Inventory composition decision optimization method based on block part blank cutting process

Abstract

The invention provides an inventory composition decision optimization method based on a block part blank cutting process, and relates to the field of intelligent manufacturing of plate cutting and blanking. The method comprises the following steps: carrying out feature extraction through size features of parts, obtaining a data set, building an Isolation Forest decision model, carrying out model training, inputting features of a to-be-tested raw material into the trained decision model, carrying out feature screening, identifying clustering centers of similar parts by adopting SOM and K-means clustering methods, calculating a raw material loss volume of each layout mode, and calculating a raw material loss volume of each layout mode. According to the method, the block part is subjected to clustering analysis and iterative optimization, population iteration is carried out on the block part as fitness, the layout result is optimized, the layout scheme with the lowest current fitness is selected as the optimal solution, data analysis and machine learning technologies are effectively integrated, the efficiency and effect of the block part cutting process are improved, more accurate resource utilization can be achieved through clustering analysis and iterative optimization, and the method is suitable for large-scale popularization and application. Therefore, raw material waste is reduced, and the material utilization rate is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing of plate cutting and blanking, and in particular to an inventory composition decision optimization method based on a block material part blank cutting process. Background Art

[0002] In the prior art, a large amount of metal rectangular block steel ingots are consumed in the production process of part blanks, and most parts are made of rectangular block blanks. Based on the complexity and diversity of products, rectangular block steel ingots of the same model and specification can be cut to produce rectangular block blank parts with various sizes.

[0003] However, in practice, enterprises adopt a single cutting method, that is, only one size of blank is cut from the same steel ingot. Due to the mismatch between the raw material size and the part blank size, some residual materials are generated, resulting in a decrease in material utilization and affecting the economic benefits of the enterprise. In theory, the material utilization rate can be increased through reasonable part layout. At present, due to some reasons in process and management, most enterprises still have some difficulties in determining reasonable raw material specifications. Summary of the invention

[0004] The main purpose of the present invention is to provide a raw material inventory composition decision optimization method considering the block blank cutting process, so as to solve the current technical problems such as the inability to reasonably cut part blanks according to raw material specifications, resulting in material waste and reduced material utilization.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: an inventory composition decision optimization method based on the block part blank cutting process, The preferred embodiment comprises the following steps: S1: Feature extraction: Obtain the dimensional features of the relevant parts in the raw materials within the preset production cycle, extract the features according to the side length of the parts, and save them into the data set; S2: Model building: Build the Isolation Forest decision model, use the data set to train the model, and retain the optimal decision model; S3: Feature screening: After extracting the features in step S1 from the raw material to be tested, input the optimal decision model, obtain the model output, screen and eliminate according to the preset size threshold, and save those within the preset size threshold range to the second data set; S4: Parts clustering: Use K-means clustering based on SOM to perform cluster analysis on the second data set and obtain the cluster center; S5: Packing iteration: According to the cluster center, the total raw material volume lost in all feasible packing methods of the parts on the raw material is used as the fitness to perform population iteration; S6: Output result: When the population iteration is completed, the current optimal arrangement result is output.

[0006] In the preferred embodiment, step S5 comprises the following steps: S51: Population initialization: The cluster center is used as the initial population, and integer coding is used, that is, according to the size of the raw material, including width and thickness, two adjacent coding bits are used to represent the size of a raw material; S52: Fitness calculation: Calculate all feasible layouts of parts on the raw material and find out the volume of raw materials wasted by all feasible solutions. The parts should ensure that the dimensions of any single rectangular box should be included in the interior of the raw material. The constraint formula is: ; In the formula, , and Indicates the length, width and height of the part. , and Indicates the length, width and height of the raw material; , and for Relative position coordinates in the axis direction; Taking the total raw material volume lost as fitness, the formula is: ; ; In the formula, To waste volume, The sum of wasted volume, is the length, width and height of the raw material, is the length, width and height of the part.

[0007] In the preferred embodiment, the calculation of fitness specifically includes the following steps: A1: Preset the size of the steel ingot to determine the layout of each part to maximize the material utilization rate; A2: Obtain the size data of various rectangular steel ingots in stock, and select the size of the steel ingot to be used for blanking for each rectangular part based on the principle of minimizing the total volume of the steel ingots required for the corresponding parts. The length of the raw material steel ingot can be considered infinite, while the width and thickness are fixed, so as to determine the length, width and thickness of the parts; A3: By calculating the different placement positions of parts on the raw materials, different cutting costs can be obtained. For each part, choose the most suitable specification and place it in the way with the least cost. Add up the volumes of all parts with the least cost, which is the fitness.

[0008] In the preferred embodiment, step S6 also includes designing a crossover operator that crosses only at even-numbered positions and a mutation operator that mutates at 3 or multiples of 3 based on the size characteristics of the raw materials to reduce the range of the solution space. When the nesting result reaches the preset utilization rate or the number of iterations reaches the upper limit, it is the current optimal nesting result.

[0009] In the preferred embodiment, in step S1, the size characteristics of relevant parts in the raw materials within the preset production cycle are extracted according to the side length of the parts, including: sorting the parts according to the size side length to obtain part characteristics, and the part characteristics include the longest side, the second longest side and the shortest side.

[0010] In the preferred solution, in step S3, the Isolation Forest decision model determines whether a part is abnormal based on the part features, and screens the parts by setting the abnormal point ratio parameter.

[0011] In the preferred solution, judging whether a part is abnormal and screening the parts specifically includes: S301: Calculate the expected value of the path length of a data point in the second data set: a data point in the second data set , the size of the second dataset is , record the path length of the data point in the tree , the path length of an isolation tree is ,in, is the expected path length, For the The harmonic series of terms, n is the size of the data set; S302: Calculate the current anomaly score of data point x using the expected value of the path length, using the formula: ; In the formula, is a normalizing constant; S303: If the current anomaly score is greater than the preset anomaly threshold, the current part is abnormal and will be removed during clustering.

[0012] In the preferred embodiment, step S4 includes the following steps: S401: Use SOM to map the topological structure of the second data set and perform fast clustering, wherein: Feature data points of the second dataset , each SOM node has a weight vector , calculate the Euclidean distance between the input data point and all nodes, and find the node BMU corresponding to the minimum distance. SOM updates the weight vector of BMU and its neighboring nodes. The formula is: ; In the formula, Is a node In the The weight vector at the iteration, is the learning rate, is the neighborhood function; S402: Take the weight vector of each node after SOM as input, use K-means clustering to obtain the cluster center, where: select Cluster Center , each cluster center represents a class, and each The Euclidean distance to each cluster center and assign the node to the closest cluster , according to the weight vectors of all nodes in the cluster, iteratively update the mean of the vector of the cluster center. The formula is: ; In the formula, is the weight vector of each node; S403: When the maximum number of iterations is reached, the original data sample is matched with the clustering result, and the clustering result is determined according to the cluster to which the BMU belongs.

[0013] In the preferred solution, the total raw material volume lost in all feasible layout methods of the parts on the raw material is used as the fitness, and the population is iterated to iterate to achieve the optimal layout result, including: S501: Integer coding is used. Starting from the position with coding position index 1, two adjacent codes represent a set of raw material specifications, and all codes are guaranteed to be divisible by 3. S502: using the obtained cluster centers as the initial population of raw material specifications; S503: according to the cutting process during blanking, the volume of the wasted raw materials is calculated by using a fitness operator based on part matching; S504: Preset a crossover operator that crosses over at an even-numbered position, and a mutation operator whose mutation value must be a multiple of 3; S505: Preset an iteration threshold, and output the result when the specified number of iterations is reached or the raw material utilization rate reaches the requirement.

[0014] In the preferred solution, in step S6, the current optimal layout result is output. Considering the material utilization rate, the objective function and constraint conditions are set as follows: B1: The objective function is to minimize the cutting loss, the formula is: ; In the formula, To waste volume, The sum of wasted volume; B2: Constraints include: The constraints required to meet all parts are as follows: ; In the formula, Respectively represent The length, width and height of the rectangular block part and the The length, width and height of a rectangular block of raw material, For the The demand for parts is a binary variable, indicating Is the part from Cut from rectangular blocks of raw material; Binary variable constraints ensure that each part is cut only from raw materials, the formula is: ; Integer variable constraint ensures that the purchase quantity is an integer. The formula is: ; In the formula, is an integer variable representing the Demand for raw materials; The raw material type constraint limits the number of raw material types used to ensure that the raw material size is selected within a certain range. The formula is: ; In the formula, is a Boolean variable indicating whether the jth raw material is used, and n is an integer indicating the upper limit of the number of raw material types allowed.

[0015] The present invention provides a method for optimizing the decision-making of raw material inventory composition in consideration of the block blank cutting process, by extracting features from the size features of the parts, obtaining a data set, then building an Isolation Forest decision model and training the model, then inputting the features of the parts to be tested into the trained decision model, obtaining the model output, and using the SOM and K-means clustering methods after feature screening to perform cluster analysis, thereby evaluating the feasible arrangement mode for each cluster, optimizing the arrangement results, and selecting the arrangement scheme with the lowest current fitness as the optimal solution after the population iteration is completed. The present invention optimizes the inventory composition decision-making method of the block blank cutting process, effectively integrates data analysis and machine learning technologies, improves the efficiency and effect of the block blank cutting process, and can achieve more accurate resource utilization through cluster analysis and iterative optimization, thereby reducing the waste of raw materials, improving material utilization, reducing costs and improving production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 is a flow chart of the overall method of the present invention; Figure 2 It is the difference raw material map of the isolation forest screening of the present invention; Figure 3 It is a K-means clustering diagram based on SOM of the present invention; Figure 4 Schematic diagram of parts placement; Figure 5 It is a schematic diagram of the method for calculating the matching degree of parts of the present invention; Figure 6 It is the individual coding diagram of the present invention; Figure 7 is an example diagram of a crossover operator and a mutation operator according to a preferred embodiment of the present invention; Figure 8 It is an iterative curve diagram of a practical case of application of the present invention. DETAILED DESCRIPTION

[0017] Example 1 This embodiment considers the raw material composition decision problem of the block part blank layout process, that is, to decide on the types and quantities of raw materials required for a production cycle of a mold steel production enterprise, and then considers the production process, through a reasonable raw material inventory plan, to improve the material utilization rate during cutting and reduce inventory costs.

[0018] The layout process theory of rectangular block parts is as follows: Figure 4 As shown in the figure, different rectangular block parts blanks are cut from the raw material rectangular block steel ingot according to different placement methods. Since the length of the raw material steel ingot is much longer than the length of the part blank, the steel ingot can be regarded as infinitely long. At the same time, due to process constraints, multiple parts can only be cut along the length direction from the raw material steel ingot, such as Figure 5 shown.

[0019] like Figure 1-8 As shown, a method for optimizing inventory composition decision based on the blank cutting process of block parts includes the following steps: S1: Feature extraction: Obtain the dimensional features of relevant parts in the raw materials within the preset production cycle, extract features according to the side length of the parts, and save them to the data set.

[0020] S2: Model building: Build an Isolation Forest decision model, use the data set to train the model, and retain the optimal decision model.

[0021] S3: Feature screening: After extracting the features in S1 from the raw materials to be tested, input the optimal decision model, and obtain the model output. Screen and eliminate according to the preset size threshold, and save those within the preset size threshold range to the second data set.

[0022] S4: Part clustering: Use SOM-based K-means clustering to perform cluster analysis on the second data set and obtain the cluster center.

[0023] S5: Packing iteration: Based on the cluster center, the total raw material volume lost in all feasible packing methods of the parts on the raw material is used as the fitness to perform population iteration.

[0024] S6: Output result: When the population iteration is completed, the current optimal arrangement result is output.

[0025] In this embodiment, the Isolation Forest used is an ensemble anomaly detection algorithm based on a binary tree, and is an unsupervised outlier detection method.

[0026] like Figure 1 As shown, this embodiment extracts features from the size features of the parts to obtain a data set, then builds an Isolation Forest decision model and performs model training, then inputs the features of the parts to be tested into the trained decision model to obtain the model output, and uses SOM (Self-Organizing Maps) and K-means clustering methods after feature screening to perform cluster analysis, thereby evaluating the feasible arrangement methods for each cluster, calculating the raw material loss volume of each arrangement method, and using this as the fitness to perform population iteration and optimize the arrangement results. After the population iteration is completed, the arrangement scheme with the lowest current fitness is selected as the optimal solution, thereby realizing the inventory composition decision optimization of the block part blank cutting process, effectively integrating data analysis and machine learning technologies, and improving the efficiency and effect of the block part blank cutting process. Through cluster analysis and iterative optimization, more accurate resource utilization can be achieved, thereby reducing the waste of raw materials, improving material utilization, reducing costs and improving production efficiency.

[0027] In this embodiment, the raw material inventory problem is specifically described as follows: for a part in a production cycle, it is necessary to purchase several rectangular block raw material steel ingots of different specifications. The length, width and height of the steel ingots can be adaptively modified, but they need to meet certain constraints, among which the width and height must be multiples of 3.

[0028] In the preferred embodiment, in step S1, the size characteristics of relevant parts in the raw materials within the preset production cycle are extracted according to the side length of the parts, including: sorting the parts according to the size side length to obtain part characteristics, and the part characteristics include the longest side, the second longest side and the shortest side.

[0029] As shown in Table 1, the dimensional data of all parts in the part set, including the length, width, and height (or the corresponding side length) of each part, are collected, extracted, and saved into a new data set to form the features of the parts, thereby extracting effective features to facilitate subsequent decision model training and analysis.

[0030] Table 1

[0031] The Isolation Forest decision model used in this embodiment is mainly used for outlier detection. The hyperparameters of the model can be optimized through grid search or random search to further improve the performance.

[0032] In the preferred solution, in step S3, the Isolation Forest decision model determines whether a part is abnormal based on the part features, and screens the parts by setting the abnormal point ratio parameter, such as Figure 2 As shown, blue represents normal values ​​and red represents abnormal points. Normal data is filtered out by setting the preset abnormal point ratio.

[0033] In the preferred solution, judging whether a part is abnormal and screening the parts specifically includes: S301: Calculate the expected value of the path length of a data point in the second data set: a data point in the second data set , the size of the second dataset is , record the path length of the data point in the tree , the path length of an isolation tree is ;in, is the expected path length, For the is the number of harmonic series, and n is the size of the data set.

[0034] S302: Calculate the current anomaly score of data point x using the expected value of the path length, using the formula: ; In the formula, is a normalizing constant.

[0035] S303: If the current anomaly score is greater than the preset anomaly threshold, the current part is abnormal and will be removed during clustering.

[0036] Through steps S301-S303, the parts that are abnormal are determined based on the expected value of the path length of the Isolation Forest model. A preset threshold, such as 0.5, can be set for screening. Finally, the abnormal parts are deleted and the remaining part data is saved as the second data set, thereby effectively identifying and screening out abnormal parts with large dimensional deviations, thereby optimizing subsequent production processes.

[0037] In the preferred embodiment, step S4 includes the following steps: S401: Use SOM to map the topological structure of the second data set and perform fast clustering, wherein: Feature data points of the second dataset , each SOM node has a weight vector , calculate the Euclidean distance between the input data point and all nodes, and find the node BMU corresponding to the minimum distance. SOM updates the weight vector of BMU and its neighboring nodes. The formula is: ; In the formula, Is a node In the The weight vector at the iteration, is the learning rate, is a neighborhood function.

[0038] S402: Take the weight vector of each node after SOM as input, use K-means clustering to obtain the cluster center, where: select Cluster Center , each cluster center represents a class, and each The Euclidean distance to each cluster center and assign the node to the closest cluster , according to the weight vectors of all nodes in the cluster, iteratively update the mean of the vector of the cluster center. The formula is: ; In the formula, is the weight vector of each node.

[0039] S403: When the maximum number of iterations is reached, the original data sample is matched with the clustering result, and the clustering result is determined according to the cluster to which the BMU belongs.

[0040] like Figure 3The figure shows the composite clustering process of SOM and K-means clustering. First, the self-SOM is used to perform feature mapping and dimensionality reduction on the size data of the parts, so that similar parts are clustered in a low-dimensional space. Then, the K-means algorithm is applied to cluster the mapped data and divide the parts into multiple clusters, and the parts in each cluster have similar size features. The above steps help to improve clustering efficiency and accuracy when processing large-scale data sets, effectively map the topological structure of the data and quickly cluster it, providing support for subsequent data analysis and optimization.

[0041] In this embodiment, in the layout iteration, the fitness function can be defined as the total raw material volume lost, and the goal is to minimize this loss and ensure that the material utilization rate is maximized, such as Figure 4 Shown are various ways of placing parts in raw materials. Figure 5 This embodiment adopts the method of carrying out along the side length.

[0042] In the preferred embodiment, step S5 comprises the following steps: S51: Population initialization: The cluster center is used as the initial population, and integer coding is used, that is, according to the size of the raw material, including width and thickness, two adjacent coding bits are used to represent the size of a raw material.

[0043] S52: Fitness calculation: Calculate all feasible layouts of parts on the raw material and find out the volume of raw materials wasted by all feasible solutions. The parts should ensure that the dimensions of any single rectangular box should be included in the interior of the raw material. The constraint formula is: ; In the formula, , and Indicates the length, width and height of the part. , and Indicates the length, width and height of the raw material; , and for Axis direction relative position coordinates.

[0044] Taking the total raw material volume lost as fitness, the formula is: ; ; In the formula, To waste volume, The sum of wasted volume, is the length, width and height of the raw material, is the length, width and height of the part.

[0045] like Figure 6 The individual encoding method shown is optimized through the genetic algorithm to effectively initialize the population and calculate the fitness, providing a basis for the subsequent selection, crossover and mutation operations of the genetic algorithm.

[0046] In the preferred embodiment, the calculation of fitness includes the following steps: A1: Preset the size of the steel ingot and determine the layout plan for each part to maximize the material utilization rate.

[0047] A2: Obtain the size data of various rectangular steel ingots in stock, and select the size of the steel ingot to be used for cutting for each rectangular block part based on the principle of minimizing the total volume of the steel ingots required to meet the needs of the corresponding parts; among them, the length of the raw material steel ingot can be regarded as infinite, and the width and thickness are fixed, thereby determining the length, width and thickness of the parts.

[0048] A3: By calculating the different placement positions of parts on the raw materials, different cutting costs can be obtained. For each part, choose the most suitable specification and place it in the way with the least cost. Add up the volumes of all parts with the least cost, which is the fitness.

[0049] Through the above steps, the calculated fitness value represents the overall performance of the selected layout scheme. The smaller the fitness, the higher the utilization rate of raw materials. The performance of different layout schemes is effectively evaluated, and the material utilization rate is optimized in the genetic algorithm.

[0050] In this embodiment, the length, width and height of the rectangular raw material steel ingots of different specifications can be adaptively changed, where the width and height must meet the multiples of 3, and the length can be selected arbitrarily. It is necessary to plan raw material steel ingots of different specifications to meet the parts processing within a production cycle and ensure the utilization rate of materials, so as to set a unique crossover operator and mutation operator to optimize the raw material arrangement results, such as Figure 4 Schematic diagram of the matching degree calculation method given.

[0051] In the preferred embodiment, step S6 also includes designing a crossover operator that crosses only at even-numbered positions and a mutation operator that mutates at 3 or multiples of 3 based on the size characteristics of the raw materials to reduce the range of the solution space. When the nesting result reaches the preset utilization rate or the number of iterations reaches the upper limit, it is the current optimal nesting result.

[0052] In the preferred solution, the total raw material volume lost in all feasible layout methods of the parts on the raw material is used as the fitness, and the population is iterated to iterate to achieve the optimal layout result, including: S501: Integer coding is used. Starting from the coding position index 1, the adjacent two-digit codes represent a set of raw material specifications, and it is ensured that all codes are divisible by 3.

[0053] S502: Using the obtained cluster centers as the initial population of raw material specifications.

[0054] S503: According to the cutting process during blanking, the volume of the wasted raw materials is calculated using a fitness operator based on part matching.

[0055] S504: A crossover operator is preset to cause crossover at an even-numbered position, and the mutation value must be a multiple of 3.

[0056] S505: Preset an iteration threshold, and output the result when the specified number of iterations is reached or the raw material utilization rate reaches the requirement.

[0057] like Figure 7 Shown are examples of the crossover operator and mutation operator used in this embodiment.

[0058] Through the iterative process, the layout plan is optimized and the optimal result required is output.

[0059] In this embodiment, in order to minimize the total cutting cost, it is necessary to design an objective function and constraint conditions to optimize the nesting results. The following is a detailed setting.

[0060] In the preferred solution, in step S6, the current optimal layout result is output. Considering the material utilization rate, the objective function and constraint conditions are set as follows: B1: The objective function is to minimize the cutting loss, the formula is: ; In the formula, To waste volume, The sum of wasted volume.

[0061] B2: Constraints include: 1) The constraints required to meet all parts are as follows: ; In the formula, Respectively represent The length, width and height of the rectangular block part and the The length, width and height of a rectangular block of raw material, For the The demand for parts is a binary variable, indicating Is the part from Rectangular blocks are cut from the raw material.

[0062] 2) Binary variable constraints ensure that each part is cut only from raw materials, the formula is: .

[0063] 3) Integer variable constraint, ensure that the purchase quantity is an integer, the formula is: ; In the formula, is an integer variable representing the The demand for raw materials.

[0064] 4) Constraint on the number of raw material types, which limits the number of raw material types used and ensures that the raw material size is selected within a certain range. The formula is: ; In the formula, is a Boolean variable, indicating whether the j-th raw material is used, with values ​​of 1 for use and 0 for not use); n is an integer, indicating the upper limit of the number of raw material types allowed.

[0065] By solving this optimization problem, we can obtain the layout plan that minimizes the total cutting cost, ensuring that the requirements of all parts are met.

[0066] As shown in Table 2 below, it is a small-scale case verification data table of this embodiment.

[0067] Table 2

[0068] In this embodiment, all cases achieved the 80% material utilization rate required by the enterprise.

[0069] The following is an actual large-scale case verification. Table 3 below shows the raw material specifications required within a production cycle and the actual utilization rate obtained after adopting the optimization method of this embodiment.

[0070] Table 3 Raw material specifications and their actual utilization rate within a production cycle

[0071] The raw material data [228,135] in Table 3 indicates a width of 228 cm and a thickness of 135 cm.

[0072] The implementation process of this embodiment is as follows: 1) Obtain the dimensional features of the relevant parts in the raw materials in Table 3, extract the features according to the side length of the parts, and save them in the data set.

[0073] 2) Input the optimal decision model, obtain the model output, and filter and remove the outliers according to the preset outlier threshold of 0.005. The outliers are represented as Figure 2 Click the red triangle and save it to the second data set.

[0074] 3) Use K-means based on SOM to perform joint clustering analysis and obtain the cluster center. The clustering results are as follows: Figure 3 shown.

[0075] 4) Perform iterative sorting according to the cluster center, such as Figure 8 shown.

[0076] 5) Output results: When the population iteration is completed, the current optimal nesting result is output. The nesting result includes 70 raw material sizes, as shown in Table 3, which raw material each part corresponds to, and the placement of the part in its corresponding raw material.

[0077] Figure 8 It is an iteration curve diagram of an actual case, in which the horizontal axis represents the number of iterations and the vertical axis represents the volume of the raw materials wasted. As the number of iterations increases, the total volume of the wasted materials continues to decrease, and after reaching the preset number of iterations of 20, it tends to converge, effectively proving the feasibility and advantages of this embodiment in the actual production process.

[0078] When in use, this embodiment converts the part information in the raw materials into part features, and screens and clusters the parts based on the part features. The cluster center of the clustering result is used as the initial population of the evolutionary algorithm. A part matching algorithm is proposed to calculate the material volume wasted on raw materials of different specifications, and an evolutionary iterative method that matches the problem is designed to obtain the final inventory plan.

[0079] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for optimizing inventory composition decision based on the blank cutting process of block parts, characterized in that: The following steps are involved: S1: Feature extraction: Obtain the dimensional features of the relevant parts in the raw materials within the preset production cycle, extract the features according to the side length of the parts, and save them into the data set; S2: Model building: Build the Isolation Forest decision model, use the data set to train the model, and retain the optimal decision model; S3: Feature screening: After extracting the features in step S1 from the raw material to be tested, input the optimal decision model, obtain the model output, screen and eliminate according to the preset size threshold, and save those within the preset size threshold range to the second data set; S4: Parts clustering: Use K-means clustering based on SOM to perform cluster analysis on the second data set and obtain the cluster center; S5: Packing iteration: According to the cluster center, the total raw material volume lost in all feasible packing methods of the parts on the raw material is used as the fitness to perform population iteration; S6: Output result: When the population iteration is completed, the current optimal arrangement result is output.

2. The inventory composition decision optimization method based on the block part blank cutting process according to claim 1 is characterized in that: Step S5 includes the following steps: S51: Population initialization: The cluster center is used as the initial population, and integer coding is used, that is, according to the size of the raw material, including width and thickness, two adjacent coding bits are used to represent the size of a raw material; S52: Fitness calculation: Calculate all feasible layouts of parts on the raw material and find out the volume of raw materials wasted by all feasible solutions. The parts should ensure that the dimensions of any single rectangular box should be included in the interior of the raw material. The constraint formula is: ； In the formula, , and Indicates the length, width and height of the part. , and Indicates the length, width and height of the raw material; , and for Relative position coordinates in the axis direction; Taking the total raw material volume lost as fitness, the formula is: ； ； In the formula, To waste volume, The sum of wasted volume, is the length, width and height of the raw material, is the length, width and height of the part.

3. The inventory composition decision optimization method based on the block part blank cutting process according to claim 2 is characterized in that: The calculation of fitness specifically includes the following steps: A1: Preset the size of the steel ingot to determine the layout of each part to maximize the material utilization rate; A2: Obtain the size data of various rectangular steel ingots in stock, and select the size of the steel ingot to be used for blanking for each rectangular part based on the principle of minimizing the total volume of the steel ingots required for the corresponding parts. The length of the raw material steel ingot can be considered infinite, while the width and thickness are fixed, so as to determine the length, width and thickness of the parts; A3: By calculating the different placement positions of parts on the raw materials, different cutting costs can be obtained. For each part, choose the most suitable specification and place it in the way with the least cost. Add up the volumes of all parts with the least cost, which is the fitness.

4. The inventory composition decision optimization method based on the block part blank cutting process according to claim 1 is characterized in that: Step S6 also includes designing a crossover operator that crosses only at even-numbered positions and a mutation operator that mutates at 3 or multiples of 3 based on the size characteristics of the raw materials to reduce the range of the solution space. When the nesting result reaches the preset utilization rate or the number of iterations reaches the upper limit, it is the current optimal nesting result.

5. The inventory composition decision optimization method based on the block part blank cutting process according to claim 1 is characterized in that: In step S1, the size features of the relevant parts in the raw materials within the preset production cycle are extracted according to the side length of the parts, including: sorting the parts according to the size side length to obtain part features, and the part features include the longest side, the second longest side and the shortest side.

6. The inventory composition decision optimization method based on the block part blank cutting process according to claim 1 is characterized in that: In step S3, the Isolation Forest decision model determines whether a part is abnormal based on its features and screens the parts by setting the abnormal point ratio parameter.

7. The inventory composition decision optimization method based on the block part blank cutting process according to claim 6 is characterized in that: Determine whether the parts are abnormal and screen the parts, including: S301: Calculate the expected value of the path length of a data point in the second data set: a data point in the second data set , the size of the second dataset is , record the path length of the data point in the tree , the path length of an isolation tree is ,in, is the expected path length, For the The harmonic series of terms, n is the size of the data set; S302: Calculate the current anomaly score of data point x using the expected value of the path length, using the formula: ； In the formula, is an outlier, is a normalizing constant; S303: If the current anomaly score is greater than the preset anomaly threshold, the current part is abnormal and will be removed during clustering.

8. The inventory composition decision optimization method based on the block part blank cutting process according to claim 1 is characterized in that: Step S4 includes the following steps: S401: Use SOM to map the topological structure of the second data set and perform fast clustering, wherein: Feature data points of the second dataset , each SOM node has a weight vector , calculate the Euclidean distance between the input data point and all nodes, and find the node BMU corresponding to the minimum distance. SOM updates the weight vector of BMU and its neighboring nodes. The formula is: ； In the formula, Is a node In the The weight vector at the iteration, is the learning rate, is the neighborhood function; S402: Take the weight vector of each node after SOM as input, use K-means clustering to obtain the cluster center, where: select Cluster Center , each cluster center represents a class, and each The Euclidean distance to each cluster center and assign the node to the closest cluster , according to the weight vectors of all nodes in the cluster, iteratively update the mean of the vector of the cluster center. The formula is: ； In the formula, is the weight vector of each node; S403: When the maximum number of iterations is reached, the original data sample is matched with the clustering result, and the clustering result is determined according to the cluster to which the BMU belongs.

9. The inventory composition decision optimization method based on the block part blank cutting process according to claim 8 is characterized in that: The total raw material volume lost in all possible layout methods of the parts on the raw material is used as the fitness, and the population is iterated to achieve the optimal layout result, including: S501: Integer coding is used. Starting from the position with coding position index 1, two adjacent codes represent a set of raw material specifications, and all codes are guaranteed to be divisible by 3. S502: using the obtained cluster centers as the initial population of raw material specifications; S503: according to the cutting process during blanking, the volume of the wasted raw materials is calculated by using a fitness operator based on part matching; S504: Preset a crossover operator that crosses over at an even-numbered position, and a mutation operator whose mutation value must be a multiple of 3; S505: Preset an iteration threshold, and output the result when the specified number of iterations is reached or the raw material utilization rate reaches the requirement.

10. The inventory composition decision optimization method based on the block part blank cutting process according to claim 9 is characterized in that: In step S6, the current optimal layout result is output. Considering the material utilization rate, the objective function and constraint conditions are set as follows: B1: The objective function is to minimize the cutting loss, the formula is: ； In the formula, To waste volume, The sum of wasted volume; B2: Constraints include: The constraints required to meet all parts are as follows: ； In the formula, Respectively represent The length, width and height of the rectangular block part and the The length, width and height of a rectangular block of raw material, For the The demand for parts is a binary variable, indicating Is the part from Cut from rectangular blocks of raw material; Binary variable constraints ensure that each part is cut only from raw materials, the formula is: ； Integer variable constraint ensures that the purchase quantity is an integer. The formula is: ； In the formula, is an integer variable representing the Demand for raw materials; The raw material type constraint limits the number of raw material types used to ensure that the raw material size is selected within a certain range. The formula is: ； In the formula, is a Boolean variable indicating whether the jth raw material is used, and n is an integer indicating the upper limit of the number of raw material types allowed.

Images