Intelligent trepanning combination optimization method and system based on deep learning
Through the intelligent nesting combination optimization method based on deep learning, deep learning technology is used to extract and encode raw materials, and a neural network model is designed for joint training, which solves the problem of cumbersome calculation and error-prone calculation of the existing nesting combination method, and realizes efficient and accurate nesting combination, reducing resource waste.
Patent Information
- Application Number
- CN202311092796.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-08-28
- Publication Date
- 2025-02-28
AI Technical Summary
The existing nesting material combination method is single, the calculation is cumbersome and error-prone, and it cannot meet customer needs, resulting in waste of resources.
Using the intelligent nesting combination optimization method based on deep learning, the raw materials are extracted and encoded through deep learning technology, and an adaptive neural network model is designed, and the hidden vector representation of the raw materials is used for joint training with the product parameters to automatically calculate the best nesting combination scheme.
An automated calculation of nesting combination scheme is realized, which improves the accuracy of calculation results and nesting efficiency, and reduces resource waste.
Smart Images
Figure CN119538692A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of nesting combination methods, and in particular to an intelligent nesting combination optimization method and system based on deep learning. Background Art
[0002] At present, in the sheet metal processing of industries such as shipbuilding, machinery, and construction, nesting combination is a key link to achieve efficient use of raw materials and reduce resource waste. The existing nesting method is single, and the nesting combination scheme is usually calculated manually. The calculation process is cumbersome and prone to errors. The actual effect is far from user requirements and cannot meet customer needs. Although some automatic nesting has been achieved, the nested sheets are not satisfactory in terms of sheet metal utilization, rationality, and cutting sequence, and manual intervention is required before they can be put into the next production link. For example, in manual cutting and mechanical cutting, the cutting workers cut in sequence according to the order of the product blanking list, and it is difficult to consider the optimization of nesting, which inevitably produces a large amount of scraps and causes waste of raw materials. In addition, there are differences in processing requirements, raw material properties, etc. for different products. Therefore, designing an intelligent nesting combination optimization method and system to automatically calculate the nesting combination scheme and optimize the nesting combination plays an important role in improving the efficiency and accuracy of the nesting combination. Summary of the invention
[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and provide an intelligent nesting combination optimization method and system based on deep learning.
[0004] The purpose of the present invention is achieved through the following technical solutions: A method for intelligent nesting combination optimization based on deep learning, comprising the following steps:
[0005] (1) Use deep learning technology to extract and encode features of different raw materials to obtain latent vector representations of the raw materials;
[0006] (2) Design an adaptive neural network model whose input is the product parameters and output is the nesting plan;
[0007] (3) The neural network model uses the latent vector representation of the raw materials and the parameters of the products for joint training. By learning from a large amount of data, the rules and optimization strategies for nesting combinations are obtained.
[0008] (4) Based on the trained neural network model, the new nesting problem is solved to obtain the optimal nesting combination solution.
[0009] A further technical solution is that in step (1), a multi-layer neural network is constructed to simulate the learning process of the human brain to achieve high-level abstraction and feature extraction of data, including data preparation, network construction, feature extraction, feature encoding, and latent vector representation.
[0010] A further technical solution is that the latent vector of the raw material converts the properties of the raw material into a specific vector representation, and the latent vector serves as the input of the neural network model.
[0011] A further technical solution is that the neural network model includes a neural network structure with multiple levels, each level has multiple neuron nodes, and a deep neural network structure is formed by continuously stacking multiple levels.
[0012] A further technical solution is that the parameters of the product include size, size tolerance, shape and position tolerance, and surface roughness.
[0013] A further technical solution is that in the process of joint training of the neural network model in step (3), the latent vector representation of the raw materials is combined with the parameters of the product, and the relationship between the raw materials and the product is modeled by learning and adjusting the weights of the parameters. A known nesting scheme is used as training data as input, and the corresponding effect is used as output. Through a large amount of training data and a back-propagation algorithm, the neural network model learns the relationship between different material combinations and corresponding effects.
[0014] A further technical solution is that the neural network model trained in step (4) evaluates various material combinations based on the previously learned relationship, and combines other algorithms or heuristic methods to obtain the best nesting combination solution.
[0015] An intelligent nesting combination optimization system based on deep learning, the optimization system adopts an intelligent nesting combination optimization method based on deep learning, and the optimization system includes the following modules: data collection module, data preprocessing module, deep learning model training, nesting combination calculation module, nesting combination optimization module, and nesting execution module.
[0016] A further technical solution is that the data preprocessing module includes data cleaning, data integration and fusion, data transformation, and data reduction, and the nesting combination calculation module includes statistical classification and typesetting nesting.
[0017] The present invention has the following advantages:
[0018] The present invention proposes an intelligent nesting combination optimization method and system based on deep learning. The deep learning technology is used to extract and encode the features of the raw materials, and joint training is performed based on a neural network model and product parameters to obtain the optimal nesting combination scheme. The present invention can automatically calculate the nesting combination scheme, has a high degree of automation, and realizes optimal selection, which effectively improves the accuracy of the calculation results and the nesting efficiency, and reduces resource waste. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] Figure 1 It is a workflow diagram of the intelligent nesting combination optimization method based on deep learning in the present invention;
[0020] Figure 2 It is a structural diagram of the intelligent nesting combination optimization system based on deep learning in the present invention;
[0021] Figure 3 It is a workflow diagram of the data preprocessing module in the present invention. DETAILED DESCRIPTION
[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be described clearly and completely below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention described and shown herein can be arranged and designed in various different configurations.
[0023] Therefore, the following detailed description of the embodiments of the present invention is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0024] like Figure 1 As shown, the present invention discloses an intelligent nesting combination optimization method based on deep learning, comprising the following steps:
[0025] (1) Use deep learning technology to extract and encode features of different raw materials to obtain latent vector representations of the raw materials. The specific processing process is as follows:
[0026] Data preparation: Collect and prepare data sets of raw materials, such as images, text, or sensor data, ensure that data samples are diverse and representative, and perform preprocessing such as data cleaning, normalization, and denoising on the collected data to facilitate subsequent feature extraction and encoding.
[0027] Network construction: Design and build a deep neural network model suitable for feature extraction and encoding tasks. You can choose convolutional neural network (CNN) or recurrent neural network (RNN) to extract features from raw data. For image data, CNN is preferred to extract visual features of the image; for text data, RNN is preferred to extract semantic features.
[0028] Feature extraction: The raw data is input, and after forward propagation, feature representation is obtained in the middle layer or output layer of the network. The feature can be a high-level abstract semantic feature or a low-level original feature such as vision or sound.
[0029] Feature encoding: The extracted features are converted into latent vector representations, and the features are encoded using methods such as autoencoders, variational autoencoders, or generative adversarial networks, mapping high-dimensional features into low-dimensional latent vector representations while retaining important information of the original data.
[0030] Latent vector application: The obtained latent vector representation of raw materials is used for subsequent tasks, such as clustering, retrieval, classification, etc. Common applications include raw material similarity calculation based on latent vector representation and the construction of recommendation system. The latent vector of raw materials can convert the properties of raw materials into specific vector representations. Each raw material has a corresponding latent vector, which can be used as input into the neural network model during the calculation process, so that the characteristics and attribute information of the raw materials can be processed and learned by the model in a unified way.
[0031] (2) By designing an adaptive neural network model, the product parameters are taken as input, and the model outputs the nesting plan. The specific processing process is as follows:
[0032] Product data collection and preprocessing: Collect data samples containing product parameters and corresponding nesting plans. Product parameters can include information such as size, shape, material, etc. The nesting plan is a layout diagram or specific cutting / assembly instructions; preprocess the collected data, including data cleaning, normalization, denoising, etc., to facilitate subsequent neural network model training.
[0033] In this embodiment, the parameters of the product include:
[0034] Size: divided into qualitative and positioning size;
[0035] Dimensional tolerance: fit tolerances need to be noted in the general assembly drawing, and tolerances need to be noted in the product drawing;
[0036] Shape and position tolerance: Indicated in the product drawing, including coaxiality, cylindricity, flatness, etc. The shapes include bar, trapezoid, and preset product specifications;
[0037] Surface roughness requirements, processing symbols and surface roughness;
[0038] Technical requirements and instructions: Generally, it refers to the requirements for heat treatment and other methods of oxidation and rust prevention of processed parts.
[0039] Feature engineering: Based on the characteristics of product parameters, feature engineering is performed to convert product parameters into a form suitable for neural network model input, such as normalizing product dimensions or using domain knowledge to construct features.
[0040] Neural network model design: Select an appropriate neural network model, such as a feedforward neural network or a convolutional neural network, and design a suitable network structure and number of layers according to the complexity of the problem and the characteristics of the data. The deep learning model is a neural network structure composed of multiple levels, each of which has multiple neuron nodes. By continuously stacking multiple levels, a deep neural intelligent nesting combination optimization method network structure is formed. After the data of the products and raw materials to be processed are transmitted to the deep learning model, the deep learning model calculates and combines different product parameters to generate a nesting combination plan that saves the most raw materials.
[0041] Model training: Use the prepared data samples to train the neural network model, take product parameters as input, and nesting schemes as labels for supervised learning, and continuously adjust model parameters to improve prediction accuracy.
[0042] Model evaluation and optimization: Use test data to evaluate the trained neural network model and calculate the prediction accuracy and performance; adjust and optimize the model based on the evaluation results, such as adjusting hyperparameters, increasing the amount of training data, introducing regularization, etc.
[0043] Nesting plan prediction: Use the trained neural network model to pass the input product parameters into the model for inference and output the corresponding nesting plan.
[0044] (3) The neural network model uses the latent vector representation of the raw materials and the parameters of the product for joint training. By learning from a large amount of data, it automatically learns the rules and optimization strategies for nesting combinations.
[0045] During the joint training process, the neural network model combines the latent vector representation of the raw materials with the parameters of the product. By learning and adjusting the weights of the parameters, the relationship between the raw materials and the products is modeled. The known nesting scheme is used as training data as input, and the corresponding effect is used as output. Through a large amount of training data and back-propagation algorithm, the neural network model learns the relationship between different material combinations and corresponding effects.
[0046] In this embodiment, according to the generated preliminary nesting combination plan, a frame is drawn in the CAD interface according to the outer dimensions of the raw materials, and the generated nesting combination graphic is inserted into the frame; in order to improve the utilization rate and cutting quality of the plate, it is necessary to set the gaps between each block and between the block and the edge of the plate. The gap between the blocks is set to 10mm, the distance between the block and the edge of the plate is 10mm, and the gap between the blocks with bevels is 15mm. The gap between the blocks and the distance between the two pieces and the edge of the plate can also be adjusted according to actual needs.
[0047] After the optimization of the nesting combination plan is completed, it is necessary to define the cutting order and the starting position of the part cutting. Before cutting, it is also necessary to determine the type and angle of the guide line of the current raw material to be cut. When cutting on the plate, cut small blocks first in a clockwise direction, then large blocks; cut the inner control blocks first, then the outer blocks; cut long strips first, then square blocks.
[0048] (4) Based on the trained neural network model, the new nesting problem is solved to obtain the optimal nesting combination solution.
[0049] The nesting solution to be solved is input into the trained neural network model. The model will evaluate various material combinations based on the previously learned relationships and give the best nesting combination solution. In order to obtain more accurate results, other algorithms or heuristic methods, such as genetic algorithms and simulated annealing, can be combined with the neural network model for collaborative solution to obtain a better nesting combination solution.
[0050] like Figure 2 As shown, the present invention proposes an intelligent nesting combination optimization system based on deep learning. The optimization system adopts an intelligent nesting combination optimization method based on deep learning. The optimization system includes the following modules: data collection module, data preprocessing module, deep learning model training, nesting combination calculation module, nesting combination optimization module, nesting execution module, wherein the functions of each module are as follows:
[0051] Data collection module: used to collect product parameters and raw material data, including product shape, size, material and other parameters and raw material size, material and other data;
[0052] Data preprocessing module: preprocesses the collected data, including data cleaning, normalization and other operations, to prepare for deep learning models;
[0053] Deep learning model training: Use preprocessed data to train a deep learning model that can output nesting combination solutions based on input product parameters and raw material data;
[0054] Nesting combination calculation module: used to automatically calculate the preliminary nesting combination plan through the deep learning model;
[0055] Nesting combination optimization module: used to optimize the preliminary nesting combination plan according to the optimization criteria and generate the final nesting combination plan.
[0056] Nesting execution module: Cut the plates according to the optimized nesting combination plan to produce the actual processed products.
[0057] In this embodiment, the process of data preprocessing after data collection is as follows: Figure 3 As shown, it includes operations such as data cleaning, data integration and fusion, data transformation, and data reduction.
[0058] In this embodiment, the structure of the deep learning model selects a convolutional neural network (CNN), which is a deep learning model for image processing and is particularly suitable for tasks such as image classification and target detection. CNN has multiple convolutional layers and pooling layers, and realizes effective classification and recognition of images by continuously extracting high-level features and reducing resolution; the model selection optimizer adopts the Adam optimizer, which is good at processing image classification problems, can process more complex models and can update all parameters at the same time; conducts combination experiments according to different parameters, data sets, network structures, etc., and finds the best combination; uses training data to train the model, performs multiple iterations, evaluates the performance of the model and optimizes it, and obtains satisfactory model performance by adjusting parameters or modifying the network structure.
[0059] In this embodiment, the nesting combination calculation module includes statistical classification and typesetting nesting. The commonly used typesetting nesting methods are as follows:
[0060] Rectangular layout method: suitable for the case where the plate shape is regular, such as rectangle, square, etc.;
[0061] Circular layout method: It is suitable for situations where circular workpieces need to be arranged. It is necessary to calculate the diameter and area of each circular workpiece and arrange the circular workpieces reasonably on the plate;
[0062] Irregular shape layout method: For irregular shaped workpieces, computer-aided design software is usually used for nesting layout.
[0063] In this embodiment, the nesting execution module includes CAD software and a calling code compilation subroutine to automatically convert the nesting diagram into numerical control code.
[0064] The present invention provides an intelligent nesting combination optimization method and system based on deep learning, which can automatically calculate the nesting combination scheme and realize the optimization selection, thereby effectively improving the nesting efficiency and reducing the waste of resources; the present invention has a high degree of automation, reduces manual intervention, and effectively improves the accuracy of the calculation results.
[0065] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. An intelligent nesting combination optimization method based on deep learning, characterized in that: The following steps are involved: (1) Use deep learning technology to extract and encode features of different raw materials to obtain latent vector representations of the raw materials; (2) Design an adaptive neural network model whose input is the product parameters and output is the nesting plan; (3) The neural network model uses the latent vector representation of the raw materials and the parameters of the products for joint training. By learning from a large amount of data, the rules and optimization strategies for nesting combinations are obtained. (4) Based on the trained neural network model, the new nesting problem is solved to obtain the optimal nesting combination solution.
2. The intelligent nesting combination optimization method based on deep learning according to claim 1 is characterized in that: In the step (1), a multi-layer neural network is constructed to simulate the learning process of the human brain, thereby achieving high-level abstraction and feature extraction of data, including data preparation, network construction, feature extraction, feature encoding, and latent vector representation.
3. The intelligent nesting combination optimization method based on deep learning according to claim 1 is characterized in that: The latent vector of the raw material converts the properties of the raw material into a specific vector representation, and the latent vector serves as the input of the neural network model.
4. The intelligent nesting combination optimization method based on deep learning according to claim 1, characterized in that: The neural network model includes a neural network structure with multiple levels, each level has multiple neuron nodes, and a deep neural network structure is formed by continuously stacking multiple levels.
5. The intelligent nesting combination optimization method based on deep learning according to claim 1, characterized in that: The parameters of the product include size, dimensional tolerance, shape and position tolerance, and surface roughness.
6. The intelligent nesting combination optimization method based on deep learning according to claim 1, characterized in that: In the step (3), during the joint training process, the neural network model combines the latent vector representation of the raw materials with the parameters of the products, and models the relationship between the raw materials and the products by learning and adjusting the weights of the parameters. A known nesting scheme is used as training data as input, and the corresponding effect is used as output. Through a large amount of training data and a back-propagation algorithm, the neural network model learns the relationship between different material combinations and corresponding effects.
7. The intelligent nesting combination optimization method based on deep learning according to claim 1 is characterized in that: The neural network model trained in step (4) evaluates various material combinations based on the previously learned relationships, and combines other algorithms or heuristic methods to obtain the best nesting combination solution.
8. An intelligent nesting combination optimization system based on deep learning, characterized by: The optimization system adopts the deep learning-based intelligent nesting combination optimization method described in any one of claims 1-7, and the optimization system includes the following modules: data collection module, data preprocessing module, deep learning model training, nesting combination calculation module, nesting combination optimization module, and nesting execution module.
9. The intelligent nesting combination optimization system based on deep learning according to claim 8, characterized in that: The data preprocessing module includes data cleaning, data integration and fusion, data transformation, and data reduction, and the nesting combination calculation module includes statistical classification and typesetting nesting.