Variable production knowledge migration intelligent layout method and system
Through the matching mechanism of shape context algorithm and twin neural network, combined with clustering of similar parts and improved positioning algorithm, the problems of high computational complexity and cumbersome rotation angle adjustment in the existing technology are solved, and efficient layout of two-dimensional irregular parts and improved material utilization are achieved.
Patent Information
- Application Number
- CN202510642163.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-19
- Publication Date
- 2025-09-19
AI Technical Summary
In the existing technology of variable-type production, the particle swarm optimization algorithm based on sequence migration and the neural network positioning strategy have the problems of high computational complexity, high computational randomness and cumbersome rotation angle adjustment, making it difficult to efficiently match and optimize the layout of two-dimensional irregular parts.
A dual matching mechanism based on shape context algorithm and twin neural network is adopted, combined with the priority clustering rule of similar parts. By defining the matching result set MS, a hybrid positioning algorithm of improved heuristic algorithm and minimum envelope rectangle increment is used to optimize the layout order and rotation angle of target task parts.
It achieves efficient matching and optimization of target task parts, reduces computational complexity, improves material utilization and production efficiency, and reduces production costs.
Smart Images
Figure CN120671887A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of variable production, and in particular relates to a method and system for intelligent layout of variable production knowledge migration. Background Art
[0002] In the steel cutting and blanking process of the heavy industry, a rotating batch production model based on variant design is often used for similar products. This production model is commonly referred to as variable-type production or variable-variety production. In this production process, nesting methods are a key technical step in improving material utilization and controlling production costs. Their quality directly affects the company's production efficiency and economic benefits.
[0003] Problems with existing technologies:
[0004] From the algorithmic level, the particle swarm optimization algorithm based on sequence migration can inherit and migrate historically excellent nesting sequences in a relatively short period of time, and some of the solution effects can be compared with certain heuristic algorithms. The use of neural network-based part similarity measurement and improved positioning algorithm to guide the optimized nesting of target tasks has certain limitations. On the one hand, the algorithm relies on the excellent nesting sequences provided by the nesting cases. When the nesting sequence of parts in the source task is not provided or the number of matching target task parts and source task parts is small, the randomness of the calculation is relatively high. On the other hand, although the positioning strategy adopted by the algorithm is better, it is also more complex, which increases the computational complexity. When adjusting and calculating the rotation angle of parts, it is not as free and time-saving as the manual nesting operation in enterprise production. Summary of the Invention
[0005] The present invention aims to provide a method and system for intelligent nesting based on knowledge transfer for variable-type production. This method provides a two-dimensional irregular nesting algorithm based on nesting diagram reuse, solving the nesting problem of two-dimensional irregular parts of varying scales and sizes. The method achieves solutions that are superior or competitive to existing nesting algorithms, meeting the needs of variable-type production and demonstrating promising application potential.
[0006] The technical solutions adopted by the present invention are as follows:
[0007] A variant production knowledge transfer intelligent layout method, comprising:
[0008] Obtain target task and source task knowledge, and store the source task knowledge of the layout in the information storage module;
[0009] Match the parts in the target task with the parts in different source tasks in the information storage module. At the same time, prioritize clustering the same type of parts with the same shape features in the target task and centrally calculate them in the nesting optimization.
[0010] Save the matching results to the collection MS;
[0011] The matching result set MS is used to predict the order and rotation angle of the parts in the target task set on the plate;
[0012] Take the historical task case with the best matching result as the current source task part set for matching;
[0013] At the same time, taking the nesting diagram as the source task knowledge, an improved heuristic algorithm for solving the two-dimensional irregular parts nesting problem is adopted, and a hybrid positioning algorithm based on fit and minimum envelope rectangle increment is used to output the optimal nesting results of the target task parts.
[0014] The source task knowledge includes original part drawings, historical excellent nesting sequences, nesting part sizes and their numbers; and a part set in a target task is obtained.
[0015] The matching of the parts in the target task with the parts in different source tasks in the information storage module specifically includes:
[0016] Define the part set in the target task as P0 = {P 01 , P 02 ,...,P 05},
[0017] The set of parts corresponding to MS is Ps={P s1 , P s2 ,...,P s5};
[0018] Match the parts in the set MS and P0 one by one, and let the parts in the target task inherit the arrangement order and rotation angle of the corresponding parts in the set MS. Then the arrangement sequence of the parts in the target task P0 on the plate is (P 01 , P 03 , P 05 , P 04 , P 02 ).
[0019] In the matching of the target task part and the source task part, for the part matching calculated based on the shape context algorithm, the source task part with the smallest matching distance value is taken as the matching result, and the currently matched source part is hidden.
[0020] The feature is that: when the matching distance value is greater than 6.1, it is considered that there is no part in the source task that matches the current target task;
[0021] The historical task cases with the smallest average distance are used as the current source task parts set for matching.
[0022] In the matching of the target task parts and the source task parts, for the parts matching based on the twin neural network, the output distribution is used as the intuitive matching result, and the historical task case with the highest reuse degree of the nesting task is used as the source task part set currently used for matching.
[0023] A variant production knowledge transfer intelligent nesting system, comprising:
[0024] The information storage module is used to store the original part drawings of the source task, historical excellent nesting sequences, nesting part sizes and their numbers;
[0025] Part matching module, used to match the target task parts with the source parts in the information storage module, and save the matching results to the set MS;
[0026] The prediction module uses the set MS to predict the nesting order and rotation angle of the target part, so that the target part inherits the nesting order and rotation angle of the corresponding parts in the set MS;
[0027] The nesting optimization module adopts an improved heuristic algorithm for solving the two-dimensional irregular parts nesting problem and outputs the optimal nesting result based on a hybrid positioning algorithm based on fit and minimum envelope rectangle increment.
[0028] The parts matching module supports matching methods based on shape context algorithm and twin neural network in the matching process, and follows the rule of prioritizing clustering of similar parts.
[0029] The sequence migration module uses the historical task cases with the smallest distance to the average value as the source task part set for matching.
[0030] The technical effects achieved by the present invention are:
[0031] The present invention achieves efficient matching of target task and source task parts through a dual matching mechanism based on shape context algorithm and twin neural network, combined with the priority clustering rule of similar parts; among them, the shape context algorithm ensures the reliability of matching results by screening historical cases with the smallest matching distance threshold and average distance; the twin neural network uses the case with the highest reuse of the pattern arrangement task as the matching benchmark, enhancing the system's adaptability to complex part matching scenarios, and effectively solving the problem of high calculation randomness when the number of matching source task parts is insufficient.
[0032] The present invention avoids repeated matching and invalid calculations by hiding the matched source parts, prioritizing clustering of similar parts and centralized calculations, thus simplifying the matching process. At the same time, the target task parts directly inherit the layout order and rotation angle of the source task matching parts, without the need to repeatedly adjust the rotation angle, which significantly reduces the time complexity of the algorithm and improves the calculation efficiency.
[0033] The present invention defines the matching result set MS and inherits the source task nesting order, so that the target task parts can be directly arranged into a sequence based on excellent historical nesting experience, thereby reducing the randomness of sequence optimization; combined with an improved heuristic algorithm with the nesting diagram as the knowledge source and a hybrid positioning algorithm based on fit and minimum envelope rectangle increment, while ensuring the rationality of nesting, the material utilization rate is improved and the production cost of steel cutting in the heavy industry is reduced; in addition, by directly inheriting the rotation angle and the rule of preferentially clustering similar parts, the algorithm is closer to the operational logic of manual nesting in part angle adjustment and grouping calculation, avoiding the cumbersome problem of rotation angle calculation in the prior art, shortening the nesting planning time, and improving production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 It is a framework diagram of the transfer learning method of the present invention;
[0035] Figure 2 It is the process of predicting the arrangement sequence of parts in the target task of the present invention;
[0036] Figure 3 It is a flow chart of the KRIH algorithm of the present invention
[0037] Figure 4 It is a system block diagram of the present invention. DETAILED DESCRIPTION
[0038] In order to make the purpose and advantages of the present invention more clearly understood, the present invention is described in detail below with reference to the following examples. It should be understood that the following text is only used to describe one or more specific embodiments of the present invention and does not strictly limit the scope of protection of the present invention.
[0039] like Figure 1-Figure 3 As shown, a variant production knowledge transfer intelligent layout method includes:
[0040] Obtain target task and source task knowledge, and store the source task knowledge of nesting in the information storage module; the source task knowledge includes the original part drawing, historical excellent nesting sequence, nesting part size and its number; obtain the part set in the target task;
[0041] Match the parts in the target task with the parts in different source tasks in the information storage module. At the same time, prioritize clustering the same type of parts with the same shape features in the target task and centrally calculate them in the nesting optimization.
[0042] Define the part set in the target task as P0 = {P 01 , P 02 ,...,P 05},
[0043] The set of parts corresponding to the set MS is Ps={P s1 , P s2 ,...,P s5}, match the parts in the set MS and P0 one by one, and let the parts in the target task inherit the arrangement order and rotation angle of the corresponding parts in the set MS, then the arrangement sequence of the parts in the target task P0 on the plate is (P 01 , P 03 , P 05 , P 04 , P 02 );
[0044] Based on nesting rules and case experience, when performing matching calculations, similar parts with the same shape features in the target task are prioritized for clustering, so as to concentrate calculations in nesting optimization. For part matching calculated based on the shape context algorithm, the source task part with the smallest matching distance value is selected as the matching result, and the currently matched source parts are hidden.
[0045] When the matching distance value is greater than 6.1, it is considered that there is no part in the source task that matches the current target task;
[0046] The historical task cases with the smallest distance to the average value are used as the current source task parts set for matching;
[0047] For parts matching based on twin neural networks, the output distribution is used as the intuitive matching result, and the historical task case with the highest nesting task reuse is used as the source task part set currently used for matching;
[0048] Save the matching results to the collection MS;
[0049] The matching result set MS is used to predict the order and rotation angle of the parts in the target task set on the plate;
[0050] Take the historical task case with the best matching result as the current source task part set for matching;
[0051] At the same time, taking the nesting diagram as the source task knowledge, an improved heuristic algorithm for solving the two-dimensional irregular parts nesting problem is adopted, and a hybrid positioning algorithm based on fit and minimum envelope rectangle increment is used to output the optimal nesting results of the target task parts.
[0052] like Figure 4 As shown, a variant production knowledge transfer intelligent nesting system includes:
[0053] The information storage module is used to store the original part drawings of the source task, historical excellent nesting sequences, nesting part sizes and their numbers;
[0054] The parts matching module is used to match the target task parts with the source parts in the information storage module and save the matching results to the set MS. The matching process supports matching methods based on shape context algorithm and twin neural network, and follows the priority clustering rule of similar parts.
[0055] For part matching based on shape context algorithm, the source task part with the smallest matching distance value is taken as the matching result, and the currently matched source parts are hidden.
[0056] When the distance value is greater than 6.1, it is considered that there is no matching part in the source task with the current target task; and the historical task case with the smallest average distance value is used as the current source task part set for matching;
[0057] The prediction module uses the set MS to predict the nesting order and rotation angle of the target part, so that the target part inherits the nesting order and rotation angle of the corresponding parts in the set MS.
[0058] According to the above, the source task knowledge for defining nesting is stored in the information storage module, also known as the knowledge base, including information such as the original part drawing, historical excellent nesting sequence, nesting part size and its number;
[0059] The results of matching the parts in the target task with the parts in the source task are saved in the set MS. The matching results MS are used to predict the order and rotation angle of the parts in the target task set on the plate, which is the key step in part sequence migration.
[0060] Each part in the target task will be matched with parts in different cases in the information storage module (ISM), and the historical task case with the best matching result will be used as the current source task part set for matching;
[0061] The successfully matched parts in the source task are saved in the set MS, which guides the optimization of the nesting sequence of the matched parts in the target task;
[0062] Define the part set in the target task as P0 = {P 01 , P 02 ,...,P 05},
[0063] The set of parts corresponding to MS is Ps={P s1 , P s2 ,...,P s5}, match the parts in the set MS and P0 one by one, and let the parts in the target task inherit the arrangement order and rotation angle of the corresponding parts in the MS set, then the arrangement sequence of the parts in the target task P0 on the plate is (P 01 , P 03 , P 05 , P04 , P 02 );
[0064] Based on nesting rules and case experience, when performing matching calculations, similar parts with the same shape features in the target task are prioritized for clustering, so as to concentrate calculations in nesting optimization;
[0065] For part matching based on shape context algorithms, the source task part with the smallest matching distance is selected as the matching result, and the currently matched source parts are hidden. When the distance value is greater than 6.1, it is considered that there is no matching part in the source task with the current target task, and the historical task case with the smallest average distance is used as the current source task part set for matching.
[0066] For parts matching based on twin neural networks, the output distribution is used as the intuitive matching result, and the historical task case with the highest nesting task reuse is used as the source task part set currently used for matching;
[0067] At the same time, with excellent nesting drawings as source task knowledge, an improved heuristic algorithm for solving the two-dimensional irregular parts nesting problem is proposed. The two-dimensional irregular parts nesting problem solving algorithm: inputs the plate and source task nesting drawing information; and matches it with the information stored in the information storage module; then performs part graphics segmentation, sequence arrangement and angle information extraction; and gives the nesting task of part shape similarity for reusability evaluation; then performs sequence migration operation on the matched previous source task parts; uses reinforcement learning to perform sequence search operation on the unmatched target task parts; and gives a hybrid positioning strategy of fit and minimum envelope rectangle increment; when the termination condition is met, the nesting result is output and the nesting result is saved in the information storage module; when the termination condition is not met, the reward and part status based on reinforcement learning optimization are updated; then the sequence information of the part is adjusted, and the part positioning is performed to calculate the nesting height and utilization rate. If the termination condition is met, the nesting result is output and stored. If not, the above operation is continued until it is met; the optimal nesting result is output using a hybrid positioning algorithm based on fit and minimum envelope rectangle increment.
[0068] Example 1:
[0069] A sheet metal processing scene
[0070] The target task part set P0 contains 5 special-shaped parts, which are matched to the historical case MS-023 through the ISM module:
[0071] The shape context matched 3 parts successfully (EMD values were 5.2, 4.7, and 5.9 respectively);
[0072] The Siamese network matched 1 part (92.3% similarity)
[0073] The remaining part generates a new arrangement sequence through reinforcement learning;
[0074] The final nesting utilization rate was 83.7%, and the calculation took 29 seconds, which is a significant improvement compared to traditional methods that do not use knowledge transfer.
[0075] The foregoing is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art may make various improvements and modifications without departing from the principles of the present invention, and such improvements and modifications are also within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained herein shall, unless otherwise specified or limited, be implemented in accordance with conventional means in the art.
Claims
1. A variant production knowledge transfer intelligent layout method, characterized by: include: Obtain target task and source task knowledge, and store the source task knowledge of the layout in the information storage module; Match the parts in the target task with the parts in different source tasks in the information storage module. At the same time, prioritize clustering the same type of parts with the same shape features in the target task and centrally calculate them in the nesting optimization. Save the matching results to the collection MS; The matching result set MS is used to predict the order and rotation angle of the parts in the target task set on the plate; Take the historical task case with the best matching result as the current source task part set for matching; At the same time, taking the nesting diagram as the source task knowledge, an improved heuristic algorithm for solving the two-dimensional irregular parts nesting problem is adopted, and a hybrid positioning algorithm based on fit and minimum envelope rectangle increment is used to output the optimal nesting results of the target task parts.
2. A variant production knowledge transfer intelligent layout method according to claim 1, characterized in that: The source task knowledge includes original part drawings, historical excellent nesting sequences, nesting part sizes and their numbers; Get the part collection in the target task.
3. The variant production knowledge transfer intelligent layout method according to claim 1, characterized in that: The matching of the parts in the target task with the parts in different source tasks in the information storage module specifically includes: Define the part set in the target task as P0={P 01 , P 02 ,...,P 05 }; The set of parts corresponding to MS is Ps={P s1 , P s2 ,...,P s5 }, match the parts in the set MS and P0 one by one, and let the parts in the target task inherit the arrangement order and rotation angle of the corresponding parts in the set MS, then the arrangement sequence of the parts in the target task P0 on the plate is (P 01 , P 03 , P 05 , P 04 , P 02 ).
4. The variant production knowledge transfer intelligent layout method according to claim 1, characterized in that: In the matching of the target task part and the source task part, for the part matching calculated based on the shape context algorithm, the source task part with the smallest matching distance value is taken as the matching result, and the currently matched source part is hidden.
5. The variant production knowledge transfer intelligent layout method according to claim 1, characterized in that: When the matching distance value is greater than 6.1, it is considered that there is no part in the source task that matches the current target task; The historical task cases with the smallest average distance are used as the current source task parts set for matching.
6. The variant production knowledge transfer intelligent layout method according to claim 1, characterized in that: In the matching of the target task parts and the source task parts, for the parts matching based on the twin neural network, the output distribution is used as the intuitive matching result, and the historical task case with the highest reuse degree of the nesting task is used as the source task part set currently used for matching.
7. A variant production knowledge transfer intelligent layout system, applied to the variant production knowledge transfer intelligent layout method according to any one of claims 1 to 6, characterized in that: include: The information storage module is used to store the original part drawings of the source task, historical excellent nesting sequences, nesting part sizes and their numbers; Part matching module, used to match the target task parts with the source parts in the information storage module, and save the matching results to the set MS; The prediction module uses the set MS to predict the nesting order and rotation angle of the target part, so that the target part inherits the nesting order and rotation angle of the corresponding parts in the set MS; The nesting optimization module adopts an improved heuristic algorithm for solving the two-dimensional irregular parts nesting problem and outputs the optimal nesting result based on a hybrid positioning algorithm based on fit and minimum envelope rectangle increment.
8. The variant production knowledge transfer intelligent nesting system according to claim 7, characterized in that: The parts matching module supports matching methods based on shape context algorithm and twin neural network in the matching process, and follows the rule of prioritizing clustering of similar parts.
9. The variant production knowledge transfer intelligent nesting system according to claim 7, characterized in that: The part matching module uses the historical task cases with the smallest distance to the average value as the source task part set for matching.