Intelligent bending and edge covering system and method for solid wood customized furniture

Abstract

The invention discloses an intelligent bending and edge covering system and method for solid wood customized furniture, and belongs to the technical field of furniture manufacturing. Comprising a data acquisition module, a material analysis module, a demand analysis module, a design optimization module, an identification positioning module, a path planning module, an edge covering strategy module, a production control module, a quality detection module, a production simulation module and a data storage module. According to the method, the number of parameter combinations needing to be evaluated is reduced, so that calculation and time cost is reduced, a more comprehensive bending path planning scheme can be provided, all design requirements are ensured to be met, the system can quickly adapt to different production requirements and process requirements, the universality and adaptability of the system are improved, and the system performance is improved. The system can flexibly cope with different production conditions and requirements, errors caused by common empirical and subjective judgment in a traditional method can be avoided, material waste is reduced, the use efficiency of materials is improved, and therefore the production cost is reduced, and economic benefits are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of furniture manufacturing, and in particular to an intelligent bending and edging system and method for custom-made solid wood furniture. Background Art

[0002] With the continuous development of the furniture industry, consumers' demand for personalized and high-quality furniture is growing. Solid wood custom furniture has always been regarded as a representative of high-end furniture due to its unique natural texture and elegant appearance. With the increasing demand of consumers for personalized furniture, the customized furniture market is growing rapidly. However, the production process of solid wood furniture faces many challenges, especially in the bending and hemming process. There are many challenges in the traditional production process. The bending and hemming process relies on manual operation, which is inefficient and unstable in quality. It is difficult to meet the needs of modern production. Its labor cost is high, production efficiency is low, and process complexity is high, especially in the bending and hemming process. These problems are particularly prominent. Bending and hemming process is a key link that affects the quality and appearance of solid wood furniture. It is difficult for traditional methods to ensure consistency and precision.

[0003] The existing intelligent bending and edging systems and methods for solid wood custom furniture require a large number of parameter combinations to be evaluated, with high calculation and time costs, and are unable to provide a more comprehensive bending path planning solution. The system has low versatility and adaptability. In addition, the existing intelligent bending and edging systems and methods for solid wood custom furniture have errors caused by the common experience and subjective judgment in traditional methods, which increases material waste and reduces material utilization efficiency. To this end, we propose an intelligent bending and edging system and method for solid wood custom furniture. Summary of the invention

[0004] The purpose of the present invention is to solve the defects in the prior art and to propose an intelligent bending and edging system and method for solid wood custom furniture.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] An intelligent bending and hemming system for solid wood custom furniture, comprising a data acquisition module, a material analysis module, a demand analysis module, a design optimization module, an identification and positioning module, a path planning module, an hemming strategy module, a production control module, a quality inspection module, a production simulation module and a data storage module;

[0007] The data acquisition module is used to collect various physical property data of solid wood materials.

[0008] As a further solution of the present invention, the physical property data of solid wood materials collected by the data acquisition module specifically include density, hardness, humidity, texture characteristics, elastic modulus, bending strength, compressive strength, toughness, shrinkage, color difference, oil content, thermal conductivity, sound absorption coefficient, specific gravity, wood type and chemical composition.

[0009] The material analysis module is used to analyze the collected material data and construct a material property map.

[0010] As a further solution of the present invention, the material analysis module constructs a material property map in the following specific steps:

[0011] Q1.1: Clean, normalize and standardize the various physical property data collected by the data acquisition module to remove noise and outliers. Treat each piece of solid wood material as a node, and the node feature vector is composed of pre-processed physical property data. Define the edges between nodes based on the similarity between materials, and the weight of the edge represents the degree of similarity;

[0012] Q1.2: Input the feature vector of each node and the weight matrix of the edge as input data into the pre-trained graph neural network model, collect the features of each group of node i’s neighbor nodes j∈Π(i), and perform feature aggregation, where Π(i) represents the set of node i’s neighbor nodes. Then, fuse the node’s own features with the aggregated neighbor features, and update the node’s feature vector through a set of nonlinear activation functions.

[0013] Q1.3: Repeat feature aggregation and feature update, gradually integrate more distant nodes’ information through multi-layer graph convolution, and the output features of each layer will become the input features of the next layer. After passing through K layers of graph convolution, the output node i integrates the information of the node itself and its multi-layer neighbor nodes, and finally represents the node embedding.

[0014] Q1.4: Embedding nodes Input into a set of regression models, obtain the predicted value of node i through the regression model, calculate the loss value between the predicted value and the actual value through the RMSE loss function, and obtain the gradient of the RMSE loss function for each layer of the regression model based on the calculated loss value. Then, through the back propagation algorithm, pass the gradient back to each layer and update the parameters of each layer. Through multiple iterative training, continuously update the model parameters until the loss function variation range reaches the preset convergence threshold;

[0015] Q1.5: The graph neural network model collects real-time data on various physical properties of solid wood materials, and processes them through the graph convolution layer to obtain real-time node embedding. Then, the predicted value of node i is obtained through the trained regression model, and it is output as the predicted material service life.

[0016] As a further solution of the present invention, the specific calculation formula of feature aggregation described in Q1.2 is as follows:

[0017]

[0018] In the formula, represents the neighbor feature aggregation result of node i at layer k; Π(i) represents the set of neighbor nodes of node i; c ij Represents the normalized coefficient between node i and its neighbor node j; Represents the feature vector of node j at layer k;

[0019] The specific calculation formula for updating the feature vector of the node described in Q1.2 is as follows:

[0020]

[0021] In the formula, represents the feature vector of node i in the k+1 layer; σ represents the nonlinear activation function, usually ReLU; W (k) Represents the weight matrix of the kth layer; Represents the feature vector of node i in the kth layer.

[0022] The demand analysis module is used to integrate customer needs, market trends and historical order data, analyze customer customization needs, and generate the best design solution;

[0023] The design optimization module is used to balance material cost, production efficiency and aesthetics, and optimize the generated design solution;

[0024] The identification and positioning module is used to accurately identify and position the solid wood material.

[0025] As a further solution of the present invention, the identification and positioning module performs the following specific steps to accurately identify and position the solid wood material:

[0026] Q2.1: Collect image data containing solid wood materials, annotate the solid wood materials in the images, generate a training data set, use the image data in the training data set as input images, input them into the pre-trained convolutional neural network, extract high-level feature representations, and obtain corresponding feature maps;

[0027] Q2.2: The extracted feature map is transformed through three different linear transformations to generate the query matrix, key matrix, and value matrix. The dot product of the query matrix and the key matrix is ​​calculated, and the calculated dot product is scaled. The softmax function is used to obtain the attention weights.

[0028] Q2.3: According to the preset number of attention heads, the query matrix, key matrix and value matrix are divided into equal parts, and then the query matrix, key matrix and value matrix are reshaped respectively. For each attention head, the attention is calculated using the corresponding query matrix, key matrix and value matrix;

[0029] Q2.4: Concatenate the outputs of all attention heads together, generate the final output through linear transformation, concatenate the features processed by the multi-head attention mechanism with the original features, and use the matching algorithm to find the position of the solid wood material in the cascade feature map. At the same time, output the position coordinates of the solid wood material in the image and its bounding box.

[0030] The path planning module is used to plan the bending path according to the material shape and position data.

[0031] As a further solution of the present invention, the specific steps of planning the bending path by the path planning module are as follows:

[0032] Q3.1: Randomly generate multiple groups of bending paths as the initial population, where each individual in the population represents a group of bending paths. Calculate each group of bending paths P based on path length, processing time and energy consumption. l The fitness value f(P l ), and then the individuals with fitness higher than the preset threshold are selected to enter the next generation through the roulette wheel selection method;

[0033] Q3.2: Randomly select two sets of parent individuals P from the population a With P b , randomly select multiple crossover points in the gene sequence of the individual, exchange the gene fragments of the parent individuals at each crossover point to generate new offspring individuals, then randomly select one or more groups of individuals from the current population, and randomly select one or more mutation points in the gene sequence of the individual, and perturb the gene value according to the Gaussian distribution;

[0034] Q3.3: Randomly generate the initial position and initial speed of each group of individuals in the initial population, calculate the fitness value of the current position of each group of individuals, and compare the fitness value of the current position of each group of individuals with the fitness value of the historical optimal position of the individual. If the current fitness value is better than the historical optimal fitness value, update the individual's optimal position to the current position;

[0035] Q3.4: Use the Gaussian process as a proxy model, fit the existing evaluation data, and randomly select a portion of individuals for real evaluation to obtain the corresponding fitness values ​​as training data. Then use the proxy model to calculate the expected improvement values ​​of all candidate individuals, and select the individual with the largest expected improvement value for actual experiments or simulations to obtain its true fitness value.

[0036] Q3.5: Add the newly evaluated individuals and their fitness values ​​to the training data set, update the proxy model, update the individual information according to the true evaluated fitness value, adjust the population, and then use the new true fitness value to update the individual's speed and position;

[0037] Q3.6: Repeat iterations and update the population until the preset maximum number of generations or the fitness change value is within the preset convergence interval. Select the bending path with the highest fitness value from the final population as the optimal solution, and apply it to the actual bending and hemming operation. At the same time, dynamically adjust the bending path according to the actual bending situation.

[0038] The hemming strategy module is used to produce the optimal hemming strategy based on real-time bending data;

[0039] The production control module is used to detect the production process in real time and adjust the production parameters;

[0040] The quality inspection module is used to perform quality inspection on finished products and identify potential quality defects;

[0041] The production simulation module is used to generate a three-dimensional model of the furniture and simulate the entire production process;

[0042] The data storage module is used to store all production data, test data and optimization parameters.

[0043] An intelligent bending and hemming method for solid wood custom furniture, the specific steps of the analysis method are as follows:

[0044] Ⅰ: Collect solid wood material data to build material property maps and identify materials;

[0045] Ⅱ: Capture images of solid wood materials and extract image feature data;

[0046] III: Plan the bending path according to the material characteristics and identification results;

[0047] IV: Design the hemming scheme and simulate the actual hemming process to verify its feasibility and effect;

[0048] V: Optimize the hemming scheme and use the intelligent bending and hemming system for actual operation;

[0049] Ⅵ: Evaluate the quality of the completed bent and edged furniture and make adjustments.

[0050] As a further solution of the present invention, the specific steps of the optimized hemming solution in step V are as follows:

[0051] Q4.1: Generate an initial hemming plan based on material properties and bending path, take the current hemming plan as the root node, and the results of different hemming strategies as child nodes, and build a corresponding search tree based on the root node and each group of child nodes;

[0052] Q4.2: Calculate the upper limit of the confidence interval of each child node, select the child node with the highest upper limit of the confidence interval from the root node, check whether the child node has any child nodes that have not been fully explored, if so, expand the node, if not, the current child node is considered to be fully expanded, and return to the previous level node, continue traversal, repeat selection and expansion until the preset depth or leaf node is reached;

[0053] Q4.3: Start from the selected child node, perform simulation, randomly select an executable action, execute the selected action to transfer to the next state, and update the current node state. Calculate the benefit of the current node through the potential function, continue simulation until reaching the preset depth or the terminated leaf node, and record the current path and the corresponding benefit;

[0054] Q4.4: Starting from the leaf node, trace the nodes and corresponding benefits in the simulation path back to the root node step by step, and for each node and action in the path, update its visit count, and update the average benefit of each node and action in the path;

[0055] Q4.5: Use the updated node benefits to calculate the potential function of each node, determine the benefit increment based on the calculated potential function value, and update the wrapping parameters in the wrapping scheme. If the update of any wrapping parameter causes the potential function to increase, keep the update, otherwise, restore the original state and try to update other variables;

[0056] Q4.6: Repeat the search and update of the wrapping parameters in the wrapping solution until the potential function change value is within the preset convergence range, then terminate the search and update, and start from the root node, select the optimal path with the highest benefit value in the current search tree as the optimal wrapping solution, record each wrapping parameter and its corresponding value in the optimal solution, and output the wrapping solution.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1. The present invention randomly generates multiple groups of bending paths as the initial population, calculates the fitness value of each group of bending paths according to the path length, processing time and energy consumption, and then selects individuals with fitness higher than a preset threshold to enter the next generation through a roulette selection method, randomly selects two groups of parent individuals from the population, performs crossover and mutation operations, randomly generates the initial position and initial speed of each group of individuals in the initial population, and calculates the fitness value of the current position of each group of individuals, compares the fitness value of the current position of each group of individuals with the fitness value of the historical optimal position of the individual, and if the current fitness value is better than the historical optimal fitness value, then the optimal position of the individual is updated to the current position, and the Gaussian process is used as a proxy model to fit the existing evaluation data, and a part of individuals are randomly selected for real evaluation to obtain the corresponding fitness value as training data, and then the proxy model is used to calculate the expected improvement value of all candidate individuals, and select The individual with the largest expected improvement value is selected for actual experiment or simulation to obtain its true fitness value, the newly evaluated individual and its fitness value are added to the training data set, the proxy model is updated, the individual information is updated according to the true evaluated fitness value, the population is adjusted, and then the speed and position of the individual are updated using the new true fitness value, the iteration is repeated, and the population is updated until the preset maximum number of generations or the fitness change value is reached within the preset convergence interval, and the bending path with the highest fitness value is selected from the final population as the optimal solution, and it is applied to the actual bending and hemming operation. At the same time, according to the actual bending situation, the bending path is dynamically adjusted to reduce the number of parameter combinations that need to be evaluated, thereby reducing the calculation and time costs, and being able to provide a more comprehensive bending path planning solution to ensure that all design requirements are met, so that the system can quickly adapt to different production requirements and process requirements, and improve the versatility and adaptability of the system.

[0059] 2. The present invention takes the current edge wrapping scheme as the root node and the results of different edge wrapping strategies as child nodes, and constructs a corresponding search tree based on the root node and each group of child nodes, calculates the upper limit value of the confidence interval of each child node, and selects the child node with the highest upper limit value of the confidence interval starting from the root node, and checks whether the child node has child nodes that have not been fully explored. If so, the node is expanded, and the selection and expansion are repeated until the preset depth or leaf node is reached. Starting from the selected child node, simulation is performed, and the nodes and corresponding benefits in the simulation path are gradually traced back to the root node, and each node and action visit times are updated. The potential function of each node is calculated using the updated node benefit, and the benefit increment is determined based on the calculated potential function value, and the edge wrapping is updated at the same time. For the hemming parameters in the scheme, if the update of any hemming parameter leads to an increase in the potential function, the update is retained; otherwise, the original state is restored and the update of other variables is tried, and the search and update of the hemming parameters in the hemming scheme are repeated until the change value of the potential function is within the preset convergence interval. Then the search and update are terminated, and starting from the root node, the optimal path with the highest benefit value in the current search tree is selected as the optimal hemming scheme, and each hemming parameter and its corresponding value in the optimal scheme are recorded, and the hemming scheme is output, so that the system can flexibly respond to different production conditions and requirements, which helps to avoid the errors caused by the common experience and subjective judgment in traditional methods, reduce material waste, and improve the utilization efficiency of materials, thereby reducing production costs and improving economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0061] Figure 1 This is a system block diagram of an intelligent bending and edging system for solid wood custom-made furniture proposed by the present invention;

[0062] Figure 2 This is a flowchart of an intelligent bending and edging method for solid wood custom furniture proposed by the present invention. DETAILED DESCRIPTION

[0063] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0064] Example 1

[0065] Reference Figure 1, an intelligent bending and hemming system for solid wood custom furniture, including a data acquisition module, a material analysis module, a demand analysis module, a design optimization module, an identification and positioning module, a path planning module, an hemming strategy module, a production control module, a quality inspection module, a production simulation module and a data storage module.

[0066] The data acquisition module is used to collect various physical property data of solid wood materials.

[0067] It should be further explained that the physical property data of solid wood materials collected by the data acquisition module specifically include density, hardness, moisture, texture characteristics, elastic modulus, bending strength, compressive strength, toughness, shrinkage, color difference, oil content, thermal conductivity, sound absorption coefficient, specific gravity, wood type and chemical composition.

[0068] The material analysis module is used to analyze the collected material data and construct material property maps.

[0069] Specifically, all kinds of physical property data collected by the data acquisition module are cleaned, normalized and standardized to remove noise and outliers. Each piece of solid wood material is regarded as a node, and the node feature vector is composed of preprocessed physical property data. The edges between nodes are defined according to the similarity between materials, and the weight of the edge represents the degree of similarity. The feature vector of each node and the weight matrix of the edge are input as input data into the pre-trained graph neural network model, and the features of each group of node i neighbor nodes j∈Π(i) are collected and feature aggregation is performed, where Π(i) represents the set of node i neighbor nodes. The node's own features are then fused with the aggregated neighbor features, and the node's feature vector is updated through a set of nonlinear activation functions. Feature aggregation and feature updating are repeated, and the information of more distant nodes is gradually integrated through multi-layer graph convolutions, and the output features of each layer will become the input features of the next layer. After passing through K layers of graph convolution, the final node embedding representation of the output node i that integrates the information of the node itself and its multi-layer neighbor nodes is output. Embed the node It is input into a set of regression models, and the predicted value of node i is obtained through the regression model. The loss value between the predicted value and the actual value is calculated through the RMSE loss function, and the gradient of the RMSE loss function for each layer of the regression model is obtained based on the calculated loss value. Then, the gradient is passed back to each layer through the back propagation algorithm, and the parameters of each layer are updated. Through multiple iterative training, the model parameters are continuously updated until the loss function variation range reaches the preset convergence threshold. The graph neural network model collects real-time physical property data of various solid wood materials, and processes it through the graph convolution layer to obtain real-time node embedding. Then, the predicted value of node i is obtained through the trained regression model, and it is output as the predicted material service life.

[0070] In this embodiment, the specific calculation formula for feature aggregation is as follows:

[0071]

[0072] In the formula, represents the neighbor feature aggregation result of node i at layer k; Π(i) represents the set of neighbor nodes of node i; c ij Represents the normalized coefficient between node i and its neighbor node j; Represents the feature vector of node j at layer k;

[0073] The specific calculation formula for updating the node's feature vector is as follows:

[0074]

[0075] In the formula, represents the feature vector of node i in the k+1 layer; σ represents the nonlinear activation function, usually ReLU; W (k) Represents the weight matrix of the kth layer; Represents the feature vector of node i in the kth layer.

[0076] The demand analysis module is used to integrate customer needs, market trends and historical order data, analyze the customer's customization needs to generate the optimal design solution; the design optimization module is used to weigh material costs, production efficiency and aesthetics, and optimize the generated design solution; the identification and positioning module is used to accurately identify and locate solid wood materials.

[0077] Specifically, image data containing solid wood materials are collected, the solid wood materials in the images are labeled, and a training data set is generated. The image data in the training data set is used as an input image and input into a pre-trained convolutional neural network to extract high-level feature representations and obtain corresponding feature maps. The extracted feature maps are transformed through three different linear transformations to generate query matrices, key matrices, and value matrices. The dot product of the query matrix and the key matrix is ​​calculated, and the calculated dot product is scaled. The attention weights are obtained through the softmax function. According to the preset number of attention heads, the query matrix, key matrix, and value matrix are divided into equal parts, and then the query matrix, key matrix, and value matrix are reshaped respectively. For each attention head, the corresponding query matrix, key matrix, and value matrix are used to calculate the attention. The outputs of all attention heads are spliced ​​together, and then the final output is generated through linear transformation. The features processed by the multi-head attention mechanism are cascaded with the original features, and the matching algorithm is used to find the position of the solid wood material in the cascade feature map, and the position coordinates of the solid wood material in the image and its bounding box are output.

[0078] The path planning module is used to plan the bending path according to the material shape and position data.

[0079] Specifically, multiple groups of bending paths are randomly generated as the initial population, where the individuals in the population represent a group of bending paths, and the P of each group of bending paths is calculated based on the path length, processing time and energy consumption. l The fitness value f(P l ), and then select individuals with fitness higher than the preset threshold to enter the next generation through the roulette wheel selection method, and randomly select two groups of parent individuals P from the population a With P b , randomly select multiple crossover points in the gene sequence of the individual, exchange the gene fragments of the parent individuals at each crossover point to generate new offspring individuals, then randomly select one or more groups of individuals from the current population, and randomly select one or more mutation points in the gene sequence of the individual, perturb the gene value according to the Gaussian distribution, randomly generate the initial position and initial speed of each group of individuals in the initial population, and calculate the fitness value of the current position of each group of individuals, compare the fitness value of the current position of each group of individuals with the fitness value of the historical optimal position of the individual, if the current fitness value is better than the historical optimal fitness value, then update the individual's optimal position to the current position, use the Gaussian process as a proxy model, fit the existing evaluation data, and randomly select a part of individuals for true Real evaluation is performed to obtain the corresponding fitness value as training data, and then the proxy model is used to calculate the expected improvement values ​​of all candidate individuals, and the individual with the largest expected improvement value is selected for actual experiment or simulation to obtain its real fitness value, and the newly evaluated individual and its fitness value are added to the training data set, and the proxy model is updated. The individual information is updated according to the real evaluated fitness value, and the population is adjusted. Then, the speed and position of the individual are updated with the new real fitness value, and the iteration is repeated to update the population until the preset maximum number of generations or the fitness change value is within the preset convergence interval. The bending path with the highest fitness value is selected from the final population as the optimal solution, and it is applied to the actual bending and hemming operation. At the same time, the bending path is dynamically adjusted according to the actual bending situation.

[0080] The hemming strategy module is used to produce the optimal hemming strategy based on real-time bending data; the production control module is used to detect the production process in real time and adjust the production parameters; the quality inspection module is used to inspect the quality of finished products and identify potential quality defects; the production simulation module is used to generate a three-dimensional model of furniture and simulate the entire production process; the data storage module is used to store all production data, inspection data and optimization parameters.

[0081] Example 2

[0082] Reference Figure 2 , an intelligent bending and hemming method for solid wood custom furniture, the specific steps of the analysis method are as follows:

[0083] Ⅰ: Collect solid wood material data to construct material property maps and identify materials.

[0084] Ⅱ: Capture images of solid wood materials and extract image feature data.

[0085] III: Plan the bending path based on the material characteristics and identification results.

[0086] IV: Design a hemming scheme and simulate the actual hemming process to verify its feasibility and effectiveness.

[0087] V: Optimize the hemming solution and use the intelligent bending and hemming system for actual operation.

[0088] Specifically, according to the material characteristics and the bending path, an initial hemming scheme is generated, the current hemming scheme is used as the root node, and the results of different hemming strategies are used as child nodes. Based on the root node and each group of child nodes, a corresponding search tree is constructed, and the confidence interval upper limit value of each child node is calculated. Starting from the root node, the child node with the highest confidence interval upper limit value is selected, and the child node is checked to see if there are any child nodes that have not been fully explored. If so, the node is expanded. If not, the current child node is regarded as fully expanded and returns to the previous level node. The traversal continues, and the selection and expansion are repeated until the preset depth or leaf node is reached. Starting from the selected child node, the simulation is performed, and an executable action is randomly selected. The selected action is executed to transfer to the next state, and the current node state is updated. The benefit of the current node is calculated through the potential function, and the simulation is continued until the preset depth or the terminated leaf node is reached, and the record is recorded. The current path and the corresponding benefits, starting from the leaf node, will gradually trace back the nodes and corresponding benefits in the simulation path to the root node, and for each node and action in the path, update its visit count, and update the average benefit of each node and action in the path, use the updated node benefits to calculate the potential function of each node, determine the benefit increment based on the calculated potential function value, and update the wrapping parameters in the wrapping scheme. If the update of any wrapping parameter causes the potential function to increase, then retain the update, otherwise, restore the original state and try to update other variables, repeat the search and update the wrapping parameters in the wrapping scheme until the potential function change value is within the preset convergence interval, then terminate the search and update, and start from the root node, select the optimal path with the highest benefit value of the current search tree as the optimal wrapping scheme, record each wrapping parameter and its corresponding value in the optimal scheme, and output the wrapping scheme.

[0089] Ⅵ: Evaluate the quality of the completed bent and edged furniture and make adjustments.

Claims

1. An intelligent bending and edging system for solid wood custom furniture, characterized in that: It includes data acquisition module, material analysis module, demand analysis module, design optimization module, identification and positioning module, path planning module, hemming strategy module, production control module, quality inspection module, production simulation module and data storage module; The data acquisition module is used to collect various physical property data of solid wood materials; The material analysis module is used to analyze the collected material data and construct a material property map; The demand analysis module is used to integrate customer needs, market trends and historical order data, analyze customer customization needs, and generate the best design solution; The design optimization module is used to balance material cost, production efficiency and aesthetics, and optimize the generated design solution; The identification and positioning module is used to accurately identify and position the solid wood material; The path planning module is used to plan the bending path according to the material shape and position data; The hemming strategy module is used to produce the optimal hemming solution based on real-time bending data; The production control module is used to detect the production process in real time and adjust the production parameters; The quality inspection module is used to perform quality inspection on finished products and identify potential quality defects; The production simulation module is used to generate a three-dimensional model of the furniture and simulate the entire production process; The data storage module is used to store all production data, test data and optimization parameters.

2. The intelligent bending and edging system for solid wood custom furniture according to claim 1, characterized in that: The physical property data of solid wood materials collected by the data acquisition module specifically include density, hardness, moisture, texture characteristics, elastic modulus, bending strength, compressive strength, toughness, shrinkage, color difference, oil content, thermal conductivity, sound absorption coefficient, specific gravity, wood type and chemical composition.

3. The intelligent bending and edging system for solid wood custom furniture according to claim 2, characterized in that: The specific steps of constructing the material property map by the material analysis module are as follows: Q1.1: Clean, normalize and standardize the various physical property data collected by the data acquisition module to remove noise and outliers. Treat each piece of solid wood material as a node, and the node feature vector is composed of pre-processed physical property data. Define the edges between nodes based on the similarity between materials, and the weight of the edge represents the degree of similarity; Q1.2: Input the feature vector of each node and the weight matrix of the edge as input data into the pre-trained graph neural network model, collect the features of each group of node i’s neighbor nodes j∈Π(i), and perform feature aggregation, where Π(i) represents the set of node i’s neighbor nodes. Then, fuse the node’s own features with the aggregated neighbor features, and update the node’s feature vector through a set of nonlinear activation functions. Q1.3: Repeat feature aggregation and feature update, gradually integrate more distant nodes’ information through multi-layer graph convolution, and the output features of each layer will become the input features of the next layer. After passing through K layers of graph convolution, the output node i integrates the information of the node itself and its multi-layer neighbor nodes, and finally represents the node embedding. Q1.4: Embedding nodes Input into a set of regression models, obtain the predicted value of node i through the regression model, calculate the loss value between the predicted value and the actual value through the RMSE loss function, and obtain the gradient of the RMSE loss function for each layer of the regression model based on the calculated loss value. Then, through the back propagation algorithm, pass the gradient back to each layer and update the parameters of each layer. Through multiple iterative training, continuously update the model parameters until the loss function variation range reaches the preset convergence threshold; Q1.5: The graph neural network model collects real-time data on various physical properties of solid wood materials, and processes them through the graph convolution layer to obtain real-time node embedding. Then, the predicted value of node i is obtained through the trained regression model, and it is output as the predicted material service life.

4. The intelligent bending and edging system for solid wood custom furniture according to claim 3, characterized in that: The specific steps of the identification and positioning module for accurately identifying and positioning solid wood materials are as follows: Q2.1: Collect image data containing solid wood materials, annotate the solid wood materials in the images, generate a training data set, use the image data in the training data set as input images, input them into the pre-trained convolutional neural network, extract high-level feature representations, and obtain corresponding feature maps; Q2.2: The extracted feature map is transformed through three different linear transformations to generate the query matrix, key matrix, and value matrix. The dot product of the query matrix and the key matrix is ​​calculated, and the calculated dot product is scaled. The softmax function is used to obtain the attention weights. Q2.3: According to the preset number of attention heads, the query matrix, key matrix and value matrix are divided into equal parts, and then the query matrix, key matrix and value matrix are reshaped respectively. For each attention head, the attention is calculated using the corresponding query matrix, key matrix and value matrix; Q2.4: Concatenate the outputs of all attention heads together, generate the final output through linear transformation, concatenate the features processed by the multi-head attention mechanism with the original features, and use the matching algorithm to find the position of the solid wood material in the cascade feature map. At the same time, output the position coordinates of the solid wood material in the image and its bounding box.

5. The intelligent bending and edging system for solid wood custom furniture according to claim 4, characterized in that: The specific steps of the path planning module for planning the bending path are as follows: Q3.1: Randomly generate multiple groups of bending paths as the initial population, where each individual in the population represents a group of bending paths. Calculate each group of bending paths P based on path length, processing time and energy consumption. l The fitness value f(P l ), and then the individuals with fitness higher than the preset threshold are selected to enter the next generation through the roulette wheel selection method; Q3.2: Randomly select two sets of parent individuals P from the population a With P b , randomly select multiple crossover points in the gene sequence of the individual, exchange the gene fragments of the parent individuals at each crossover point to generate new offspring individuals, then randomly select one or more groups of individuals from the current population, and randomly select one or more mutation points in the gene sequence of the individual, and perturb the gene value according to the Gaussian distribution; Q3.3: Randomly generate the initial position and initial speed of each group of individuals in the initial population, calculate the fitness value of the current position of each group of individuals, and compare the fitness value of the current position of each group of individuals with the fitness value of the historical optimal position of the individual. If the current fitness value is better than the historical optimal fitness value, update the individual's optimal position to the current position; Q3.4: Use the Gaussian process as a proxy model, fit the existing evaluation data, and randomly select a portion of individuals for real evaluation to obtain the corresponding fitness values ​​as training data. Then use the proxy model to calculate the expected improvement values ​​of all candidate individuals, and select the individual with the largest expected improvement value for actual experiments or simulations to obtain its true fitness value. Q3.5: Add the newly evaluated individuals and their fitness values ​​to the training data set, update the proxy model, update the individual information according to the true evaluated fitness value, adjust the population, and then use the new true fitness value to update the individual's speed and position; Q3.6: Repeat iterations and update the population until the preset maximum number of generations or the fitness change value is within the preset convergence interval. Select the bending path with the highest fitness value from the final population as the optimal solution, and apply it to the actual bending and hemming operation. At the same time, dynamically adjust the bending path according to the actual bending situation.

6. An intelligent bending and edging method for solid wood custom furniture, used to realize the intelligent bending and edging system function for solid wood custom furniture as claimed in any one of claims 1 to 5, characterized in that: The specific steps of this analysis method are as follows: Ⅰ: Collect solid wood material data to build material property maps and identify materials; Ⅱ: Capture images of solid wood materials and extract image feature data; III: Plan the bending path according to the material characteristics and identification results; IV: Design the hemming scheme and simulate the actual hemming process to verify its feasibility and effect; V: Optimize the hemming scheme and use the intelligent bending and hemming system for actual operation; Ⅵ: Evaluate the quality of the completed bent and edged furniture and make adjustments.

7. The intelligent bending and edging method for solid wood custom furniture according to claim 6, characterized in that: The specific steps of optimizing the hemming scheme described in step V are as follows: Q4.1: Generate an initial hemming plan based on material properties and bending path, take the current hemming plan as the root node, and the results of different hemming strategies as child nodes, and build a corresponding search tree based on the root node and each group of child nodes; Q4.2: Calculate the upper limit of the confidence interval of each child node, select the child node with the highest upper limit of the confidence interval from the root node, check whether the child node has any child nodes that have not been fully explored, if so, expand the node, if not, the current child node is considered to be fully expanded, and return to the previous level node, continue traversal, repeat selection and expansion until the preset depth or leaf node is reached; Q4.3: Start from the selected child node, perform simulation, randomly select an executable action, execute the selected action to transfer to the next state, and update the current node state. Calculate the benefit of the current node through the potential function, continue simulation until reaching the preset depth or the terminated leaf node, and record the current path and the corresponding benefit; Q4.4: Starting from the leaf node, trace the nodes and corresponding benefits in the simulation path back to the root node step by step, and for each node and action in the path, update its visit count, and update the average benefit of each node and action in the path; Q4.5: Use the updated node benefits to calculate the potential function of each node, determine the benefit increment based on the calculated potential function value, and update the wrapping parameters in the wrapping scheme. If the update of any wrapping parameter causes the potential function to increase, keep the update, otherwise, restore the original state and try to update other variables; Q4.6: Repeat the search and update of the wrapping parameters in the wrapping solution until the potential function change value is within the preset convergence range, then terminate the search and update, and start from the root node, select the optimal path with the highest benefit value in the current search tree as the optimal wrapping solution, record each wrapping parameter and its corresponding value in the optimal solution, and output the wrapping solution.

Images