solutions, load balancing penalties, and population diversity, it determines whether a cross-elimination mechanism needs to be added to optimize the solution quality and algorithm; Auxiliary exploration module: Combined with the fitness value improvement rate and convergence speed, it is determined whether to use the simulated annealing algorithm to assist in the exploration of the search space; Iteration termination module: Through multi-generation optimization, iterative termination judgment conditions are set, and the individuals with the best fitness are selected as the next generation population to obtain the optimal path for collaborative cutting of multiple laser machines.

[0015] The technical effects and advantages of the industrial cutting path self-learning generation system and method of the present invention are as follows: Comparing cutting path planning to the multi-traveling salesman problem, optimizing the path through a heuristic algorithm can effectively reduce the path length and improve cutting efficiency. Through the penalty mechanism for load balancing, it ensures that the workload of all laser machines is relatively balanced, thereby improving overall work efficiency. New solutions are generated through crossover and mutation operations, and the quality of the solution is optimized using the crossover elimination mechanism, which helps to avoid local optimal solutions and improve the quality of the final solution. The introduction of the simulated annealing algorithm can help escape the local optimal solution and further improve the convergence speed and solution quality, especially in problems with a large search space. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a structural diagram of the industrial cutting path self-learning generation system and method of the present invention.

[0017] Figure 2 This is a workflow diagram of the analog collaborative module of the present invention.

[0018] Figure 3 A flow chart of adding a crossover elimination mechanism to the present invention to optimize the solution quality and algorithm. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] The present invention is collected.

[0021] Example 1 The present invention discloses a method for self-learning and generating an industrial cutting path, comprising the steps of: Step S1: analogize cutting path planning to the traveling salesman problem, and analogize parallel multi-machine collaborative cutting to the multiple traveling salesman problem. Use a heuristic algorithm to solve the multiple traveling salesman problem, establish a fitness function, combine parent individuals into offspring individuals through crossover operations, perform mutation operations on the newly generated offspring, generate new offspring through selection, crossover and mutation, and merge the offspring with the current population; Step S2: By monitoring the frequency of illegal solutions, load balancing penalties, and population diversity, determine whether a crossover elimination mechanism needs to be added to optimize the solution quality and algorithm; Step S3: Determine whether to use the simulated annealing algorithm to assist in exploring the search space based on the fitness value improvement rate and the convergence speed; Step S4: Through multi-generation optimization, set the iterative termination judgment condition, select the individual with the best fitness as the next generation population to obtain the optimal path for multi-laser collaborative cutting.

[0022] In step S1, the cutting path planning is analogized to the traveling salesman problem, and the parallel multi-machine collaborative cutting is analogized to the multiple traveling salesman problem. A heuristic algorithm is used to solve the multiple traveling salesman problem. A fitness function is established to combine parent individuals into offspring individuals through crossover operations. The newly generated offspring are mutated, and new offspring are generated through selection, crossover, and mutation. The offspring are then merged with the current population. The specific contents include: Cutting path planning is analogized to the traveling salesman problem. The path nodes of the workpiece to be cut, including the cutting starting point, end point or intermediate point on the workpiece, are analogized to cities. The moving trajectory of the cutting head is analogized to the path of the traveling salesman. Multiple cutting heads are analogized to multiple traveling salesmen. The goal is to minimize the total moving distance of the cutting head, that is, to solve the shortest path. For a weighted complete graph ;in, is the set of all nodes that can be passed through, n is the total number of nodes, and the edge set for The set of lines connecting any two points in , represents the connection between node i and node j, Yes Arrive distance, satisfy and , at the same time , indicating that the distance is symmetrical and positive, there is no infinite distance, r represents the weight between nodes, then the mathematical model is established as: ;in, Yes Arrive distance, yes Corresponding point Arrive The distance weight of It should be noted that in a complete graph, any two different nodes are connected by an edge, that is, if there are n nodes in the graph, then the graph has 2n(n−1) edges; a weighted complete graph is based on a complete graph, and each edge is assigned a weight. The weight usually represents the "cost" or "distance" of connecting the two points. The weight can be the distance between the two nodes, the transmission time, or other metrics. In a weighted complete graph, each edge Connects node i and node j and has a weight associated with it ; The constraints for establishing the mathematical model are: 1. Each node is visited only once. For each target node j, the sum of the distances from i to j for all edges pointing to it is 1: ; 2. Path continuity: ; Among them, the graph formed by randomly selecting several vertices from the vertex set V of the graph G and the corresponding edges between these vertices in the edge set E of the original graph G is called an arbitrary subgraph S; 3. All nodes must be covered: .

[0023] Determine the path of each cutting head. In parallel multi-machine collaborative cutting, for each cutting head, the movement from one node to the next node is determined in sequence according to the set of path nodes assigned to it. The path nodes are It is a cutting head The number of path nodes), then the cutting head Path length It is calculated by accumulating the distances between adjacent nodes, that is, ;in, Here is the node To Node The distance weight, is the actual distance between the two nodes; Calculate the total path length D: the total path length of all cutting heads It is the sum of the path lengths of each cutting head, that is ;in, is the number of cutting heads.

[0024] The process of using heuristic algorithms to solve the multi-traveling salesman problem is as follows: Step A1: Initialize the population A certain number of individuals, denoted as PopSize, are randomly generated. Each individual represents a path allocation scheme for a cutting head. The specific generation method can be a random allocation method, that is, for each cutting head, a node is randomly selected from its unvisited node set as the starting point, and then the nearest neighbor strategy rule is followed, that is, the unvisited node closest to the current node is selected each time, and subsequent nodes are selected in turn until all nodes are assigned to a cutting head or certain task allocation conditions are met; Step A2: Calculate fitness For each individual in the population, its fitness value is calculated according to the above fitness function. The fitness value is obtained by calculating the total distance weighted value of each cutting head path and the load balancing penalty term. This is used to evaluate the individual's performance in optimizing the goal, taking into account the minimization of the total path length and the load balancing penalty: Indicates the The load of the cutting head is the average of all cutting head loads. , then the load balancing penalty is expressed as: ; Where m is the number of cutting heads, is the penalty coefficient, and the fitness function is expressed as: ; Where D is the total path length; Step A3: Select an Action Select a certain number of individuals (denoted as SelSize) from the current population as parents based on their fitness values. Use the roulette wheel selection method. The probability of an individual being selected is proportional to its fitness value. Individuals with high fitness are more likely to be selected. Calculate the sum of the fitness values ​​of all individuals in the population, SumFitness. For each individual i, calculate its probability of being selected. , then generate a random number Ram in the range [0,1], if , then the individual is selected as the parent; Step A4: Crossover Operation Perform a crossover operation on the selected parent individuals to generate offspring individuals. Taking single-point crossover as an example, a crossover point is randomly selected and the parts of the two parent individuals after the point are exchanged to obtain new offspring individuals. In this way, the offspring individuals inherit some of the good characteristics of the parent individuals and may produce better ones. Step A5: Mutation Operation The newly generated offspring individuals are recorded with a certain mutation probability as Perform a mutation operation to randomly select a node and change its position in the path. The mutation operation can increase the diversity of the population and prevent the algorithm from falling into a local optimal solution. When performing mutation, it is necessary to ensure that the mutated path still meets the constraints of the problem; Step A6: Population Update The newly generated offspring individuals are merged with the current population to form a new population. The merged population is sorted according to the fitness value, and the individuals with the highest fitness are retained to form the next generation population, and subsequent iterative optimization is continued.

[0025] Step A7: Iteration termination judgment Check whether the current number of iterations reaches the preset upper limit MaxIter. If so, the algorithm terminates and outputs the optimal individual in the current population, that is, the optimal path allocation plan for all cutting heads. If not, return to the fitness calculation step A2 and continue the selection, crossover, mutation and other operations until the termination condition is met.

[0026] In step S2, by monitoring the frequency of illegal solutions, load balancing penalties, and population diversity, it is determined whether a crossover elimination mechanism needs to be added to optimize the solution quality and algorithm. The specific contents include: Illegal solution frequency monitoring and analysis: Definition of illegal solution: refers to a solution that does not satisfy the constraints in the mathematical model; For example, the constraint of "each node is visited only once" is not satisfied, that is, a node is repeatedly visited by multiple cutting heads; or the path continuity constraint is not satisfied. (in It is a subgraph composed of a number of vertices and their corresponding edges randomly selected from the vertex set of the original graph, which leads to path discontinuity, etc. Illegal solution frequency calculation: During the algorithm iteration process, each generation of population is generated and the number of illegal solutions is counted. And calculate the frequency of illegal solutions: ; Where PopSize is the population size. The changing trend of the frequency of illegal solutions is recorded through multiple iterations. If the frequency of illegal solutions remains high for several consecutive generations, that is, exceeds a pre-set threshold, , such as 0.1 or 0.2, and the specific threshold can be adjusted according to the actual problem, then it means that the solution quality generated by the current algorithm is poor; Analysis of load balancing penalties: A larger load balancing penalty value means that the load difference between the cutting heads is larger, that is, the path may not be balanced enough. If the load balancing penalty value continues to be large during the algorithm iteration, that is, it exceeds the pre-set threshold , such as 100 or set according to the specific problem scale and requirements, it means that the algorithm does not consider load balancing well when planning paths; Population diversity monitoring and analysis: Method for measuring population diversity: Calculate the Hamming distance between individuals in the population. For two individuals, that is, the path allocation scheme of two cutting heads, its encoding is represented as a binary string, and whether each node is visited by a cutting head can be expressed as as a measure of population diversity; Relationship between population diversity and optimization: If, during the algorithm iteration process, it is found that the population diversity continues to decrease for several consecutive generations When the drop exceeds a certain preset ratio threshold β, Or it is set according to the actual situation, which means that the algorithm may converge too early and fall into a local optimal solution; When any one or more of the indicators, such as the frequency of illegal solutions, the load balancing penalty value, or the population diversity, reaches a pre-set threshold, a cross-elimination mechanism is added to optimize the solution quality and algorithm. The cross-elimination mechanism is implemented in the following ways: A sequential crossover operation is used to generate offspring, and then the paths of the offspring individuals are partially adjusted according to the load balancing penalty, such as exchanging some path nodes of certain cutting heads to make the load more balanced. At the same time, it is checked whether the adjusted individuals meet the constraints. If not, further adjustments are made until a legal individual is obtained, thereby optimizing the quality of the solution.

[0027] In step S3, the fitness value improvement rate and convergence speed are combined to determine whether to use the simulated annealing algorithm to assist in exploring the search space. The specific contents include: The fitness value improvement rate refers to the average improvement of the fitness value of the best individual in the population in several consecutive iterations. The fitness value of the best individual in the generation population is , No. The fitness value of the best individual in the generation population is , then the fitness value improvement rate It can be calculated as: ;in, is the number of iterations for continuous statistics, and cs is the starting number of iterations for starting to count the fitness value improvement rate; right Itness setting threshold ,when itness≥ When , it means that the algorithm is constantly improving the quality of the solution, and the fitness value increases rapidly: itness When it approaches 0, it indicates that the algorithm may have fallen into a local optimal solution and the fitness value increases slowly. The convergence speed is measured by the number of iterations required. The number of iterations when the fitness value of the best individual in the population is less than or equal to the set fitness threshold for the first time is recorded as the convergence speed. A threshold is set for the convergence speed. When the value is greater than or equal to the threshold, the convergence speed is fast, indicating that the algorithm can quickly find a better solution. On the contrary, when the value is less than the threshold, the convergence speed is very slow, and it may take a long time to find a better solution, and the algorithm efficiency is low. The rules for determining whether to use the simulated annealing algorithm to assist in exploring the search space are as follows: Rule 1: When the fitness value improvement rate is low and the convergence speed is fast, use the simulated annealing algorithm to assist in exploring the search space. This indicates that the algorithm may have fallen into a local optimal solution and quickly converged to this local optimal area. In this case, it is necessary to introduce the simulated annealing algorithm to assist in exploring the search space. The simulated annealing algorithm accepts worse solutions with a certain probability, which can help the algorithm escape the current local optimal area and continue to explore a wider search space, potentially finding a better global optimal solution. Rule 2: When the fitness value increases rapidly but convergence is slow, use simulated annealing to assist in exploring the search space. This indicates that while the algorithm can continuously improve the quality of the solution, it is inefficient. Therefore, the simulated annealing algorithm is introduced to leverage its ability to explore the search space and accelerate convergence.

[0028] Rule 3: When both the fitness improvement rate and convergence speed are ideal, do not use simulated annealing to assist in exploring the search space. This indicates that the current heuristic algorithm is already solving the problem well. In this case, it is advisable to not use simulated annealing and continue searching using the original algorithm to avoid introducing additional computational complexity and uncertainty.

[0029] Rule 4: When the fitness value improvement rate and convergence speed are not ideal, use the simulated annealing algorithm to assist in exploring the search space.

[0030] The steps for using the simulated annealing algorithm to assist in exploring the search space are as follows: Step C1: Encode the moving path of the cutting head into a node sequence: ;in, is a node on the path; Step C2: Initialize parameters Initial temperature : Setting a higher initial temperature controls the probability of the algorithm accepting a poor solution at the beginning. The higher the temperature, the greater the possibility of accepting a poor solution, which is conducive to the algorithm exploring a larger search space. Termination temperature : Set a sufficiently small termination temperature. When the current temperature drops to this value, the algorithm stops. The termination temperature determines the final stage of the algorithm. When the temperature is low enough, the algorithm mainly performs a fine search in the neighborhood of the better solution. Cooling rate :Usually the value is Between, such as 95, etc. After each iteration, the temperature is The algorithm gradually shifts from extensive exploration to local optimization as the iteration proceeds. Initial solution : Randomly generate an initial path solution, or use other heuristic algorithms to generate a relatively good initial solution.

[0031] Step C3: Calculate the objective function value For the path planning problem, the objective function is the total length of the path, and the objective function value of the current solution Y is calculated. .

[0032] Step C4: Iterative search process Generate a new solution through certain neighborhood operations ,For path planning problems, common neighborhood operations include: ,randomly selecting two nodes in the path and exchanging their positions, ,e.g. becomes Randomly select a node and insert it into another random position in the path, such as Insert into Path Into , calculate the objective function value of the new solution: calculate the new solution The objective function value of Decision to accept the new solution: Decide whether to accept the new solution based on the Metropolis criterion, and calculate ,if , then directly accept the new solution ; if , then according to the probability Accept the new interpretation; among them, is the current temperature. A random number generator can be used to generate a A random number θ between Accept the new solution when , otherwise keep the current solution Y; Step C5: Temperature Update After each iteration, the cooling rate Update temperature, i.e. ; Step C6: Termination condition judgment Check the current temperature Is it less than the termination temperature? If yes, the algorithm stops and outputs the current optimal solution; otherwise, it returns to step C4 to continue the iterative search.

[0033] In step S4, through multi-generation optimization, the iterative termination judgment condition is set, and the individual with the best fitness is selected as the next generation population to obtain the optimal path for multi-laser collaborative cutting. The specific contents include: The optimization algorithm is terminated by controlling the timing of termination through dual judgment criteria, balancing computational efficiency and solution quality. The algorithm terminates when either the hard condition or the soft condition is met: Hard condition (upper limit of iterations): When the number of iterations reaches the preset maximum value MaxIter, the algorithm is forced to terminate to prevent it from falling into an infinite loop and ensure that a feasible solution is obtained within a limited time; Flexible conditions (convergence stability): A threshold value of ε = 0.1% is defined for the fitness improvement rate of the optimal solution for 30 consecutive generations. If the fitness improvement rate of the optimal solution for 30 consecutive generations is less than this threshold value of 0.1%, the algorithm is terminated early to avoid invalid calculations when the algorithm enters the "plateau" period, that is, when the optimization effect of the algorithm is no longer significantly improved. The algorithm is automatically stopped to save computing resources. The optimal chromosome encoding structured path set after multiple generations of optimization , ] indicates the The node access sequence of the laser machine, The number of cutting nodes allocated to each laser machine (must meet ;in, ).

[0034] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0035] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0036] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application of the technical solution and the invention constraints. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0037] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0038] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0039] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The industrial cutting path self-learning generation method is characterized by: Including steps: Step S1: analogize cutting path planning to the traveling salesman problem, and analogize parallel multi-machine collaborative cutting to the multiple traveling salesman problem. Use a heuristic algorithm to solve the multiple traveling salesman problem, establish a fitness function, combine parent individuals into offspring individuals through crossover operations, perform mutation operations on the newly generated offspring, generate new offspring through selection, crossover and mutation, and merge the offspring with the current population; Step S2: By monitoring the frequency of illegal solutions, load balancing penalties, and population diversity, determine whether a crossover elimination mechanism needs to be added to optimize the solution quality and algorithm; Step S3: Determine whether to use the simulated annealing algorithm to assist in exploring the search space based on the fitness value improvement rate and the convergence speed; Step S4: Through multi-generation optimization, set the iterative termination judgment condition, select the individual with the best fitness as the next generation population to obtain the optimal path for multi-laser collaborative cutting.

2. The method for self-learning and generating an industrial cutting path according to claim 1, characterized in that: Cutting path planning is analogized to the traveling salesman problem. The path nodes of the workpiece to be cut, including the cutting starting point, end point or intermediate point on the workpiece, are analogized to cities. The moving trajectory of the cutting head is analogized to the path of the traveling salesman. Multiple cutting heads are analogized to multiple traveling salesmen. The goal is to minimize the total moving distance of the cutting head, that is, to solve the shortest path. For a weighted complete graph ;in, is the set of all nodes, n is the total number of nodes, and the edge set for The set of lines connecting any two points in , represents the connection between node i and node j, Yes Arrive distance, satisfy and , at the same time , r represents the weight between nodes, then the mathematical model is established as: ;in, Yes Arrive distance, yes Corresponding point Arrive The distance weight of The constraints for establishing the mathematical model are: Condition 1: Each node is visited exactly once, for each target node j: ; Condition 2: Path continuity: ; Among them, the graph formed by randomly selecting several vertices from the vertex set V of the graph G and the corresponding edges between these vertices in the edge set E of the original graph G is called an arbitrary subgraph S; Condition 3: All nodes must be covered: .

3. The method for self-learning and generating an industrial cutting path according to claim 2, characterized in that: Determine the path of each cutting head. In parallel multi-machine collaborative cutting, for each cutting head, determine the movement from one node to the next node in turn. The path nodes are ;in, It is a cutting head The number of path nodes, cutting head Path length It is calculated by accumulating the distances between adjacent nodes, that is, ;in, Here is the node To Node The distance weight, is the actual distance between the two nodes; Calculate the total path length D: the total path length of all cutting heads It is the sum of the lengths of the paths of the various cutting heads, i.e. ; where m is the number of cutting heads.

4. The method for self-learning and generating an industrial cutting path according to claim 3, characterized in that: The process of using heuristic algorithms to solve the multi-traveling salesman problem is as follows: Step A1: Randomly generate a certain number of individuals, denoted as PopSize, each of which represents a path allocation scheme for a cutting head. For each cutting head, randomly select a node from its unvisited node set as the starting point. According to the nearest neighbor strategy rule, that is, each time select the unvisited node closest to the current node, and select subsequent nodes in turn until all nodes are assigned to a cutting head; Step A2: For each individual in the population, calculate its fitness value according to the fitness function to evaluate the individual's performance in optimizing the objective, taking into account the minimization of the total path length and the load balancing penalty: Indicates the The load of the cutting head is the average of all cutting head loads. , then the load balancing penalty is expressed as: ; Where m is the number of cutting heads, is the penalty coefficient, and the fitness function is expressed as: ;Where, D is the total path length; Step A3: Select a certain number of individuals (denoted as SelSize) from the current population as parents based on their fitness values. Use the roulette wheel selection method. The probability of an individual being selected is proportional to its fitness value. Calculate the sum of the fitness values ​​of all individuals in the population, SumFitness. For each individual i, calculate its probability of being selected. , generate a random number Ram in the range [0,1], if , then the individual is selected as the parent; Step A4: Perform a crossover operation on the selected parent individuals to generate offspring individuals. Randomly select a crossover point and exchange the parts of the two parent individuals after the point to obtain a new offspring individual. Step A5: The newly generated offspring individuals are recorded as Perform mutation operation, randomly select a node and change its position in the path; Step A6: Merge the newly generated offspring individuals with the current population to form a new population. Sort the merged population according to the fitness value, and retain the individuals with the highest fitness to form the next generation population. Step A7: Check whether the current number of iterations reaches the preset upper limit MaxIter. If so, the algorithm terminates and outputs the optimal individual in the current population. If not, return to step A2 to calculate the fitness and continue operations such as selection, crossover, and mutation until the termination condition is met.

5. The industrial cutting path self-learning generation method according to claim 1, characterized in that: Illegal solution frequency monitoring and analysis: each generation of population is counted to determine the number of illegal solutions And calculate the frequency of illegal solutions: ; Among them, PopSize is the population size, and the changing trend of the frequency of illegal solutions is recorded through multiple iterations. If it exceeds a pre-set threshold in several consecutive generations , it means that the solution quality generated by the current algorithm is poor; Analysis of load balancing penalties: Exceeding a pre-set threshold When , it indicates that the algorithm does not consider load balancing well during path planning; Population diversity monitoring and analysis: It was found that population diversity continued to decrease for several generations. When the drop exceeds a certain pre-set ratio threshold β, the algorithm converges prematurely and falls into a local optimal solution; When any one or more indicators of illegal solution frequency, load balancing penalty value or population diversity reach a pre-set threshold, a cross-elimination mechanism is added to optimize the solution quality and algorithm; The offspring is generated by sequential crossover operation. According to the load balancing penalty, the path of the offspring individual is partially adjusted. It is checked whether the adjusted individual meets the constraint conditions. If not, further adjustments are made until a legal individual is obtained.

6. The method for self-learning and generating an industrial cutting path according to claim 1, characterized in that: Set up the first The fitness value of the best individual in the generation population is , No. The fitness value of the best individual in the generation population is , then the fitness value improvement rate The calculation is expressed as: ;in, is the number of iterations for continuous statistics, and cs is the starting number of iterations for starting to count the fitness value improvement rate; right Itness setting threshold ,when itness≥ When , it means that the fitness value increases rapidly: when itness When it is even close to 0, it means that the fitness value increases slowly; The convergence speed is measured by the number of iterations required. The number of iterations when the fitness value of the best individual in the population is less than or equal to the set fitness threshold for the first time is recorded as the convergence speed. A threshold is set for the convergence speed. When it is greater than or equal to the threshold, it means the convergence speed is fast. On the contrary, when it is less than the threshold, it means the convergence speed is slow. The rules for determining whether to use the simulated annealing algorithm to assist in exploring the search space are as follows: Rule 1: When the fitness value improvement rate is low and the convergence speed is fast, use the simulated annealing algorithm to assist in exploring the search space; Rule 2: When the fitness value improvement rate is high but the convergence speed is slow, use the simulated annealing algorithm to assist in exploring the search space; Rule 3: When the fitness value improvement rate and convergence speed are both ideal, simulated annealing algorithm is not used to assist in exploring the search space; Rule 4: When the fitness value improvement rate and convergence speed are not ideal, use the simulated annealing algorithm to assist in exploring the search space.

7. The method for self-learning and generating an industrial cutting path according to claim 6, characterized in that: The steps for using the simulated annealing algorithm to assist in exploring the search space are as follows: Step C1: Encode the moving path of the cutting head into a node sequence: ;in, is a node on the path; Step C2: Initialize parameters, including initial temperature , termination temperature , cooling rate and the initial solution ; Step C3: For the path planning problem, the objective function is the total length of the path. Calculate the objective function value of the current solution Y. ; Step C4: Generate a new solution by randomly selecting two nodes in the path and exchanging their positions or randomly selecting a node and inserting it into the neighborhood of another random position in the path, and calculate the objective function value of the new solution ; According to the Metropolis criterion, decide whether to accept the new solution and calculate ,if , then directly accept the new solution ; if , then according to the probability Accept the new interpretation; among them, is the current temperature, generated using a random number generator A random number θ between Accept the new solution when , otherwise keep the current solution Y; Step C5: After each iteration, follow the cooling rate Update temperature, i.e. ; Step C6: Check the current temperature Is it less than the termination temperature? If yes, the algorithm stops and outputs the current optimal solution; otherwise, it returns to step C4 to continue the iterative search.

8. The method for self-learning and generating an industrial cutting path according to claim 1, wherein: The optimization algorithm is terminated by controlling the timing of termination through dual judgment criteria, balancing computational efficiency and solution quality. The algorithm terminates when either the hard condition or the soft condition is met: Hard condition (upper limit of iterations): When the number of iterations reaches the preset maximum value MaxIter, it is forced to terminate; Flexible condition (convergence stability): The threshold value of the fitness improvement rate of the optimal solution for 30 consecutive generations is ε = 0.1%. If the fitness improvement rate of the optimal solution for 30 consecutive generations is less than 0.1% of this threshold, the algorithm will be terminated early. The best chromosome encoding structured path set after multiple generations of optimization , ] indicates the The node access sequence of the laser machine, The number of cutting nodes assigned to each laser machine.

9. An industrial cutting path self-learning generation system, for implementing the industrial cutting path self-learning generation method according to any one of claims 1 to 9, characterized in that: Analogy collaboration module: The cutting path planning is analogized to the traveling salesman problem. Parallel multi-machine collaborative cutting is analogized to the multiple traveling salesman problem. A heuristic algorithm is used to solve the multiple traveling salesman problem. A fitness function is established to combine parent individuals into offspring individuals through crossover operations. The newly generated offspring are mutated. New offspring are generated through selection, crossover, and mutation, and the offspring are merged with the current population. Mechanism Optimization Module: By monitoring the frequency of illegal solutions, load balancing penalties, and population diversity, it determines whether a cross-elimination mechanism needs to be added to optimize the solution quality and algorithm; Auxiliary exploration module: Combined with the fitness value improvement rate and convergence speed, it is determined whether to use the simulated annealing algorithm to assist in the exploration of the search space; Iteration termination module: Through multi-generation optimization, iterative termination judgment conditions are set, and the individuals with the best fitness are selected as the next generation population to obtain the optimal path for collaborative cutting of multiple laser machines.

Images