Intelligent packaging line production management system and method

Abstract

The invention relates to the technical field of production management, and discloses an intelligent packaging line production management system and method, and the method comprises the steps: obtaining a product information vector V1 of a to-be-boxed product and a box body information vector V2 of a target packaging box; dividing the to-be-boxed products into three types of products according to the product information vector V1; generating a three-dimensional geometric model and a virtual box body model, and constructing a three-dimensional boxing simulation environment; an optimal boxing scheme is obtained through a boxing scheme learning model; the packaging efficiency and the product safety can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of production management, and more specifically, to an intelligent packaging line production management system and method. Background Art

[0002] With the rapid development of the packaging industry, intelligent packaging production lines are increasingly being used in industrial production. Intelligent packaging lines can effectively improve packaging efficiency, reduce labor costs, and ensure the consistency of product packaging quality. However, existing intelligent packaging technology still has some shortcomings and it is difficult to fully meet the flexibility requirements of different products and different packaging requirements.

[0003] The Chinese patent with publication number CN116552890A discloses a packaging production line and packaging method, which includes a bagging module, a labeling module and a boxing module, and the connection and control of each module are realized through a conveyor line and a monitoring system; although the invention solves the problems of high difficulty and low efficiency in operation, it lacks an intelligent classification and packing optimization mechanism for different products. The Chinese patent with authorization announcement number CN109573199B discloses a control method for an e-commerce intelligent box selection and packaging production line, which selects appropriate cartons according to the size of the goods and calculates the packing sequence and buffer filling amount; but this method is mainly aimed at the e-commerce field, and does not fully consider the packing requirements of industrial products such as robots, such as the material and fragility of the product. The Chinese patent with authorization announcement number CN112977985B discloses an intelligent transportation packaging line, which realizes the automatic adjustment of the packaging process through visual inspection and speed feedback control. Although the invention improves the intelligence level of the packaging line, it lacks fine-grained division of different types of workpieces to be packaged, and the optimization space of the packing scheme is limited.

[0004] In summary, the existing packing needs for robot-related products urgently require an intelligent packaging management system and method that can comprehensively consider the geometric and physical properties of products, reasonably classify different products, automatically build packing plans and dynamically optimize them. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides an intelligent packaging line production management system and method, which divides the products to be packed into different categories by obtaining the product information vector of the products to be packed and the box information vector of the target packaging box, and obtains the optimal packing solution by using a machine learning model in a three-dimensional simulation environment.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] Intelligent packaging line production management methods, including:

[0008] Obtain product information vector V1 of the product to be packed and box body information vector V2 of the target packaging box;

[0009] According to the product information vector V1 of the product to be packed, the products to be packed are divided into a first product category C1, a second product category C2 and a third product category C3;

[0010] According to the product information vector V1, a three-dimensional geometric model is generated for each product to be packed in the first product category C1, the second product category C2 and the third product category C3; according to the box information vector V2, a virtual box model adapted to the internal dimensions of the target packaging box is generated; according to the three-dimensional geometric model of the product to be packed and the virtual box model, a three-dimensional packing simulation environment is constructed;

[0011] In a 3D packing simulation environment, the optimal packing solution is obtained through the pre-built packing solution learning model.

[0012] Furthermore, the step of obtaining the product information vector V1 of the product to be packed includes:

[0013] Acquire first geometric parameters of the products to be packed, wherein the first geometric parameters include the length L1, width W1 and height H1 of each product to be packed;

[0014] Acquire first physical property parameters of the products to be packed, where the first physical property parameters include weight M1, material and fragility F1 of each product to be packed;

[0015] Parameter association is performed on the acquired first geometric parameter and first physical property parameter of the product to be packed, so as to generate a product information vector V1 of the product to be packed.

[0016] Furthermore, the step of obtaining the box body information vector V2 of the target packaging box includes:

[0017] Acquire second geometric parameters of the target packaging box, wherein the second geometric parameters include an inner length L2, an inner width W2, and an inner height H2 of the target packaging box;

[0018] Obtain a second physical property parameter of the target packaging box, wherein the second physical property parameter includes a maximum bearing weight M2 of the target packaging box max and compressive strength P2;

[0019] The acquired second geometric parameter and second physical property parameter of the target packaging box are parameter-associated to generate a box body information vector V2 of the target packaging box.

[0020] Further, the dividing the products to be packed into the first product category C1, the second product category C2 and the third product category C3 includes:

[0021] Extract features from the product information vector V1 to obtain the volume S1, weight M1 and fragility F1 of each product to be packed;

[0022] Preliminarily classify the products to be packed according to their volume S1 and weight M1 to obtain n1 preliminary product categories, where n1 is an integer greater than 0;

[0023] Each preliminary product category is further divided according to the fragility F1 of the product to be packed, and n2 subdivided product categories are obtained, n2 ≥ n1;

[0024] The n2 subdivided product categories are divided into a first product category C1, a second product category C2 and a third product category C3.

[0025] Furthermore, the further classification of each preliminary product category according to the fragility F1 of the product to be packed includes:

[0026] For each preliminary product category, calculate the mean vulnerability level F1 according to the vulnerability level F1 of the products in it. mean and vulnerability variance F1 var ;

[0027] If the fragility variance F1 var Greater than the preset variance threshold var thr , then the preliminary product category is subdivided into k subcategories, k ≥ n2;

[0028] All subcategories obtained from the preliminary product category segmentation are merged to form n2 subdivided product categories.

[0029] Furthermore, the subcategories obtained by combining all preliminary product categories include:

[0030] Each subcategory is regarded as an independent cluster, and the distance between any two clusters is calculated, where the distance between any two clusters is the absolute value of the difference between the mean values ​​of the vulnerability of the two clusters;

[0031] Among all current clusters, find the two closest clusters C p and C q , merged into a new cluster C r ;

[0032] Update the merged cluster C r The distance to other existing clusters;

[0033] Repeat the merging operation until the number of clusters is reduced to n2.

[0034] Furthermore, the method of dividing the n2 subdivided product categories into the first product category C1, the second product category C2 and the third product category C3 is:

[0035] For each product category, extract its vulnerability mean F1 mean and size mean S1 mean Features, set the vulnerability threshold F1 θ , size threshold S1 θ ;

[0036] If F1 mean ≥F1 θ And S1 mean ≥S1 θ , then the subdivided product category is divided into the first product category C1;

[0037] If F1 mean <F1 θ And S1 mean <S1 θ , then the subdivided product category is divided into the second product category C2;

[0038] If F1 mean ≥F1 θ And S1 mean <S1 θ , or F1 mean <F1 θ And S1 mean ≥S1 θ , then the subdivided product category is divided into the third product category C3.

[0039] Furthermore, obtaining the optimal packing solution through the pre-built packing solution learning model includes:

[0040] In a three-dimensional packing simulation environment, a pre-built packing solution learning model is used to obtain a first packing solution;

[0041] Obtaining a first constraint condition of the packing process, introducing the first constraint condition into a packing solution learning model, and obtaining a second packing solution that satisfies the first constraint condition;

[0042] According to the second packing plan, the comprehensive physical property parameters after packing are calculated, and the comprehensive physical property parameters after packing are compared with the second physical property parameters of the target packaging box to determine whether the second packing plan meets the packing safety requirements; if not, the packing safety requirements are used as the second constraint condition, and the second constraint condition is introduced into the packing plan learning model to obtain the third packing plan; the third packing plan is used as the optimal packing plan.

[0043] Furthermore, the first constraint condition includes a weight balance threshold ωb , Placement gap threshold d g 、The upper limit of stacking layers N s and the total constraint cost C total ; Total constraint cost C total Based on the weight balance threshold ω b , Placement gap threshold d g And the upper limit of stacking layers N s calculated;

[0044] The comprehensive physical property parameters include the total weight M of the packed goods c and the compressive stress P of the bottom layer of cargo b ;

[0045] The packing safety requirements are: c ≤M2 max And P b ≤P2.

[0046] An intelligent packaging line production management system, which is used to implement the above-mentioned intelligent packaging line production management method, the system includes:

[0047] Vector extraction module: used to obtain the product information vector V1 of the product to be packed and the box information vector V2 of the target packaging box;

[0048] Product classification module: used for classifying the products to be packed into a first product category C1, a second product category C2 and a third product category C3 according to the product information vector V1 of the products to be packed;

[0049] 3D packing simulation environment construction module: used to generate 3D geometric models for each product to be packed in the first product category C1, the second product category C2 and the third product category C3 according to the product information vector V1; generate a virtual box model that matches the internal dimensions of the target packaging box according to the box information vector V2; and construct a 3D packing simulation environment according to the 3D geometric models of the product to be packed and the virtual box model;

[0050] Packing plan generation module: used to obtain the optimal packing plan through the pre-built packing plan learning model in a three-dimensional packing simulation environment.

[0051] An electronic device comprises a memory, a central processing unit and a computer program stored in the memory and executable on the central processing unit, wherein the central processing unit implements the above-mentioned intelligent packaging line production management method when executing the computer program.

[0052] A computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the above-mentioned intelligent packaging line production management method when executed.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The present invention realizes automatic packing scheme generation through accurate product information extraction and classification, significantly improves packaging efficiency and reduces manual intervention, adopts 3D simulation environment and learning model to quickly obtain the optimal packing scheme and reduces trial and error time; refined product classification and segmentation strategy enhances product safety, especially provides better protection for highly fragile products, and introduces physical property constraints in the packing scheme to ensure stability and safety; the application of intelligent classification and optimization algorithm reasonably arranges product position, maximizes the use of packaging box space, reduces packaging materials and logistics costs, and thus improves packaging quality; accurate 3D modeling and simulation ensure that each product remains in the best condition during the packaging process and reduces the breakage rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 It is a principle flow chart of the intelligent packaging line production management method in the present invention;

[0057] Figure 2 A flow chart of a method for obtaining product information vectors of products to be boxed in the intelligent packaging line production management method of the present invention;

[0058] Figure 3 A flow chart of a method for identifying the shape category of a ceramic package substrate in the intelligent packaging line production management method of the present invention;

[0059] Figure 4 It is a flow chart of a method for classifying products to be boxed into a first product category C1, a second product category C2 and a third product category C3 in the intelligent packaging line production management method of the present invention;

[0060] Figure 5 A flowchart of a method for further dividing each preliminary product category to obtain n2 subdivided product categories in the intelligent packaging line production management method of the present invention;

[0061] Figure 6 This is a functional module diagram of the intelligent packaging line production management system in the present invention;

[0062] Figure 7 A schematic diagram of the structure of an electronic device provided in one embodiment of the present application;

[0063] Figure 8 A schematic diagram of the computer-readable storage medium structure provided for one embodiment of the present application. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] Example 1

[0066] See also Figure 1 As shown, this embodiment provides a smart packaging line production management method, including:

[0067] Step S1000, obtaining a product information vector of a product to be boxed and a box body information vector of a target box;

[0068] Furthermore, step S1000 includes:

[0069] Step S1100, obtaining product information vectors of products to be packed.

[0070] Furthermore, if Figure 2 As shown, step S1100 includes:

[0071] Step S1110, obtaining first geometric parameters of the products to be packed, wherein the first geometric parameters include the length L1, width W1 and height H1 of each product to be packed;

[0072] Step S1120, obtaining first physical property parameters of the products to be packed, wherein the first physical property parameters include the weight M1, material and fragility F1 of each product to be packed;

[0073] Step S1130 , parameter association is performed on the acquired first geometric parameter and first physical property parameter of the product to be packed, so as to generate a product information vector V1 of the product to be packed.

[0074] Specifically, step S1100 is to obtain key parameters such as size and weight of the product to be packed by an automated stereoscopic vision scanning device. The device integrates hardware such as structured light cameras and weight sensors, and can quickly perform three-dimensional modeling and weight detection on the product. The point cloud data is processed by a visual algorithm to extract the length L1, width W1 and height H1 of the product. For products with complex shapes or internal structures, a complete three-dimensional model can be obtained by point cloud stitching, surface reconstruction and other technologies, and then its geometric parameters can be calculated. The weight M1 is directly measured by a high-precision weight sensor. The material and fragility F1 of the product can be obtained by manual input or image recognition. For example, for glass products, the material is "glass" and the fragility can be defined as "high"; for plastic products, the material is "plastic" and the fragility can be defined as "low". After obtaining the above parameters, they are organized according to a predefined data structure to generate a product information vector V1 that represents the product characteristics. V1 contains the geometric and physical properties of a single product to be packed, which is the basis for subsequent product classification and packing optimization.

[0075] Step S1100 efficiently obtains the key parameters of the product to be packed in an automated and intelligent manner, replacing the inefficiency and errors of manual measurement. The use of machine vision to quickly collect three-dimensional data and perform intelligent processing greatly improves the speed and accuracy of data collection. Through the organic integration of geometric parameters and physical properties, the characteristics of the product to be packed are fully characterized, providing data support for subsequent intelligent operations. The product information vector V1 generated by parameter association realizes the standardized representation of heterogeneous product data, which facilitates the unified processing of the packing algorithm. The vectorized product information also creates conditions for the subsequent introduction of machine learning algorithms.

[0076] Step S1200, obtaining the box information vector of the target packaging box.

[0077] Furthermore, if Figure 3 As shown, step S1200 includes:

[0078] Step S1210, obtaining second geometric parameters of the target packaging box, wherein the second geometric parameters include an inner length L2, an inner width W2, and an inner height H2 of the target packaging box;

[0079] Step S1220, obtaining a second physical property parameter of the target packaging box, wherein the second physical property parameter includes a maximum bearing weight M2 of the target packaging box. max and compressive strength P2;

[0080] Step S1230: perform parameter association on the acquired second geometric parameter and second physical property parameter of the target packaging box to generate a box body information vector V2 of the target packaging box.

[0081] Specifically, step S1200 is to input the parameter information of the target packaging box through the human-computer interaction interface. Since the packaging box is usually a standard part, its size and physical properties can be directly obtained from the product manual or order information. The operator selects the model of the target packaging box in the interface, and automatically associates the corresponding length L2, width W2, height H2, and maximum load M2. max and compressive strength P2. For non-standard packaging boxes, a 3D model can be constructed using CAD modeling tools, and relevant physical parameters can be manually input. The obtained geometric and physical property parameters are associated to generate a box information vector V2 that characterizes the characteristics of the target packaging box. Similar to the product information vector V1, V2 uses a standardized data structure to encapsulate the packaging box information, which is convenient for subsequent modeling of packing constraints and packing optimization solutions.

[0082] Step S1200 realizes the rapid entry and vectorized representation of the target packaging box information. Through human-computer interaction and parameter association, the scattered box data is integrated to improve the efficiency of information management. The standardized box information vector V2, together with the product information vector V1, constitutes the input condition of the intelligent packing problem. Based on the vectorized description, the mathematical modeling of the box constraint conditions can be easily carried out, and the necessary boundary conditions can be provided for the generation of the packing scheme. At the same time, the parameterized box description also provides a basis for the rapid evaluation and screening of the packing scheme.

[0083] Step S2000: Classify the products to be packed into a first product category C1, a second product category C2 and a third product category C3 according to the product information vector V1 of the products to be packed.

[0084] Furthermore, if Figure 4 As shown, step S2000 includes:

[0085] Step S2100, extracting features from the product information vector V1 to obtain the volume S1, weight M1 and fragility F1 of each product to be packed;

[0086] Step S2200, preliminarily classifying the products to be packed according to their volumes S1 and weights M1, obtaining n1 preliminary product categories, where n1 is an integer greater than 0;

[0087] Step S2300, further dividing each preliminary product category according to the fragility F1 of the product to be packed, to obtain n2 subdivided product categories, n2≥n1;

[0088] Furthermore, if Figure 5 As shown, step S2300 includes:

[0089] Step S2310: for each preliminary product category, calculate the fragility mean F1 according to the fragility F1 of the products therein mean and vulnerability variance F1 var ;

[0090] Step S2320, if the fragility variance F1 var Greater than the preset variance threshold var thr , then the preliminary product category is subdivided into k subcategories, k ≥ n2; if the vulnerability variance F1 var Less than or equal to the preset variance threshold var thr , it is considered that the products within the preliminary product category have little difference in vulnerability and no further subdivision is required;

[0091] Step S2330, merging all subcategories obtained from the preliminary product category segmentation to form n2 segmented product categories.

[0092] Further, step S2330 includes:

[0093] Step S2331, treating each subcategory as an independent cluster, and calculating the distance between any two clusters, where the distance between any two clusters is the absolute value of the difference between the mean values ​​of the vulnerability of the two clusters;

[0094] Step S2332: Find the two clusters C that are closest to each other among all the current clusters. p and C q , merged into a new cluster C r ;

[0095] Step S2333, update the merged cluster C r The distance to other existing clusters;

[0096] Step S2334, repeat steps S21332 and S21333 until the number of clusters is reduced to n2.

[0097] Specifically, step S2330 uses a hierarchical clustering method to merge all subcategories obtained by the preliminary product category segmentation, and finally forms n2 subcategories. Hierarchical clustering is a bottom-up clustering method. By continuously merging the two clusters with the closest distance, the required number of categories is finally obtained. The value of n2 can be predetermined according to actual needs. For example, n2=5 means that 5 subcategories are finally obtained; each subcategories contains a group of products with similar fragility, and the homogeneity within the category is enhanced. Subsequent packaging scheme design can formulate differentiated packaging strategies according to the characteristics of different subcategories of product categories, which helps to improve packaging efficiency and product safety. Step S2330 makes full use of the product fragility information, and adaptively generates the required number of subcategories of product categories through the division and merging of subcategories. This method effectively improves the consistency within the category, so that the products in each subcategories are highly similar in terms of fragility. This product segmentation method based on fragility provides important prior information for subsequent packaging scheme optimization. By formulating personalized packaging strategies for different subcategories, product safety can be maximized and packaging efficiency can be improved. In addition, the subdivision of product categories is also conducive to the rational allocation of packaging resources, avoiding the problem of excessive or insufficient packaging.

[0098] Step S2300 adopts a two-stage classification method, which is to subdivide products according to their vulnerability level based on the preliminary classification. For each preliminary product category, the mean value F1 of its vulnerability level is calculated. mean and variance F1 var Mean F1 mean Reflects the overall vulnerability level of products in this category, variance F1 var It represents the difference in the fragility of products within the category. var It is relatively large, indicating that the vulnerability of products within this category varies and needs to be further subdivided.

[0099] In order to determine whether segmentation is needed, the vulnerability variance threshold var is introduced thr When F1 var Greater than var thrWhen the initial product category is considered to have large differences within it, a clustering algorithm based on the Gaussian mixture model (GMM) is used for segmentation. GMM is an unsupervised learning algorithm that can adaptively divide data into multiple sub-clusters, each of which obeys a Gaussian distribution. Unlike the K-means algorithm, which requires the number of clusters to be specified in advance, GMM can automatically adjust the number of clusters according to the data distribution. Here, the BIC criterion is used to determine the optimal number of sub-clusters k to achieve automatic segmentation of categories. After segmentation, the original preliminary product category is divided into k sub-categories, and the fragility of the products within each sub-category is closer. Then all the sub-categories obtained by the segmentation of the preliminary product category are merged to finally form n2 subdivided product categories. These subdivided product categories are more homogeneous in terms of fragility, providing a more refined grouping basis for the subsequent generation of packing solutions. Through two-stage classification, the combined effects of product size, weight and fragility are fully considered, which not only ensures the computational efficiency of the preliminary classification, but also takes into account the differences in fragility.

[0100] Subdividing product categories can be used for refined packing optimization. For example, for products with high fragility, they can be placed in the center of the box first and adopt shock-absorbing and buffering measures; while products with low fragility can be placed on the bottom or periphery. The reasonable division of product categories helps to formulate differentiated packing strategies and improve packing efficiency and cargo safety. In addition, product classification information can also be used in smart picking, inventory management and other links to achieve refined control of the entire product life cycle.

[0101] Step S2400: divide n2 subdivided product categories into a first product category C1, a second product category C2, and a third product category C3.

[0102] The method for dividing the n2 subdivided product categories into the first product category C1, the second product category C2 and the third product category C3 is:

[0103] For each product category, extract its vulnerability mean F1 mean and size mean S1 mean Features, set the vulnerability threshold F1 θ , size threshold S1 θ ;

[0104] If F1 mean ≥F1 θ And S1 mean ≥S1 θ , then the subdivided product category is divided into the first product category C1;

[0105] If F1 mean <F1 θ And S1 mean <S1θ , then the subdivided product category is divided into the second product category C2;

[0106] If F1 mean ≥F1 θ And S1 mean <S1 θ , or F1 mean <F1 θ And S1 mean ≥S1 θ , then the subdivided product category is divided into the third product category C3.

[0107] Specifically, in step S2300, n2 product categories are obtained through two-stage classification. These categories take into account the product's size, weight, and fragility, and achieve refined grouping of products. In order to further summarize the product categories and facilitate the subsequent formulation of differentiated packaging strategies, step S2400 divides the n2 product categories into three categories: the first product category C1, the second product category C2, and the third product category C3. The basis for division is the mean fragility value F1 of each product category. mean and size mean S1 mean . Mean fragility F1 mean Indicates the average level of fragility of all products in this category. The larger the value, the more fragile the product is. mean It reflects the average volume size of products in this category. The larger the value, the larger the product size.

[0108] Threshold F1 θ and S1 θ The selection needs to be combined with practical experience and business needs. Usually, historical packing data can be analyzed to statistically analyze the relationship between product breakage rate, fragility and size, and then determine a reasonable threshold. The setting of the threshold will affect the final product category division result, and it needs to be repeatedly tuned to achieve the best balance. Through step S2400, the n2 subdivided product categories are further summarized into three major categories. Each major category contains subdivided product categories with similar characteristics, which makes the category characteristics more obvious and is conducive to the formulation of subsequent packaging strategies. This division fully considers the differences in product characteristics and is conducive to the formulation of packaging strategies according to categories. Through threshold selection and optimization, the product category characteristics can be made more significant, which is conducive to improving packaging efficiency.

[0109] Step S2000 uses the clustering algorithm of unsupervised machine learning to realize the intelligent classification of the products to be packed. The results of product classification can be directly applied to the subsequent selection of packing containers, generation of packing schemes and other links. For example, for C1 products with high fragility and high size, large turnover boxes can be selected, equipped with cushioning packaging such as air bags and shock pads, and packed separately as much as possible; for C2 products with low fragility and low size, small cartons can be selected, and combined packing can be used to improve space utilization; for C3 products, targeted packaging measures can be taken according to their fragility and size, and packaging containers and cushioning materials can be flexibly selected. Step S2000 makes full use of the geometric and physical attribute information of the products, and realizes the intelligent grouping of products through feature extraction and classification algorithms. Clustering models such as K-means and GMM can adaptively divide cluster boundaries and continuously iterate and optimize to discover the intrinsic distribution pattern of data. Through hierarchical segmentation, while meeting business needs, it takes into account computational efficiency, which is not only conducive to the efficient implementation of subsequent packaging optimization, but also provides important prior knowledge for the generation of intelligent packing schemes. In addition, product classification can also be used in other links of the intelligent logistics system, such as sorting, transportation, warehousing, etc. Through category induction and feature analysis, it can provide data support and decision-making basis for intelligent upgrades.

[0110] For example, suppose there is a robot manufacturer that produces various robot joints, actuators and other hardware components. These components need to be classified before packing so that targeted packaging measures can be taken. The method described in step S2000 can be applied to intelligently classify robot hardware products.

[0111] First, for each robot component to be packed, obtain its key attributes such as size, weight and fragility through the product information vector V1. The product information vector can come from parameter collection in the production process, or it can be obtained in real time through 3D scanning and other methods. Assume that the following product parameters are obtained:

[0112] Joint A: size (100x80x60) mm, weight 1.2 kg, fragility 0.6;

[0113] Actuator B: size (200x120x90) mm, weight 3.5 kg, fragility 0.8;

[0114] Sensor C: size (50x40x30) mm, weight 0.5 kg, fragility level 0.9;

[0115] According to step S2100, feature extraction is performed on the product information vector V1 to obtain the volume S1, weight M1 and fragility F1 of each product. Next, according to step S2200, preliminary classification is performed based on the volume and weight of the product. The volume and weight thresholds can be set, for example, the volume threshold is 1x10 5mm 3 , the weight threshold is 2kg. Then the above three products can be divided into two categories:

[0116] Preliminary Category I: Joint A (volume 48x10 4 mm 3 , weight 1.2kg);

[0117] Preliminary Category II: Actuator B (volume 216x10 4 mm 3 , weight 3.5kg), sensor C (volume 6x10 4 mm 3 , weight 0.5kg);

[0118] Then, according to step S2300, each preliminary product category is further divided according to the vulnerability level, and the mean and variance of the vulnerability levels of preliminary categories I and II are calculated:

[0119] Preliminary Category I: F1 mean =0.6, F1 var =0;

[0120] Preliminary Category II: F1 mean =0.85, F1 var =0.005;

[0121] Set the vulnerability variance threshold var thr =0.01, it can be seen that the variance of preliminary category I is 0, which is less than the threshold and does not need to be subdivided; while the variance of preliminary category II is 0.005, which is also less than the threshold and does not need to be subdivided. Therefore, the original two preliminary categories do not need to be further subdivided, n2=n1=2. Two subdivided product categories are obtained.

[0122] Finally, according to step S2400, the two subdivided product categories are divided into the first, second, and third product categories. Assume that the vulnerability threshold F1 is set θ =0.7, size threshold S1 θ =1x10 5 mm 3 For subdivision category I, the mean vulnerability value F1 mean =0.6<F1 θ , size mean S1 mean <S1 θ , so it is classified as the second product category C2. For sub-category II, the mean vulnerability level F1 mean =0.85≥F1 θ , but the size mean S1 mean For sensor C less than S1 θ , for actuator B is greater than S1 θ, and is therefore classified as the third product category C3.

[0123] In summary, by applying step S2000, robot hardware products are successfully divided into two categories:

[0124] Second product category C2: joint A, low fragility and low size products;

[0125] Third product category C3: Actuators B and sensors C, products with high fragility or high dimensions.

[0126] Subsequent packaging solutions can be designed for these two categories. For C2-type low-fragility and low-size robot joints, small packaging boxes can be used and filled with conventional shockproof materials; while for C3-type high-fragility or high-size actuators and sensors, large-size packaging boxes need to be used and filled with a large amount of shockproof fillers to avoid collision damage. In addition, highly fragile sensors can also be individually placed in shockproof bags to provide additional protection.

[0127] From the above case analysis, it can be seen that the classification method of step S2000 can be well applied to the intelligent packaging of robot hardware products. This method fully considers the differences in geometric and physical properties of products and realizes refined product classification through hierarchical classification. Customized packaging solutions can not only improve packaging efficiency, but also maximize the safe transportation of hardware. In addition, product classification information can also be used in production scheduling optimization, inventory management and other links to achieve intelligent full-process control.

[0128] Step S3000: Generate a three-dimensional geometric model for each product to be packed in the first product category C1, the second product category C2 and the third product category C3 according to the product information vector V1; generate a virtual box model that matches the internal dimensions of the target packaging box according to the box information vector V2; and construct a three-dimensional packing simulation environment based on the three-dimensional geometric model of the product to be packed and the virtual box model.

[0129] Specifically, step S3000 uses computer graphics and virtual reality technology to build a realistic three-dimensional packing simulation environment. First, according to the product geometric parameters in the product information vector V1, a three-dimensional geometric model of each product to be packed is generated by a parametric modeling method. For example, a WebGL framework such as THREE.js can be used to dynamically generate basic geometric elements such as cubes and cylinders, and a three-dimensional model similar to the actual product appearance can be constructed through transformation operations such as translation, rotation, and scaling. For products with irregular shapes, they can also be approximated by methods such as triangular meshes. Then, according to the target packaging box size parameters in the box information vector V2, a virtual box model is generated. The box model is also dynamically generated in a parametric manner to adapt to packaging boxes of different sizes. The internal dimensions of the box model correspond to L2, W2, and H2 in V2, which are used to constrain the filling of the product model. Finally, the generated product geometric model and box model are imported into a three-dimensional physical simulation engine to build a packing simulation environment. By introducing mechanisms such as collision detection and contact constraints in the physical engine, the physical interaction between products and between products and packaging boxes is simulated to improve the authenticity of the packing simulation.

[0130] The core of step S3000 is to build a virtual packing environment in the computer for the generation and optimization of packing solutions. Through three-dimensional geometric modeling and parametric design, this step realizes the digital description of product shape and packaging box size, so that the packing problem can be solved in virtual space. The three-dimensional visualization of the product model not only intuitively displays the geometric characteristics of the product to be packed, but also provides a vivid reference for the subsequent spatial layout optimization. The introduction of a professional three-dimensional physics engine can simulate the packing process of the product in a virtual environment. Through the force analysis and collision detection of the product, the physics engine can calculate the interaction force between the products and predict the stability problems that may occur during the packing process. This physical simulation technology greatly improves the accuracy and reliability of the packing solution evaluation and avoids unforeseen problems in the actual application of the solution. In addition, the virtual packing environment also provides the necessary data interface and operation platform for the implementation of the packing optimization algorithm. Various heuristic rules, operations research models and artificial intelligence algorithms can be customized and quickly verified in this environment. The packing solution generated by the algorithm can be rendered into a three-dimensional scene in real time to intuitively display the packing effect. The comparison and screening between multiple solutions have also become more convenient and efficient.

[0131] In general, the construction of the 3D packing simulation environment realizes the digitization, visualization and intelligence of the packing problem solving process. Professional modeling tools and physical engines ensure the accuracy and feasibility of the design. Convenient algorithm development and testing platforms promote the continuous innovation and improvement of packing optimization technology. This virtual-real packing optimization method represents the latest development direction in this field and has broad application prospects.

[0132] Step S4000: In a three-dimensional packing simulation environment, an optimal packing solution is obtained through a pre-built packing solution learning model.

[0133] Furthermore, step S4000 includes:

[0134] Step S4100, in a three-dimensional packing simulation environment, obtaining a first packing solution through a pre-built packing solution learning model;

[0135] Specifically, the packing solution learning model adopts the classic DQN (Deep Q-Network) architecture in the deep reinforcement learning algorithm, which includes a policy network and a value network. The policy network is used to select the packing action according to the environment state, and the value network is used to evaluate the long-term return of the state-action pair. Both networks are implemented using deep convolutional neural networks, which can effectively extract the spatial features of the packing state. Initialize the policy network and value network of the deep reinforcement learning algorithm, and set hyperparameters such as the exploration rate ε, learning rate α, and discount factor γ; the selection of hyperparameters will directly affect the convergence speed and solution quality of the algorithm, and is usually set based on experience and adjusted appropriately during the training process. The exploration rate ε controls the probability of the agent selecting a random action, the learning rate α determines the step size of the network parameter update, and the discount factor γ reflects the importance of future rewards.

[0136] In each simulation step, the agent in DQN randomly selects a packing action with a probability of ε according to the ε-greedy strategy, or selects the optimal action given by the current policy network with a probability of 1-ε. The packing action includes selecting a product to be packed and determining its placement and direction in the box. After the action is executed, the environment gives an immediate reward based on the packing result, indicating the quality of the action. At the same time, the state of the environment also shifts to the next time step. The agent will (s t ,a t ,r (t+1) ,s (t+1)) as a transfer sample and stored in the experience replay pool for subsequent network training. The experience replay pool is a fixed-size queue used to cache transfer samples generated by the interaction between the agent and the environment. In each training step, a batch of samples are randomly sampled from the replay pool, and the network parameters are updated through time difference learning. The time difference error is defined as the difference between the target Q value and the predicted Q value, where the target Q value is calculated according to the Bellman optimal equation and the predicted Q value is output by the value network. By minimizing the time difference error, the predictive ability of the value network is continuously improved, and the long-term return of each state-action pair can be more accurately evaluated. The policy network is trained using the policy gradient algorithm, which optimizes the action selection strategy in the direction of higher returns based on the evaluation results of the value network.

[0137] As the packing simulation progresses, the agent gradually learns a better packing strategy and can obtain a first packing plan with high packing efficiency while satisfying various constraints. This plan includes the placement and order of each product, providing a good initial solution for subsequent packing plan optimization. The packing plan learning model based on the deep reinforcement learning algorithm selects a near-optimal solution from countless possible packing plans through massive simulation attempts. Its generalization ability and solution efficiency far exceed those of traditional heuristic algorithms. At the same time, the algorithm is adaptive and can automatically adjust the strategy according to different products and packaging boxes without the need for manual design of complex rules.

[0138] For example, for robot parts to be packed, their CAD models can be loaded into a three-dimensional simulation environment. By analyzing the CAD model, the geometric parameters (such as length, width, height) and physical property parameters (such as weight, material, fragility, etc.) of each part can be extracted to construct realistic virtual parts. The internal dimensions of the target packaging box are set according to the box information vector V2 obtained in step S1000. The packing scheme learning model is used to train the component packing process, and a better packing strategy is found through tens of thousands of attempts. For example, for large and fragile parts of type C1, the strategy tends to be placed in the center of the bottom layer of the box and try to avoid contact with other parts; for small and sturdy parts of type C2, the strategy tends to use the remaining space in the box and place them in a more stable interlayer. Through visualization technology, the packing process and results can be intuitively observed, problems can be found in time, and strategies can be adjusted. After repeated training, a balanced solution that can take into account packing stability, component safety, and packing efficiency is found. This solution can provide an important reference for subsequent packing optimization.

[0139] Step S4200, obtaining a first constraint condition of the packing process, introducing the first constraint condition into the packing solution learning model, and obtaining a second packing solution that satisfies the first constraint condition; the first constraint condition includes a weight balance threshold ω b, Placement gap threshold d g 、The upper limit of stacking layers N s and the total constraint cost C total ; Total constraint cost C total Based on the weight balance threshold ω b , Placement gap threshold d g And the upper limit of stacking layers N s calculated;

[0140] Specifically, step S4200 further introduces constraints of the packing process based on the first packing solution obtained in step S4100, and optimizes the packing solution through the reward mechanism. The first constraint mainly includes four indicators: weight balance threshold ω b , Placement gap threshold d g 、The upper limit of stacking layers N s and the total constraint cost C total Among them, the weight balance threshold ω b Indicates the allowable range of the center of gravity offset in the packaging box after packing, which is usually set to within 10% of the diagonal length of the bottom surface of the packaging box. Exceeding this range may cause the packed goods to overturn, affecting transportation safety. Placement gap threshold d g Indicates the maximum allowable gap size between adjacent products and between products and the box wall, usually set to within 10% of the minimum product size. Too large a gap will reduce the stability of the box, while too small a gap may cause extrusion damage. The upper limit of the number of stacking layers is N. s It indicates the maximum number of layers that products can be stacked in a packaging box. It is usually determined based on the pressure-bearing capacity of the product and the strength of the packaging box. Exceeding the upper limit of the number of layers may cause the box to deform or the product to be damaged. The total constraint cost is used to measure the constraint satisfaction of the packing solution. The total constraint cost C total Comprehensive consideration is given to weight balance, reasonable placement spacing and safety of stacking layers.

[0141] Total constraint cost C total The calculation formula is:

[0142]

[0143] in:

[0144] ω b : Weight balance threshold, reflecting the impact of center of gravity offset on the total constraint cost;

[0145] Δx, Δy, Δz: the components of the center of gravity offset of the packing on the three axes;

[0146] L: length of the packaging box;

[0147] W: width of the packaging box;

[0148] H: height of the packaging box;

[0149] d g : Placement gap threshold, reflecting the impact of gap on total constraint cost;

[0150] g i : The actual gap of the i-th product;

[0151] s i : The minimum size of the i-th product;

[0152] n: total number of products;

[0153] N s : The upper limit of the number of stacking layers, reflecting the impact of the number of stacking layers on the total constraint cost;

[0154] h max : The maximum height of the current stack;

[0155] H box : The internal height of the packaging box;

[0156] ω b , d g , N s It is used to measure the satisfaction degree of the corresponding constraint conditions of the packing scheme. In the calculation formula of the total constraint cost, the first item measures the ratio of the center of gravity offset to the size of the packaging box. The larger the offset, the higher the degree of constraint violation. The second item measures the sum of the ratios of the gaps between all products and their minimum size. The larger the gap, the higher the degree of constraint violation. The third item measures the ratio of the current stacking height to the internal height of the packaging box. The higher the stack, the higher the degree of constraint violation. b , d g , N s The value of is determined according to the actual packing requirements and is usually set to the critical value of the corresponding constraint condition.

[0157] The total constraint cost C total Introducing the reward function of the packing solution learning model can more comprehensively evaluate the pros and cons of the packing solution. The reward function R(s,a) represents the immediate reward for selecting action a in state s, which consists of two parts: packing efficiency reward and constraint penalty. total The introduction of reward function can more comprehensively evaluate the pros and cons of the packing solution. total When C is small, it indicates that the packing scheme performs well in terms of weight balance, compact placement, and stacking safety, and a higher positive reward should be given; on the contrary, when C total When it is large, it is necessary to punish unreasonable packing actions through negative rewards. During the simulation training process, the intelligent experience will weigh the packing efficiency and the total constraint cost, constantly adjust the strategy, and finally obtain a balanced packing solution that performs well in all constraint indicators.

[0158] Step S4200 introduces a weight balance threshold ω b , Placement gap threshold d g 、The upper limit of stacking layers N s and the total constraint cost C total Mechanisms such as weight balance threshold ω b By controlling the center of gravity offset, the risk of overturning is reduced and the stability of the goods during transportation is ensured; the gap threshold d g By limiting the gaps between products and between them and the box wall, space utilization is optimized and the stability of the box is improved to avoid extrusion damage; the upper limit of the number of stacking layers is N s According to the pressure bearing capacity of the product and the strength of the packaging box, the number of stacking layers is limited to prevent the box from deforming and the product from being damaged, thereby enhancing the safety of stacking. By comprehensively considering the center of gravity offset, placement gap and number of stacking layers, the total constraint cost C total It provides a comprehensive evaluation basis for the packing solution. Introducing the total constraint cost into the reward function helps the agent optimize the packing efficiency while satisfying the constraints and find the best packing solution. This series of optimization mechanisms ensures the best state of the packing solution in terms of safety, stability and efficiency, adapts to different products and packaging requirements, and improves the flexibility and practicality of the solution.

[0159] Step S4300, according to the second packing plan, calculate the comprehensive physical property parameters after packing, compare the comprehensive physical property parameters after packing with the second physical property parameters of the target packaging box, and determine whether the second packing plan meets the packing safety requirements; if not, take the packing safety requirements as the second constraint condition, return to step S4200, introduce the second constraint condition into the packing plan learning model, and continue to optimize until a third packing plan that meets all constraints is obtained; take the third packing plan as the optimal packing plan.

[0160] The comprehensive physical property parameters include the total weight M of the packed goods c and the compressive stress P of the bottom layer of cargo b ;

[0161] The packing safety requirements are: c ≤M2 max And P b ≤P2;

[0162] Specifically, step S4300 further considers the safety of the packed goods on the basis of satisfying the first constraint condition. In step S1220, the second physical attribute parameters of the target packaging box have been obtained, including the maximum load-bearing weight M2 maxand compressive strength P2, so these parameters can be directly used to evaluate the safety of the packing scheme. First, according to the second packing scheme, calculate the total weight M of the packed goods c And the actual load-bearing capacity of each layer stacked M s . c Maximum load-bearing capacity of the target packaging box M2 max Compare and judge whether the total weight of the packing exceeds the limit. Then, calculate the compressive stress P of the bottom layer of cargo b , and compare it with the compressive strength P2 of the target packaging box to determine whether the bottom layer of goods is at risk of being crushed. If the total weight of the packed goods exceeds the maximum load-bearing weight, or the compressive stress of the bottom layer of goods exceeds the compressive strength, it is considered that the second packing scheme does not meet the packing safety requirements and needs to be further optimized.

[0163] Introducing the packing safety requirement as the second constraint into the packing solution learning model can significantly improve the degree of product protection and reduce the damage rate during transportation by limiting the stacking height and other measures. The safety constraint optimizes the packing layout, such as placing heavy objects at the bottom and fragile items in the center, which improves the overall stability. This constraint also realizes differentiated packaging strategies and formulates suitable packaging solutions for different product categories. The model seeks a balance between efficiency and safety, ensuring safety without excessive waste of space, which helps to optimize packaging costs. The introduction of safety constraints improves the practicality and executability of the model, reduces the need for manual adjustments, and provides a more comprehensive decision-making basis for the intelligent packing system. In addition, safer packing solutions reduce logistics risks, reduce compensation expenses and additional processing, and can also enhance corporate brand image and customer loyalty. Safety constraints also support compliance to ensure that the packing solution complies with relevant regulations and standards. Finally, by analyzing the impact of safety constraints on packing results, companies can continuously optimize their packaging strategies and material selection to achieve long-term efficiency improvements and cost reductions. In general, the introduction of this constraint makes the packing solution more comprehensive, practical and safe, providing companies with new ideas for optimizing packaging strategies, bringing positive impacts in efficiency, cost and customer satisfaction.

[0164] Step S4300 improves the safety, reliability and adaptability of the packing scheme through comprehensive evaluation and dynamic optimization. This step establishes a feedback loop mechanism to achieve continuous improvement of the packing scheme, maximize space utilization while ensuring the safety of the goods, thereby improving packing efficiency and reducing the loss rate. By considering a variety of physical property parameters, the generated packing scheme can adapt to complex actual situations and enhance the robustness of the scheme. In addition, the data and experience accumulated in this process provide valuable decision-making support for intelligent manufacturing and promote the development of packing technology to a higher degree of intelligence and automation. In general, step S4300 achieves comprehensive optimization of the packing scheme by incorporating safety requirements into the optimization process, which not only makes significant progress in efficiency and safety, but also makes important contributions to the development of the entire intelligent manufacturing field.

[0165] Step S4000 seeks the optimal packing scheme by establishing a packing scheme learning model, comprehensively considering multiple objectives such as packing efficiency and product safety, and constructing the objective function by weighted summation, so that the packing scheme can take into account the integrity rate of goods while improving space utilization, and achieve multi-objective balanced optimization. The comprehensiveness of the constraints ensures the feasibility and practicality of the packing scheme, prevents the packed goods from exceeding the carrying capacity of the packaging box, and avoids unreasonable stacking that causes damage to the goods. Furthermore, the packing optimization model can be solved quickly, providing efficient decision support for the packing tasks of multi-batch and multi-specification products, reducing the difficulty and error rate of manual planning. In addition, the model solution process will dynamically adjust the decision variables, so it has a certain robustness and can adapt to the fluctuations of product parameters and packing requirements. Through mathematical modeling and intelligent optimization, step S4000 plays a positive role in improving packing efficiency and product safety, and provides new ideas and methods for the intelligent packaging industry.

[0166] Step S5000: Based on the optimal packing solution, the packing operation is executed and dynamically optimized.

[0167] Furthermore, step S5000 includes:

[0168] Step S5100, performing actual packing operations according to the optimal packing solution, collecting container process data in real time, comparing the packing process data with the virtual packing results of the optimal packing solution, and calculating the packing error;

[0169] Step S5200: If the packing error exceeds a preset packing error threshold, an alarm mechanism is triggered, and the packing error data is fed back to the packing solution learning model of step S4000 to continuously optimize the packing solution learning model.

[0170] Specifically, step S5000 is to apply the optimal packing solution generated in step S4000 to the actual automated packing operation. Packing equipment is usually composed of industrial robots, robotic arms, material transmission lines, etc., which can realize a series of actions such as product positioning, grasping, and loading. As the motion control instructions of the robot, the packing solution needs to be converted into executable code, such as G code. The robot control system parses the packing code, plans the motion trajectory and operation sequence of the robotic arm, and accurately controls the robotic arm to complete the packing action. The selection of packing equipment needs to comprehensively consider factors such as the size, weight, and material of the packing object, and select a robot body and end effector with matching performance. The robot's payload, degree of freedom, working range and other parameters must meet the requirements of the packing operation. In addition, the packing equipment must also be equipped with reliable safety protection devices, such as safety light curtains, emergency stop switches, etc., to ensure the safety of man and machine during the operation process.

[0171] By installing visual sensors such as RGB-D cameras on the packing equipment, real-time point cloud data of the container parts can be collected to reconstruct the three-dimensional scene of the loading status. The obtained three-dimensional loading results are compared with the theoretical packing plan, and the deviation between the two is calculated to obtain the packing error. The packing error can be measured by indicators such as Euclidean distance and Hausdorff distance, which reflects the difference between the actual loading posture and the target posture. The introduction of packing error evaluation can quantify the accuracy of packing execution and provide a basis for the adaptive optimization of the packing process. When the packing error is small, it means that the packing action is in good agreement with the theoretical plan and the packing accuracy is high; when the packing error exceeds the preset threshold, it means that the actual packing effect is significantly different from the expected one and needs to be adjusted in time.

[0172] Step S5100 realizes the transformation of the packing scheme into the actual packing action, which is a key link for the implementation of the intelligent packaging system. Through reliable robot hardware and precise motion control, products can be loaded into the packaging box efficiently and orderly, saving labor to the maximum extent and improving packaging efficiency. Based on the guidance of the three-dimensional packing scheme, the robot can flexibly respond to the packaging requirements of different batches and specifications, and has good versatility and adaptability. Through visual sensing technology and three-dimensional reconstruction algorithm, real-time monitoring and quantitative evaluation of the packing process are realized. Compared with offline detection, online visual detection can detect packing deviations earlier and provide timely feedback to the control system for adjustment, thereby improving packing accuracy and stability. Through the collection and analysis of point cloud data, complete three-dimensional information of the loading state can be obtained, including the posture and gap of the product, providing richer data support for packing optimization. By comparing the measured loading results with the theoretical packing scheme, the deviation of packing execution can be intuitively quantified, providing quantitative indicators for adaptive optimization. The setting of the packing error threshold needs to balance the packing efficiency and packing accuracy. Too high a threshold may cause frequent downtime, while too low a threshold makes it difficult to activate the optimization mechanism. Reasonable threshold selection needs to be combined with factors such as product type and packaging requirements, and obtained through experience summary and repeated debugging.

[0173] Step S5200 is the key to the adaptive optimization of the packing process. Through packing error feedback and algorithm fine-tuning, the continuous improvement of the packing solution learning model is achieved. When visual inspection finds that the packing error exceeds the threshold, the system will automatically alarm and prompt manual intervention. At the same time, the packing error data is fed back to the deep reinforcement learning algorithm of step S4000 and included in the model training as a new training sample. By continuously enriching the sample space, learning error cases, and optimizing the packing strategy, the algorithm can adapt to more abnormal situations. On the basis of sample training, it is also necessary to fine-tune the algorithm model, such as correcting the incentive function, adjusting the network structure, updating the loss weight, etc., to improve the model fitting ability. Through enhanced training and parameter tuning, the packing model can continue to evolve and generate more accurate and robust packing solutions. Step S5200 builds a closed-loop feedback mechanism to achieve self-iteration and optimization of the packing model. The packing error is obtained through visual inspection, and the deviation is fed back in time to trigger the retraining and fine-tuning of the model. Continuously accumulate packing experience data, enrich the sample library, and enable the model to learn more packing patterns and abnormal situations. At the same time, by adjusting the algorithm parameters and optimizing the network structure, the generalization ability and robustness of the model are improved. Closed-loop optimization enables the packing system to have the ability to self-improve and continuously evolve, constantly adapt to new packing requirements, and improve the packing level.

[0174] Packing error feedback and adaptive optimization are the key to achieving closed-loop control of the packing solution. Through online visual inspection, execution deviations are fed back in a timely manner to avoid error accumulation. Incorporating error data into training can expand the experience of the algorithm model and enable it to learn more packing skills. Through algorithm fine-tuning, the adaptability and robustness of the packing strategy can be improved. Long-term self-learning and optimization can enable the packing system to continuously evolve and better cope with various packing challenges.

[0175] Example 2

[0176] This embodiment provides an intelligent packaging line production management system based on the first embodiment. Figure 6 As shown, including:

[0177] Vector extraction module: used to obtain the product information vector V1 of the product to be packed and the box information vector V2 of the target packaging box;

[0178] Product classification module: used for classifying the products to be packed into a first product category C1, a second product category C2 and a third product category C3 according to the product information vector V1 of the products to be packed;

[0179] 3D packing simulation environment construction module: used to generate 3D geometric models for each product to be packed in the first product category C1, the second product category C2 and the third product category C3 according to the product information vector V1; generate a virtual box model that matches the internal dimensions of the target packaging box according to the box information vector V2; and construct a 3D packing simulation environment according to the 3D geometric models of the product to be packed and the virtual box model;

[0180] Packing plan generation module: used to obtain the optimal packing plan through the pre-built packing plan learning model in a three-dimensional packing simulation environment.

[0181] The vector extraction module comprises:

[0182] Product information vector extraction unit: used to obtain product information vectors of products to be packed;

[0183] Box information vector extraction unit: used to obtain the box information vector of the target packaging box.

[0184] The step of obtaining the product information vector of the product to be packed includes:

[0185] Step S1110, obtaining first geometric parameters of the products to be packed, wherein the first geometric parameters include the length L1, width W1 and height H1 of each product to be packed;

[0186] Step S1120, obtaining first physical property parameters of the products to be packed, wherein the first physical property parameters include the weight M1, material and fragility F1 of each product to be packed;

[0187] Step S1130 , parameter association is performed on the acquired first geometric parameter and first physical property parameter of the product to be packed, so as to generate a product information vector V1 of the product to be packed.

[0188] The step of obtaining the box body information vector of the target packaging box includes:

[0189] Step S1210, obtaining second geometric parameters of the target packaging box, wherein the second geometric parameters include an inner length L2, an inner width W2, and an inner height H2 of the target packaging box;

[0190] Step S1220, obtaining a second physical property parameter of the target packaging box, wherein the second physical property parameter includes a maximum bearing weight M2 of the target packaging box. max and compressive strength P2;

[0191] Step S1230: perform parameter association on the acquired second geometric parameter and second physical property parameter of the target packaging box to generate a box body information vector V2 of the target packaging box.

[0192] The product classification module includes:

[0193] Product feature extraction unit: used to extract features from the product information vector V1 to obtain the volume S1, weight M1 and fragility F1 of each product to be packed;

[0194] Preliminary classification unit: used to preliminarily classify the products to be packed according to their volume S1 and weight M1, and obtain n1 preliminary product categories, where n1 is an integer greater than 0;

[0195] Category subdivision unit: used to further divide each preliminary product category according to the fragility F1 of the product to be packed, and obtain n2 subdivided product categories, n2≥n1;

[0196] Product classification unit: used to divide n2 subdivided product categories into the first product category C1, the second product category C2 and the third product category C3.

[0197] The category subdivision unit is used to further divide each preliminary product category according to the fragility F1 of the product to be packed, including:

[0198] Step S2310: for each preliminary product category, calculate the fragility mean F1 according to the fragility F1 of the products therein mean and vulnerability variance F1 var ;

[0199] Step S2320, if the fragility variance F1 var Greater than the preset variance threshold varthr , then the preliminary product category is subdivided into k subcategories, k ≥ n2; if the vulnerability variance F1 var Less than or equal to the preset variance threshold var thr , it is considered that the products within the preliminary product category have little difference in vulnerability and no further subdivision is required;

[0200] Step S2330, merging all subcategories obtained from the preliminary product category segmentation to form n2 segmented product categories.

[0201] The subcategories obtained by combining all preliminary product category breakdowns include:

[0202] Step S2331, treating each subcategory as an independent cluster, and calculating the distance between any two clusters, where the distance between any two clusters is the absolute value of the difference between the mean values ​​of the vulnerability of the two clusters;

[0203] Step S2332: Find the two clusters C that are closest to each other among all the current clusters. p and C q , merged into a new cluster C r ;

[0204] Step S2333, update the merged cluster C r The distance to other existing clusters;

[0205] Step S2334, repeat steps S21332 and S21333 until the number of clusters is reduced to n2.

[0206] The method for dividing the n2 subdivided product categories into the first product category C1, the second product category C2 and the third product category C3 is:

[0207] For each product category, extract its vulnerability mean F1 mean and size mean S1 mean Features, set the vulnerability threshold F1 θ , size threshold S1 θ ;

[0208] If F1 mean ≥F1 θ And S1 mean ≥S1 θ , then the subdivided product category is divided into the first product category C1;

[0209] If F1 mean <F1 θ And S1 mean <S1 θ, then the subdivided product category is divided into the second product category C2;

[0210] If F1 mean ≥F1 θ And S1 mean <S1 θ , or F1 mean <F1 θ And S1 mean ≥S1 θ , then the subdivided product category is divided into the third product category C3.

[0211] The packing scheme generating module comprises:

[0212] A first packing solution generating unit is used to obtain a first packing solution by using a pre-built packing solution learning model in a three-dimensional packing simulation environment;

[0213] The second packing solution generating unit is used to obtain the first constraint condition of the packing process, introduce the first constraint condition into the packing solution learning model, and obtain the second packing solution that satisfies the first constraint condition; the first constraint condition includes the weight balance threshold ω b , Placement gap threshold d g 、The upper limit of stacking layers N s and the total constraint cost C total ; Total constraint cost C total Based on the weight balance threshold ω b , Placement gap threshold d g And the upper limit of stacking layers N s calculated;

[0214] Optimal packing solution generation unit: used to calculate the comprehensive physical property parameters after packing according to the second packing solution, compare the comprehensive physical property parameters after packing with the second physical property parameters of the target packaging box, and determine whether the second packing solution meets the packing safety requirements; if not, the packing safety requirements are used as the second constraint condition, and the second constraint condition is introduced into the packing solution learning model, and the optimization is continued until a third packing solution that meets all constraints is obtained; the third packing solution is used as the optimal packing solution.

[0215] Example 3

[0216] This embodiment discloses an electronic device 500, such as Figure 7 As shown, the electronic device 500 may include one or more processors and one or more memories, wherein the memories store computer readable codes, and when the computer readable codes are run by the one or more processors, the intelligent packaging line production management method as described above may be executed.

[0217] The method or system according to the embodiment of the present application can also be used by Figure 7 The electronic device architecture shown in FIG. Figure 7 As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, a read-only memory (ROM) 503, a random access memory (RAM) 504, a communication port 505 connected to a network, an input / output component 506, a hard disk 507, etc. The storage device in the electronic device 500, such as the ROM 503 or the hard disk 507, may store the intelligent packaging line production management method provided in the present application. The production management method of the intelligent packaging line may, for example, include: obtaining the product information vector V1 of the product to be boxed and the box information vector V2 of the target box; dividing the product to be boxed into a first product category C1, a second product category C2 and a third product category C3 according to the product information vector V1 of the product to be boxed; generating a three-dimensional geometric model for each product to be boxed in the first product category C1, the second product category C2 and the third product category C3 according to the product information vector V1; generating a virtual box model that matches the internal dimensions of the target box according to the box information vector V2; constructing a three-dimensional packing simulation environment according to the three-dimensional geometric model of the product to be boxed and the virtual box model; in the three-dimensional packing simulation environment, obtaining the optimal packing solution through a pre-built packing solution learning model. Furthermore, the electronic device 500 may also include a user interface 508. Of course, Figure 7 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 7 One or more components of an electronic device are shown.

[0218] Example 4

[0219] This embodiment discloses a computer-readable storage medium 600, such as Figure 8 As shown, a computer-readable storage medium 600 stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the intelligent packaging line production management method according to the embodiment of the present application described with reference to the above figures can be executed. The storage medium 600 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0220] In addition, according to the implementation of the present application, the process described with reference to the flowchart above can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, the non-transitory machine-readable storage medium stores machine-readable instructions, the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided in the present application, for example: obtaining a product information vector V1 of the product to be boxed and a box information vector V2 of the target packaging box; according to the product information vector V1 of the product to be boxed, the product to be boxed is divided into a first product category C1, a second product category C2 and a third product category C3; according to the product information vector V1, a three-dimensional geometric model is generated for each product to be boxed in the first product category C1, the second product category C2 and the third product category C3; according to the box information vector V2, a virtual box model adapted to the internal dimensions of the target packaging box is generated; according to the three-dimensional geometric model of the product to be boxed and the virtual box model, a three-dimensional packing simulation environment is constructed; in the three-dimensional packing simulation environment, the optimal packing solution is obtained through a pre-constructed packing solution learning model. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are performed.

[0221] The methods, systems, and devices of the present application may be implemented in many ways. For example, the methods, systems, and devices of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified. In addition, in some embodiments, the present application may also be implemented as a program recorded in a recording medium, which includes machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers a recording medium storing a program for executing the method according to the present application.

[0222] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0223] The specific implementation modes described above further describe the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. Intelligent packaging line production management method, characterized in that: The method comprises: Obtain product information vector V1 of the product to be packed and box body information vector V2 of the target packaging box; According to the product information vector V1 of the product to be packed, the products to be packed are divided into a first product category C1, a second product category C2 and a third product category C3; According to the product information vector V1, a three-dimensional geometric model is generated for each product to be packed in the first product category C1, the second product category C2 and the third product category C3; according to the box information vector V2, a virtual box model adapted to the internal dimensions of the target packaging box is generated; according to the three-dimensional geometric model of the product to be packed and the virtual box model, a three-dimensional packing simulation environment is constructed; In a 3D packing simulation environment, the optimal packing solution is obtained through the pre-built packing solution learning model.

2. The intelligent packaging line production management method according to claim 1, characterized in that: The step of obtaining the product information vector V1 of the product to be packed includes: Acquire first geometric parameters of the products to be packed, wherein the first geometric parameters include the length L1, width W1 and height H1 of each product to be packed; Acquire first physical property parameters of the products to be packed, where the first physical property parameters include weight M1, material and fragility F1 of each product to be packed; Parameter association is performed on the acquired first geometric parameter and first physical property parameter of the product to be packed, so as to generate a product information vector V1 of the product to be packed.

3. The intelligent packaging line production management method according to claim 1, characterized in that: The step of obtaining the box body information vector V2 of the target packaging box includes: Acquire second geometric parameters of the target packaging box, wherein the second geometric parameters include an inner length L2, an inner width W2, and an inner height H2 of the target packaging box; Obtain a second physical property parameter of the target packaging box, wherein the second physical property parameter includes a maximum bearing weight M2 of the target packaging box max and compressive strength P2; The acquired second geometric parameter and second physical property parameter of the target packaging box are parameter-associated to generate a box body information vector V2 of the target packaging box.

4. The intelligent packaging line production management method according to claim 1, characterized in that: The classification of the products to be packed into a first product category C1, a second product category C2 and a third product category C3 comprises: Extract features from the product information vector V1 to obtain the volume S1, weight M1 and fragility F1 of each product to be packed; Preliminarily classify the products to be packed according to their volume S1 and weight M1 to obtain n1 preliminary product categories, where n1 is an integer greater than 0; Each preliminary product category is further divided according to the fragility F1 of the product to be packed, and n2 subdivided product categories are obtained, n2 ≥ n1; The n2 subdivided product categories are divided into a first product category C1, a second product category C2 and a third product category C3.

5. The intelligent packaging line production management method according to claim 4, characterized in that: The further classification of each preliminary product category according to the fragility F1 of the product to be packed includes: For each preliminary product category, calculate the mean vulnerability level F1 according to the vulnerability level F1 of the products in it. mean and vulnerability variance F1 var ; If the fragility variance F1 var Greater than the preset variance threshold var thr , then the preliminary product category is subdivided into k subcategories, k ≥ n2; All subcategories obtained from the preliminary product category segmentation are merged to form n2 subdivided product categories.

6. The intelligent packaging line production management method according to claim 5, characterized in that: The subcategories obtained by combining all preliminary product category breakdowns include: Each subcategory is regarded as an independent cluster, and the distance between any two clusters is calculated, where the distance between any two clusters is the absolute value of the difference between the mean values ​​of the vulnerability of the two clusters; Among all current clusters, find the two closest clusters C p and C q , merged into a new cluster C r ; Update the merged cluster C r The distance to other existing clusters; Repeat the merging operation until the number of clusters is reduced to n2.

7. The intelligent packaging line production management method according to claim 4, characterized in that: The method for dividing the n2 subdivided product categories into the first product category C1, the second product category C2 and the third product category C3 is: For each product category, extract its vulnerability mean F1 mean and size mean S1 mean Features, set the vulnerability threshold F1 θ , size threshold S1 θ ; If F1 mean ≥F1 θ And S1 mean ≥S1 θ , then the subdivided product category is divided into the first product category C1; If F1 mean <F1 θ And S1 mean <S1 θ , then the subdivided product category is divided into the second product category C2; If F1 mean ≥F1 θ And S1 mean <S1 θ , or F1 mean <F1 θ And S1 mean ≥S1 θ , then the subdivided product category is divided into the third product category C3.

8. The intelligent packaging line production management method according to claim 1, characterized in that: The method of obtaining the optimal packing solution through the pre-built packing solution learning model includes: In a three-dimensional packing simulation environment, a pre-built packing solution learning model is used to obtain a first packing solution; Obtaining a first constraint condition of the packing process, introducing the first constraint condition into a packing solution learning model, and obtaining a second packing solution that satisfies the first constraint condition; According to the second packing plan, the comprehensive physical property parameters after packing are calculated, and the comprehensive physical property parameters after packing are compared with the second physical property parameters of the target packaging box to determine whether the second packing plan meets the packing safety requirements; if not, the packing safety requirements are used as the second constraint condition, and the second constraint condition is introduced into the packing plan learning model to obtain the third packing plan; the third packing plan is used as the optimal packing plan.

9. The intelligent packaging line production management method according to claim 8, characterized in that: The first constraint condition includes a weight balance threshold ω b , Placement gap threshold d g 、The upper limit of stacking layers N s and the total constraint cost C total ; Total constraint cost C total Based on the weight balance threshold ω b , Placement gap threshold d g And the upper limit of stacking layers N s calculated; The comprehensive physical property parameters include the total weight M of the packed goods c and the compressive stress P of the bottom layer of cargo b ; The packing safety requirements are: c ≤M2 max And P b ≤P2.

10. An intelligent packaging line production management system, which is used to implement the intelligent packaging line production management method according to any one of claims 1 to 9, characterized in that: The system comprises: Vector extraction module: used to obtain the product information vector V1 of the product to be packed and the box information vector V2 of the target packaging box; Product classification module: used for classifying the products to be packed into a first product category C1, a second product category C2 and a third product category C3 according to the product information vector V1 of the products to be packed; 3D packing simulation environment construction module: used to generate 3D geometric models for each product to be packed in the first product category C1, the second product category C2 and the third product category C3 according to the product information vector V1; generate a virtual box model that matches the internal dimensions of the target packaging box according to the box information vector V2; and construct a 3D packing simulation environment according to the 3D geometric models of the product to be packed and the virtual box model; Packing plan generation module: used to obtain the optimal packing plan through the pre-built packing plan learning model in a three-dimensional packing simulation environment.

Images