A production line real-time data analysis and energy-saving optimization system and method

Abstract

The present invention relates to the field of intelligent manufacturing technology, and more specifically, to a real-time data analysis and energy-saving optimization system and method for a production line, the method comprising: obtaining all product orders from a database; analyzing all product orders to check whether there is a printing pattern requirement in the order, and if there is a printing pattern requirement in the product order, sorting the printing pattern; inputting the printing pattern into a pre-built first machine learning model to output product printing data; if there is no printing pattern requirement, inputting the entity performance data in the product order into a pre-built second machine learning model to output the entity production data; inputting the product printing data and the entity performance data in the product order into a pre-built third machine learning model to output a conflict identification result; ensuring that the produced products meet the order requirements, reducing the waste of raw materials, and achieving energy-saving effects.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and more specifically, to a production line real-time data analysis and energy-saving optimization system and method. Background Art

[0002] With the development of digitalization, the field of intelligent manufacturing has been increasingly valued by the country, and the formulaic production line has been the first to be affected. Because of its assembly line production model, if the production line is transformed into digitalization, it will bring significant efficiency improvement and cost reduction.

[0003] The production line digitalization technologies are as follows:

[0004] The Chinese patent with publication number CN116562500A discloses an online operation and maintenance management system for enterprise production lines. The invention belongs to the field of online production line management and includes: a data acquisition unit for acquiring enterprise production line data; a data transmission module for transmitting enterprise production line data to a data analysis module in real time through wireless communication technology; a data analysis module for analyzing enterprise production line data in real time to obtain the operation status of the enterprise production line, which includes normal status and abnormal status; an abnormality processing module for processing the enterprise production line data corresponding to the abnormal status based on the abnormal status to restore the normal operation status of the enterprise production line. The invention can monitor the operation of each device and process status of the enterprise production line in real time online. If an abnormal situation occurs in the enterprise production line, the abnormal situation can be handled in time. At the same time, the abnormal position of the entire production line can be grasped at the first time, and the specific position of the abnormal operation can be accurately determined.

[0005] However, the existing related technologies do not take into account that during the manufacturing process of products with printing pattern requirements, the production line system may not be able to accurately identify and extract the required product printing data; secondly, during the production process, the conflict between the product printing data and the body performance data is ignored, which may easily lead to production errors and the production of products that do not meet the order or even defective products, which will cause serious losses and waste of resources. Summary of the invention

[0006] The present invention provides a production line real-time data analysis and energy-saving optimization method, the method comprising:

[0007] Get all product orders from the database;

[0008] Analyze all product orders to see if there is a need for printing patterns in the orders. If there is a need for printing patterns in the product orders, arrange the printing patterns;

[0009] Input the printing pattern into the pre-built first machine learning model to output product printing data; if there is no printing pattern demand, input the body performance data in the product order into the pre-built second machine learning model to output body production data, which includes body raw material ratio, mixing parameters and molding parameters;

[0010] Input the product printing data and the ontology performance data in the product order into a pre-built third machine learning model to output a conflict identification result, where the conflict identification result includes no conflict and conflict;

[0011] The conflicting product printing data and entity performance data are input into the pre-trained fourth machine learning model to output corrected production data, wherein the corrected production data includes corrected product printing data and corrected entity production data.

[0012] Furthermore, the method for constructing the first machine learning model is:

[0013] The first machine learning model includes a main network and a sub-network; wherein the main network includes four convolution blocks, which are respectively denoted as the first convolution block, the second convolution block, the third convolution block and the fourth convolution block, each of which includes a convolution layer and an activation layer, and the second convolution block and the fourth convolution block also include a pooling layer; the fourth convolution block is sequentially connected to a drop layer, a first fully connected layer, a second fully connected layer and a softmax function layer; the four convolution blocks and the drop layer in the sub-network and the main network are connected in parallel; the sub-network includes two convolution blocks, a fully connected layer and an activation layer, the two convolution blocks are sequentially denoted as the fifth convolution block and the sixth convolution block, the sub-network includes a fully connected layer denoted as the third fully connected layer, and an activation layer is connected to the back of the third fully connected layer.

[0014] Furthermore, the first machine learning model training process comprises the following steps:

[0015] Step 1: Get Printed pattern, training set and validation set Two subsets, each containing multiple printed patterns;

[0016] Let the first machine learning model be , the first machine learning model is initialized, and the initialization parameters satisfy the Gaussian distribution. During the training process, each time from the training set Randomly select The printed pattern was fed into the first machine learning model ,set up The printed patterns are , The corresponding product printing data are: , the first machine learning model The cross entropy loss function corresponding to each training process is ,but The calculation formula is as follows:

[0017] ;

[0018] In the formula, For the Printed pattern, For the The product printing data corresponding to the printed pattern, For the Prediction product printing data of printed patterns, is a hyperparameter, is the regularization of the weights of the first fully connected layer and the second fully connected layer;

[0019] Step 2: From the training set Repeated random sampling The printed pattern was fed into the first machine learning model During the training process, the first machine learning model was trained using the Adam gradient descent optimization algorithm with a learning rate of 0.002. When converged, the first machine learning model stops training;

[0020] Step 3: After training, use the validation set The performance of the first machine learning model is evaluated on

[0021] Furthermore, the third machine learning model building method includes:

[0022] A conflict identification data set is collected in advance, wherein the conflict identification data includes identification feature data and conflict identification results corresponding to the identification feature data, wherein the identification feature data includes entity performance data and product printing data; the conflict identification data set is divided into a training set and a test set, and a regression network is constructed. The identification feature data in the training set is used as input data of the regression network, and the conflict identification results in the training set are used as output data of the regression network. The regression network is trained to obtain an initial regression network for predicting conflict identification results. The initial regression network is tested using the test set, and a regression network that satisfies a preset error value is output as a third machine learning model.

[0023] Furthermore, the construction method of the second machine learning model is:

[0024] The second machine learning model uses a random forest regression model to collect M historical ontology data, which includes ontology performance data and ontology production data corresponding to the ontology performance data; N tree subsets are sampled from the M historical ontology data, each tree subset contains a preset proportion of historical ontology data, and the N tree subsets will be used to train each decision tree:

[0025] The single decision tree construction method includes:

[0026] Initialize model parameters: determine the number of trees in the forest N_trees; the maximum depth of each tree max_depth; the minimum number of samples for internal nodes min_samples_split; the minimum number of samples for leaf nodes min_samples_leaf;

[0027] Use the CART algorithm to recursively build the first decision tree:

[0028] Divide the data in the tree subset into a training set and validation set , For the training set Middle A specific historical ontology data, using features Indicates one of high barrier properties, heat resistance, low temperature resistance, environmental protection, oil resistance, antistatic properties, tear resistance and light weight. Features Information gain rate :

[0029] For each feature The corresponding information gain rate Calculate and select the maximum information gain rate The corresponding features As a node, and according to the maximum information gain rate The corresponding features The training set Divide into decision subsets so that the number of historical ontology data in each decision subset is the same;

[0030] According to the maximum information gain rate Divide the decision subsets and repeat the algorithm recursive process until the decision subsets cannot be divided into smaller decision subsets; the trained decision tree model is tested on the validation set. The performance evaluation is carried out on the network, and finally the construction of the first decision tree model is completed.

[0031] Furthermore, the random forest construction method includes:

[0032] Collect M pieces of historical ontology data, which include ontology performance data and ontology production data corresponding to the ontology performance data; sample N tree subsets from the M pieces of historical ontology data, and the N tree subsets will be used to train each decision tree;

[0033] Use the other N-1 tree subsets in turn to train N-1 decision trees; the generation process of each decision tree is independent and based on the CART algorithm; finally, a random forest of N decision trees is constructed.

[0034] Furthermore, the second machine learning model training process:

[0035] Calculate the loss function value on all N decision trees; input the training data into each decision tree and record the decision output, which is the ontology production data corresponding to the ontology performance data; calculate the mean square loss of the error between the average output of the N decision trees and the true value; reduce the mean square loss through parameter tuning, and select the model parameters with the smallest mean square loss as the model parameters of the second machine learning model.

[0036] Furthermore, the fourth machine learning model training method includes:

[0037] Collecting a correction data set in advance, wherein the correction data includes conflicting characteristic data and corrected production data corresponding to the conflicting characteristic data;

[0038] The corrected data set is divided into a training set and a test set, and a regression network is constructed. The conflict feature data in the training set is used as the input data of the regression network, and the corrected production data in the training set is used as the output data of the regression network. The regression network is trained to obtain an initial regression network for generating corrected production data. The initial regression network is tested with the test set, and the output regression network that satisfies a preset error value is used as the fourth machine learning model.

[0039] A production line real-time data analysis and energy-saving optimization system, implementing the production line real-time data analysis and energy-saving optimization method, the system includes:

[0040] Collection module, used to obtain all product orders from the database;

[0041] The first analysis module is used to analyze all product orders to check whether there is a printing pattern requirement in the order. If there is a printing pattern requirement in the product order, the printing pattern is sorted;

[0042] The second analysis module is used to input the printing pattern into the pre-built first machine learning model to output product printing data; if there is no printing pattern demand, the body performance data in the product order is input into the pre-built second machine learning model to output body production data, and the body production data includes body raw material ratio, mixing parameters and molding parameters;

[0043] A third analysis module inputs the product printing data and the ontology performance data in the product order into a pre-built third machine learning model to output a conflict identification result, where the conflict identification result includes no conflict and conflict;

[0044] A fourth analysis module, inputting the conflicting product printing data and the body performance data into a pre-trained fourth machine learning model to output corrected production data, wherein the corrected production data includes corrected product printing data and corrected body production data;

[0045] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a production line real-time data analysis and energy-saving optimization method is implemented.

[0046] A computer-readable storage medium stores a computer program, and when the computer program is executed, the method for real-time data analysis and energy-saving optimization of a production line is implemented.

[0047] The present invention provides a production line real-time data analysis and energy-saving optimization system and method, and has the following effects and advantages:

[0048] Effect: Through the collection module, the system can obtain all product orders from the database to ensure the integrity and accuracy of the orders. Then, the second analysis module analyzes the orders to identify whether there are product orders with printing pattern requirements. If so, the printing pattern is input into the first machine learning model to output product printing data; the product printing data and the entity performance data in the product order are input into the pre-built third machine learning model to output the conflict identification results, which include no conflict and conflict; the conflicting product printing data and entity performance data are input into the pre-trained fourth machine learning model to output the corrected production data, which includes the corrected product printing data and the corrected entity production data; and the corrected production data is sent to the printing equipment and the entity production equipment.

[0049] Advantages: Through machine learning technology and automated processes, the automatic collection, analysis and execution of product orders are realized, which greatly reduces the workload and time consumption of manual processing; it can quickly and accurately analyze the required product printing data when there is a printing pattern requirement in the product order, and automatically identify whether there is a conflict between the product printing data and the body performance data. In the event of a conflict, it automatically generates corrected production data, which makes the production process more efficient and reduces the possibility of human error. At the same time, it also indirectly reduces the waste of raw materials and achieves energy-saving effects; through the input and output of the machine learning model, it can provide more accurate and reliable data to ensure that the products produced meet the user's product requirements; this system can process product orders in real time and in large quantities, and send instructions according to demand to ensure the timeliness and flexibility of production. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A transmission schematic diagram of a production line real-time data analysis and energy-saving optimization system of the present invention;

[0051] Figure 2 A flow chart of a production line real-time data analysis and energy-saving optimization method of the present invention;

[0052] Figure 3 is a schematic diagram of an electronic device of the present invention;

[0053] Figure 4 Schematic diagram of the storage medium of the present invention. DETAILED DESCRIPTION

[0054] 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.

[0055] Example 1

[0056] like Figure 1 As shown, a production line real-time data analysis and energy-saving optimization system includes a collection module, a first analysis module, a second analysis module, a third analysis module and a fourth analysis module; wherein each module is connected by wired or wireless means:

[0057] The collection module is used to obtain all product orders from the database.

[0058] The first analysis module is used to analyze all product orders to check whether there is a printing pattern requirement in the order. If there is a printing pattern requirement in the product order, the printing pattern is sorted.

[0059] The second analysis module is used to input the printing pattern into the pre-built first machine learning model to output product printing data; if there is no printing pattern demand, the body performance data in the product order is input into the pre-built second machine learning model to output the body production data, and the body production data is sent to the body production equipment. The body production data is used to control the body production equipment. The production equipment includes body batching equipment, mixing equipment, and molding equipment. The body production data includes body raw material ratio, mixing parameters and molding parameters. The body raw material ratio is used to control the body batching equipment, the mixing parameters are used to control the mixing equipment, and the molding parameters are used to control the molding equipment.

[0060] In this embodiment, taking the product order as a food packaging order as an example, and the main body as a packaging bag body as an example, the description is made. The performance data of the main body include high barrier properties, heat resistance, low temperature resistance, environmental protection, oil resistance, anti-static property, tear resistance and lightweight. The high barrier property can effectively block oxygen, water vapor, odor, etc., to maintain the shelf life of food, and the applicable raw materials include EVOH (ethylene-vinyl alcohol copolymer), PVDC (polyvinylidene chloride), etc.; heat-resistant food packaging is used for food packaging heated in microwave ovens or hot water, and the applicable raw materials include PET (polyethylene terephthalate) and PP (polypropylene); environmentally friendly food packaging uses degradable or recyclable packaging materials, and the applicable raw materials include PLA (polylactic acid), degradable polyethylene, etc. The raw materials and manufacturing processes corresponding to each type of demand are different.

[0061] The method for determining whether a product order has a demand for printed patterns includes automatic analysis and manual confirmation processes;

[0062] Automatic analysis includes:

[0063] Get all product order data from the database.

[0064] Check the product order for any fields or labels related to printing, such as "Printing pattern" or "Printing requirements." Product orders usually include a specific description of the product's appearance, including color, pattern, etc.

[0065] Use the keyword matching method to detect whether the product order contains descriptions related to printing. For example, words such as "printing", "pattern", "logo", and "design" can be used to match the product order content.

[0066] For product orders with printing descriptions, further check whether there are pattern files, design drafts or clear pattern requirements attached.

[0067] Based on the specific structure of the product order, determine how the pattern requirements are expressed, such as through a separate pattern file or text description.

[0068] If the automatic analysis cannot accurately determine whether there is a pattern requirement, it returns error 0 and starts the manual confirmation process to manually confirm the order with uncertain description.

[0069] Product printing data includes printing ink type, ink ratio parameters, printing effect, and printing control parameters; printing control parameters include roller pressure, temperature, speed and other parameters. Printing control parameters are the control parameters of the printing equipment. Printing effects include gloss, matte, embossing, hot stamping, etc. Different effects require different inks.

[0070] Printed patterns can show pattern structure and effects, as shown below:

[0071] Example 1, geometric shape (pattern structure) + gloss (printing effect), the corresponding printing ink type is high gloss ink, the printing effect is light effect, the roller pressure is high to ensure uniform coverage of the high gloss ink, the temperature is moderate to prevent the ink from losing gloss due to high temperature, the speed is medium to ensure that the ink is fully dried and maintains gloss. High gloss ink requires higher roller pressure and moderate temperature to avoid gloss reduction while ensuring color saturation; ink ratio parameters: acrylate copolymer: 60%, light stabilizer: 10%, leveling agent: 5%, solvent: 25%.

[0072] Example 2, soft gradient pattern + matte effect, the corresponding printing ink type is matte ink, which absorbs light and reduces reflection. The roller pressure is low to ensure uniform coverage of the matte ink, the temperature is low to prevent the matte ink from becoming shiny due to excessive temperature, and the speed is low to ensure stable drying of the matte ink. Matte ink requires lower roller pressure and temperature to ensure that the pattern surface remains soft and matte; ink ratio parameters: matting agent (such as silica): 50%, anti-thermal denaturation component: 15%, solvent: 20%, to ensure the uniformity and coverage of the ink, pigment: 15%.

[0073] Example 3, three-dimensional pattern + relief effect, the corresponding printing ink type is high-viscosity ink to ensure the thickness and three-dimensional effect of the ink in the relief area, the roller pressure is high pressure to form a clear relief effect, the temperature is medium and low to prevent the ink from losing its three-dimensional effect due to excessive temperature flow, the speed is medium speed to ensure the integrity and clarity of the relief effect, high-viscosity ink combined with high pressure and moderate temperature helps to form a clear relief effect; ink ratio parameters: thickener (such as polyurethane resin): 40%, plasticizer: 20%, filler: 10%, pigment: 30%.

[0074] Example 4, abstract pattern + glossy hot stamping effect, the corresponding printing ink type is glossy metallic ink, which highlights the color and metallic effect. The roller pressure is medium-high pressure to enhance the gloss and hot stamping effect. The temperature is medium-high temperature to ensure the perfect fusion of gloss and hot stamping ink. The speed is medium speed to prevent the pattern from shifting or distorting. The combination of glossy metallic ink and hot stamping film requires precise control to ensure the visual impact of the pattern. Ink ratio parameters: metallic pigment: 45%, antioxidant: 10%, resin: 30%, solvent: 15%.

[0075] Among them, the printed patterns in the product order may be affected by light, blur, and low contrast; these problems may cause information loss in the printed pattern or reduce the usability of the printed pattern. Processing low-quality printed patterns may have a negative impact on the accuracy of subsequent tasks. In addition, the collected printed patterns may have different sizes and proportions, which may lead to difficulties in subsequent processing and analysis. Therefore, it is necessary to use printed pattern sorting technology to sort the printed patterns in the product order. The process of sorting the printed patterns includes uniformly cropping the printed patterns in the product order to 256×256 pixels, adjusting the contrast and denoising. The denoising method uses bilateral filtering. Bilateral filtering retains edge information while smoothing the printed pattern, which is suitable for maintaining the clarity and details of the printed pattern; it takes into account the spatial distance between pixels and the difference between pixel values, making the filtering result more natural.

[0076] The method for constructing the first machine learning model is:

[0077] The first machine learning model includes a main network and a sub-network; wherein the main network includes four convolution blocks, which are respectively denoted as the first convolution block, the second convolution block, the third convolution block and the fourth convolution block, each of which includes a convolution layer and an activation layer, and the second convolution block and the fourth convolution block also include a pooling layer; the fourth convolution block is sequentially connected to a drop layer, a first fully connected layer, a second fully connected layer and a softmax function layer; the four convolution blocks and the drop layer in the sub-network and the main network are connected in parallel; the sub-network includes two convolution blocks, a fully connected layer and an activation layer, the two convolution blocks are sequentially denoted as the fifth convolution block and the sixth convolution block, the sub-network includes a fully connected layer denoted as the third fully connected layer, and an activation layer is connected to the back of the third fully connected layer.

[0078] The first machine learning model training process comprises the following steps:

[0079] Step 1: Get Printed pattern, training set and validation set Two subsets, each subset contains multiple printed patterns.

[0080] Let the first machine learning model be , the first machine learning model is initialized, and the initialization parameters satisfy the Gaussian distribution. During the training process, each time from the training set Randomly select The printed pattern was fed into the first machine learning model ,set up The printed patterns are , The corresponding product printing data are: , the first machine learning model The cross entropy loss function corresponding to each training process is ,but The calculation formula is as follows:

[0081] ;

[0082] In the formula, For the Printed pattern, For the The product printing data corresponding to each printed pattern (collected in advance, that is, each printed pattern corresponds to a set of product printing data), For the Prediction product printing data of printed patterns, is a hyperparameter, is the regularization of the weights of the first and second fully connected layers.

[0083] Step 2: From the training set Repeated random sampling The printed pattern was fed into the first machine learning model During the training process, the first machine learning model was trained using the Adam gradient descent optimization algorithm with a learning rate of 0.002. When converged, the first machine learning model stops training.

[0084] Step 3: After training, use the validation set The performance of the first machine learning model is evaluated on

[0085] The construction method of the second machine learning model is:

[0086] The second machine learning model uses a random forest regression model to collect M historical ontology data, which include ontology performance data and ontology production data corresponding to the ontology performance data; N tree subsets are sampled from the M historical ontology data, each tree subset contains a preset proportion of historical ontology data, and the N tree subsets will be used to train each decision tree.

[0087] Initialize model parameters: determine the number of trees in the forest N_trees; the maximum depth of each tree max_depth; the minimum number of samples for internal nodes min_samples_split; the minimum number of samples for leaf nodes min_samples_leaf.

[0088] Use the CART algorithm to recursively build the first decision tree:

[0089] Divide the data in the tree subset into a training set and validation set , For the training set Middle A specific historical ontology data, using features Indicates one of high barrier properties, heat resistance, low temperature resistance, environmental protection, oil resistance, antistatic properties, tear resistance and light weight. Features Information gain rate , The calculation formula is as follows:

[0090] ;

[0091] In the formula, , For the training set According to the characteristics The decision subsets divided into for The information entropy of For the training set The information entropy of and The calculation methods include:

[0092] ;

[0093] In the formula, For the training set The amount of data contained in For the training set Middle The number of specific historical ontology data.

[0094] ;

[0095] In the formula, express Middle The ontology production data corresponding to the specific ontology performance data; for each feature The corresponding information gain rate Calculate and select the maximum information gain rate The corresponding features As a node, and according to the maximum information gain rate The corresponding features The training set Divide into decision subsets so that the number of historical ontology data in each decision subset is the same.

[0096] According to the maximum information gain rate Divide the decision subsets and repeat the algorithm recursive process until the decision subsets cannot be divided into smaller decision subsets; the trained decision tree model is tested on the validation set. The performance evaluation is carried out on the network, and finally the construction of the first decision tree model is completed.

[0097] Use the other N-1 tree subsets in turn to train N-1 decision trees; the generation process of each decision tree is independent and based on the CART algorithm; finally, a random forest of N decision trees is constructed.

[0098] The second machine learning model training process:

[0099] Calculate the loss function value on all N decision trees; input the training data into each decision tree and record the decision output, which is the ontology production data corresponding to the ontology performance data; calculate the mean square loss of the error between the average output of the N decision trees and the true value; reduce the mean square loss through parameter tuning, and select the model parameters with the smallest mean square loss as the model parameters of the second machine learning model.

[0100] The second analysis module inputs the product printing data and the entity performance data in the product order into a pre-built third machine learning model to output a conflict identification result, and the conflict identification result includes no conflict and conflict.

[0101] The third machine learning model building method includes:

[0102] A conflict identification data set is collected in advance, wherein the conflict identification data includes identification feature data and a conflict identification result corresponding to the identification feature data, wherein the identification feature data includes body performance data and product printing data; examples of conflict identification data are as follows:

[0103] Case 1: Conflict between glossy ink + glossy effect + high-barrier packaging bags: High-barrier packaging bags contain 50% EVOH and 50% PVDC, which may reduce the adhesion of glossy ink and cause uneven gloss effect.

[0104] Case 2: Conflict between matte ink + matte effect + heat-resistant packaging bags: The heat-resistant packaging bags contain 60% polyamide and have a high surface smoothness, which makes it difficult for matte ink to adhere evenly and affects the matte effect.

[0105] Case 3: Conflict between metallic ink + hot stamping effect + low-temperature resistant packaging bags: Low-temperature resistant packaging bags contain 70% polyvinyl alcohol (PVA). At low temperatures, the metallic ink may become dull, affecting the hot stamping effect.

[0106] Case 4: Conflict between high-reflectivity glossy ink + glossy effect + environmentally friendly packaging bags: Environmentally friendly packaging bags usually use biodegradable materials, such as 70% polylactic acid (PLA), which may chemically react with high-reflectivity glossy ink, resulting in a decrease in glossiness.

[0107] Case 5: Conflict between matte embossing ink + embossing effect + oil-proof packaging bags: The oil-proof packaging bags contain 80% polyethylene (PE) coating, which makes it difficult for matte embossing ink to adhere and the embossing effect is poor.

[0108] Case 6: Conflict between high-contrast color ink + matte effect + anti-static packaging bags: 5% antistatic agents are added to anti-static packaging bags. These additives may affect the color performance of high-contrast color inks, resulting in unsatisfactory matte effects.

[0109] Case 7: Conflict between glossy metallic ink + hot stamping effect + tear-resistant packaging bags: Tear-resistant packaging bags contain 60% polyethylene terephthalate (PET), whose high strength and rigidity may cause cracks in the glossy metallic ink during hot stamping, affecting the effect.

[0110] Case 8: Conflict between high-viscosity ink + embossing effect + lightweight packaging bags: Lightweight packaging bags use 90% polyethylene (PE) thin-wall material. High-viscosity ink may be difficult to apply evenly during the embossing printing process, affecting the effect.

[0111] Case 9: Conflict between matte ink + matte effect + high barrier packaging bags: The material ratio of high barrier packaging bags is 50% EVOH and 50% PVDC. These materials may interact with matte ink, resulting in unstable matte effect.

[0112] Case 10: Glossy ink + glossy effect + tear-resistant packaging bags do not conflict.

[0113] The conflict identification data set is divided into a training set and a test set, and a regression network is constructed. The identification feature data in the training set is used as the input data of the regression network, and the conflict identification results in the training set are used as the output data of the regression network. The regression network is trained to obtain an initial regression network for predicting conflict identification results. The initial regression network is tested using the test set, and the output regression network that satisfies a preset error value is used as the third machine learning model.

[0114] This step screens out conflicting product orders and proceeds to the next step to avoid direct production based on product orders, which would cause a decline in product quality and reduce losses and waste of resources. The product printing data and body production data in the non-conflicting product orders are sent to the printing equipment and body production equipment in the production line.

[0115] The third analysis module inputs the conflicting product printing data and entity performance data into the pre-trained fourth machine learning model to output corrected production data, wherein the corrected production data includes corrected product printing data and corrected entity production data; and sends the corrected production data to the printing equipment and the entity production equipment.

[0116] The fourth machine learning model training method includes:

[0117] Collect a correction data set in advance, wherein the correction data includes conflicting characteristic data and corrected production data corresponding to the conflicting characteristic data; examples of the correction data are as follows:

[0118] Case 1: Solution: Adjust the ink formula: Add 15% adhesion promoter (such as acrylate copolymer) to the gloss ink to improve the adhesion of the ink on the PVDC surface and improve the uniformity of the gloss effect; adjust the material ratio: reduce the EVOH ratio to 40% and increase the PVDC to 60% to enhance the performance of the gloss ink.

[0119] Case 2: Solution: Ink formula adjustment: Add 5% matting agent (such as silica) and 10% fine filler to the matte ink to increase the surface roughness to improve the matte effect; Material ratio adjustment: Reduce the polyamide ratio from 60% to 50%, and add 10% polypropylene copolymer to make the surface rougher and improve the adhesion of the ink.

[0120] Case 3: Solution: Adjust the ink formula: Add 10% cold-resistant additives (such as fluoride compounds) and 5% antioxidants to the metallic ink to maintain the gloss at low temperatures; Adjust the material ratio: Reduce the PVA ratio from 70% to 60%, and add 10% ethylene-vinyl acetate copolymer (EVA) to provide better low-temperature toughness and reduce the impact on the gloss of the metallic ink.

[0121] Case 4: Solution: Adjust the ink formula: Use biodegradable glossy ink and add 20% polycaprolactone (PCL) as ink binder to enhance compatibility with PLA; Adjust the material ratio: Reduce the PLA ratio to 60% and add 10% bio-based polyester (PBAT) to improve the glossiness of the packaging bag surface and reduce chemical reactions.

[0122] Case 5: Solution: Ink formula adjustment: Add 10% of oil-resistant additives (such as fluorinated polymers) and 5% of tackifiers to the matte relief ink to enhance its adhesion on the PE surface; Material ratio adjustment: Reduce the PE coating ratio to 70% and increase 10% of ethylene-vinyl acetate copolymer (EVA) to improve the ink adhesion and relief performance.

[0123] Case 6: Solution: Ink formula adjustment: Add 3% antistatic additives (such as quaternary ammonium salts) to high-contrast color inks to improve their compatibility with antistatic materials, and increase pigment concentration by 10% to ensure color saturation; Material ratio adjustment: Reduce the proportion of antistatic agent to 3% to reduce the impact on ink color while maintaining antistatic performance.

[0124] Case 8: Solution: Adjust the ink formula: reduce the proportion of thickener to 5% in high-viscosity ink, and add 2% leveling agent to improve coating performance and ensure the uniformity of the relief effect; adjust the material ratio: reduce the thickness of the PE film to 85%, and add 5% polypropylene (PP) layer to improve surface flatness and improve relief printing effect.

[0125] Case 7: Solution: Adjust the ink formula: Add 5% flexible resin (such as ethylene-vinyl acetate copolymer EVA) and 5% plasticizer to the glossy metallic ink to improve the elasticity of the ink and prevent cracks; adjust the material ratio: reduce the PET ratio to 50%, and add 10% polyethylene (PE) to improve the flexibility of the packaging material and reduce the stress on the ink.

[0126] Case 9: Solution: Adjust the material ratio: reduce the EVOH ratio to 45% and increase the PVDC to 55% to reduce the chemical activity on the material surface and make the matte ink performance more consistent; adjust the ink formula: add 5% matte agent (such as titanium dioxide) and 3% surfactant to improve the stability and matte effect of the ink.

[0127] The corrected data set is divided into a training set and a test set, and a regression network is constructed. The conflict feature data in the training set is used as the input data of the regression network, and the corrected production data in the training set is used as the output data of the regression network. The regression network is trained to obtain an initial regression network for generating corrected production data. The initial regression network is tested with the test set, and the output regression network that satisfies a preset error value is used as the fourth machine learning model.

[0128] Through the collection module, the system can obtain all product orders from the database to ensure the integrity and accuracy of the orders. Then, the second analysis module analyzes the orders to identify whether there are product orders with printing pattern requirements. If so, the printing pattern is input into the first machine learning model to output product printing data; the product printing data and the entity performance data in the product order are input into the pre-built third machine learning model to output the conflict identification result, which includes no conflict and conflict; the conflicting product printing data and entity performance data are input into the pre-trained fourth machine learning model to output the corrected production data, which includes the corrected product printing data and the corrected entity production data; and the corrected production data is sent to the printing equipment and the entity production equipment.

[0129] Through machine learning technology and automated processes, the automatic collection, analysis and execution of product orders are realized, which greatly reduces the workload and time consumption of manual processing; it can quickly and accurately analyze the required product printing data when there is a printing pattern requirement in the product order, and automatically identify whether there is a conflict between the product printing data and the body performance data. In the event of a conflict, it automatically generates corrected production data, which makes the production process more efficient and reduces the possibility of human error, while also indirectly reducing the waste of raw materials and achieving energy-saving effects; through the input and output of the machine learning model, it can provide more accurate and reliable data to ensure that the products produced meet the user's needs for the product; this system can process product orders in real time and in large quantities, and send instructions according to demand to ensure the timeliness and flexibility of production.

[0130] Example 2

[0131] like Figure 2 As shown, a production line real-time data analysis and energy-saving optimization method.

[0132] Get all product orders from the database;

[0133] Analyze all product orders to see if there is a need for printing patterns in the orders. If there is a need for printing patterns in the product orders, arrange the printing patterns;

[0134] Input the printing pattern into the pre-built first machine learning model to output product printing data; if there is no printing pattern demand, input the body performance data in the product order into the pre-built second machine learning model to output body production data, which includes body raw material ratio, mixing parameters and molding parameters;

[0135] Input the product printing data and the ontology performance data in the product order into a pre-built third machine learning model to output a conflict identification result, where the conflict identification result includes no conflict and conflict;

[0136] The conflicting product printing data and the entity performance data are input into a pre-trained fourth machine learning model to output corrected production data, wherein the corrected production data includes the corrected product printing data and the corrected entity production data;

[0137] Furthermore, the method for constructing the first machine learning model is:

[0138] The first machine learning model includes a main network and a sub-network; wherein the main network includes four convolution blocks, which are respectively denoted as the first convolution block, the second convolution block, the third convolution block and the fourth convolution block, each of which includes a convolution layer and an activation layer, and the second convolution block and the fourth convolution block also include a pooling layer; the fourth convolution block is sequentially connected to a drop layer, a first fully connected layer, a second fully connected layer and a softmax function layer; the four convolution blocks and the drop layer in the sub-network and the main network are connected in parallel; the sub-network includes two convolution blocks, a fully connected layer and an activation layer, the two convolution blocks are sequentially denoted as the fifth convolution block and the sixth convolution block, the sub-network includes a fully connected layer denoted as the third fully connected layer, and an activation layer is connected to the back of the third fully connected layer.

[0139] Furthermore, the first machine learning model training process comprises the following steps:

[0140] Step 1: Get Printed pattern, training set and validation set Two subsets, each containing multiple printed patterns;

[0141] Let the first machine learning model be , the first machine learning model is initialized, and the initialization parameters satisfy the Gaussian distribution. During the training process, each time from the training set Randomly select The printed pattern was fed into the first machine learning model ,set up The printed patterns are , The corresponding product printing data are: , the first machine learning model The cross entropy loss function corresponding to each training process is ,but The calculation formula is as follows:

[0142] ;

[0143] In the formula, For the Printed pattern, For the The product printing data corresponding to the printed pattern, For the Prediction product printing data of printed patterns, is a hyperparameter, is the regularization of the weights of the first fully connected layer and the second fully connected layer;

[0144] Step 2: From the training set Repeated random sampling The printed pattern was fed into the first machine learning model During the training process, the first machine learning model was trained using the Adam gradient descent optimization algorithm with a learning rate of 0.002. When converged, the first machine learning model stops training;

[0145] Step 3: After training, use the validation set The performance of the first machine learning model is evaluated on

[0146] Furthermore, the third machine learning model building method includes:

[0147] A conflict identification data set is collected in advance, wherein the conflict identification data includes identification feature data and conflict identification results corresponding to the identification feature data, wherein the identification feature data includes entity performance data and product printing data; the conflict identification data set is divided into a training set and a test set, and a regression network is constructed. The identification feature data in the training set is used as input data of the regression network, and the conflict identification results in the training set are used as output data of the regression network. The regression network is trained to obtain an initial regression network for predicting conflict identification results. The initial regression network is tested using the test set, and a regression network that satisfies a preset error value is output as a third machine learning model.

[0148] Furthermore, the construction method of the second machine learning model is:

[0149] The second machine learning model uses a random forest regression model to collect M historical ontology data, which includes ontology performance data and ontology production data corresponding to the ontology performance data; N tree subsets are sampled from the M historical ontology data, each tree subset contains a preset proportion of historical ontology data, and the N tree subsets will be used to train each decision tree:

[0150] The single decision tree construction method includes:

[0151] Initialize model parameters: determine the number of trees in the forest N_trees; the maximum depth of each tree max_depth; the minimum number of samples for internal nodes min_samples_split; the minimum number of samples for leaf nodes min_samples_leaf;

[0152] Use the CART algorithm to recursively build the first decision tree:

[0153] Divide the data in the tree subset into a training set and validation set , For the training set Middle A specific historical ontology data, using features Indicates one of high barrier properties, heat resistance, low temperature resistance, environmental protection, oil resistance, antistatic properties, tear resistance and light weight. Features Information gain rate :

[0154] For each feature The corresponding information gain rate Calculate and select the maximum information gain rate The corresponding features As a node, and according to the maximum information gain rate The corresponding features The training set Divide into decision subsets so that the number of historical ontology data in each decision subset is the same;

[0155] According to the maximum information gain rate Divide the decision subsets and repeat the algorithm recursive process until the decision subsets cannot be divided into smaller decision subsets; the trained decision tree model is tested on the validation set. The performance evaluation is carried out on the network, and finally the construction of the first decision tree model is completed.

[0156] Furthermore, the random forest construction method includes:

[0157] Collect M pieces of historical ontology data, which include ontology performance data and ontology production data corresponding to the ontology performance data; sample N tree subsets from the M pieces of historical ontology data, and the N tree subsets will be used to train each decision tree;

[0158] Use the other N-1 tree subsets in turn to train N-1 decision trees; the generation process of each decision tree is independent and based on the CART algorithm; finally, a random forest of N decision trees is constructed.

[0159] Furthermore, the second machine learning model training process:

[0160] Calculate the loss function value on all N decision trees; input the training data into each decision tree and record the decision output, which is the ontology production data corresponding to the ontology performance data; calculate the mean square loss of the error between the average output of the N decision trees and the true value; reduce the mean square loss through parameter tuning, and select the model parameters with the smallest mean square loss as the model parameters of the second machine learning model.

[0161] Furthermore, the fourth machine learning model training method includes:

[0162] Collecting a correction data set in advance, wherein the correction data includes conflicting characteristic data and corrected production data corresponding to the conflicting characteristic data;

[0163] The corrected data set is divided into a training set and a test set, and a regression network is constructed. The conflict feature data in the training set is used as the input data of the regression network, and the corrected production data in the training set is used as the output data of the regression network. The regression network is trained to obtain an initial regression network for generating corrected production data. The initial regression network is tested with the test set, and the output regression network that satisfies a preset error value is used as the fourth machine learning model.

[0164] Example 3

[0165] See also Figure 3 As shown, according to another aspect of the present application, an electronic device 500 is also provided. The electronic device 500 may include one or more processors and one or more memories. The memories store computer readable codes, and when the computer readable codes are executed by one or more processors, a production line real-time data analysis and energy-saving optimization method as described above may be executed.

[0166] The method or system according to the embodiment of the present application can also be used Figure 3 The electronic device architecture shown in FIG. Figure 3As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, ROM 503, RAM 504, a communication port 505 connected to a network, an input / output 506, a hard disk 507, etc. The storage device in the electronic device 500, such as ROM 503 or hard disk 507, may store a production line real-time data analysis and optimization method provided by the present application. Furthermore, the electronic device 500 may also include a user interface 508. Of course, Figure 3 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 3 One or more components of an electronic device are shown.

[0167] Example 4

[0168] See also Figure 4 As shown, it is a computer storage medium according to an embodiment of the present application. Computer-readable instructions are stored on the computer storage medium. When the computer-readable instructions are executed by the processor, a production line real-time data analysis and energy-saving optimization method according to an embodiment of the present application described with reference to the above figures can be executed. The storage medium 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.

[0169] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the attached claims.

[0170] Finally: The above description is only a preferred embodiment 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. A production line real-time data analysis and energy-saving optimization method, characterized in that: Methods include: Get all product orders from the database; Analyze all product orders to see if there is a need for printing patterns in the orders. If there is a need for printing patterns in the product orders, arrange the printing patterns; Input the printing pattern into the pre-built first machine learning model to output product printing data; the first machine learning model is a convolutional neural network model; the product printing data includes printing ink type, ink ratio parameters, printing effect, and printing control parameters; if there is no printing pattern demand, the body performance data in the product order is input into the pre-built second machine learning model to output the body production data, which includes the body raw material ratio, mixing parameters, and molding parameters; the body performance data includes high barrier properties, heat resistance, low temperature resistance, environmental protection, oil resistance, antistatic properties, tear resistance, and lightweight; the second machine learning model is a random forest regression model; Input the product printing data and the ontology performance data in the product order into a pre-built third machine learning model to output a conflict identification result, wherein the conflict identification result includes no conflict and conflict; the third machine learning model is a regression network model; Inputting conflicting product printing data and entity performance data into a pre-trained fourth machine learning model to output corrected production data, the corrected production data including corrected product printing data and corrected entity production data; the fourth machine learning model is a regression network model; The method for constructing the first machine learning model is: The first machine learning model includes a main network and a sub-network; wherein the main network includes four convolution blocks, which are respectively denoted as the first convolution block, the second convolution block, the third convolution block and the fourth convolution block, each of which includes a convolution layer and an activation layer, and the second convolution block and the fourth convolution block also include a pooling layer; the fourth convolution block is sequentially connected to a drop layer, a first fully connected layer, a second fully connected layer and a softmax function layer; the four convolution blocks and the drop layer in the sub-network and the main network are connected in parallel; the sub-network includes two convolution blocks, a fully connected layer and an activation layer, the two convolution blocks are sequentially denoted as the fifth convolution block and the sixth convolution block, the sub-network includes a fully connected layer denoted as the third fully connected layer, and an activation layer is connected to the back of the third fully connected layer.

2. A production line real-time data analysis and energy-saving optimization method according to claim 1, characterized in that: The first machine learning model training process comprises the following steps: Step 1: Get Printed pattern, training set and validation set Two subsets, each containing multiple printed patterns; Let the first machine learning model be , the first machine learning model is initialized, and the initialization parameters satisfy the Gaussian distribution. During the training process, each time from the training set Randomly select The printed pattern was fed into the first machine learning model ,set up The printed patterns are , The corresponding product printing data are: , the first machine learning model The cross entropy loss function corresponding to each training process is ,but The calculation formula is as follows: ； In the formula, For the Printed pattern, For the The product printing data corresponding to the printed pattern, For the Prediction product printing data of printed patterns, is a hyperparameter, is the regularization of the weights of the first fully connected layer and the second fully connected layer; Step 2: From the training set Repeated random sampling The printed pattern was fed into the first machine learning model During the training process, the first machine learning model was trained using the Adam gradient descent optimization algorithm with a learning rate of 0.

002. When converged, the first machine learning model stops training; Step 3: After training, use the validation set The performance of the first machine learning model is evaluated on 3. A production line real-time data analysis and energy-saving optimization method according to claim 1, characterized in that: The third machine learning model building method includes: A conflict identification data set is collected in advance, wherein the conflict identification data includes identification feature data and conflict identification results corresponding to the identification feature data, wherein the identification feature data includes entity performance data and product printing data; the conflict identification data set is divided into a training set and a test set, and a regression network is constructed. The identification feature data in the training set is used as input data of the regression network, and the conflict identification results in the training set are used as output data of the regression network. The regression network is trained to obtain an initial regression network for predicting conflict identification results. The initial regression network is tested using the test set, and a regression network that satisfies a preset error value is output as a third machine learning model.

4. A production line real-time data analysis and energy-saving optimization method according to claim 1, characterized in that: The construction method of the second machine learning model is: The second machine learning model uses a random forest regression model to collect M historical ontology data, which includes ontology performance data and ontology production data corresponding to the ontology performance data; N tree subsets are sampled from the M historical ontology data, each tree subset contains a preset proportion of historical ontology data, and the N tree subsets will be used to train each decision tree: The single decision tree construction method includes: Initialize model parameters: determine the number of trees in the forest N_trees; the maximum depth of each tree max_depth; the minimum number of samples for internal nodes min_samples_split; the minimum number of samples for leaf nodes min_samples_leaf; Use the CART algorithm to recursively build the first decision tree: Divide the data in the tree subset into a training set and validation set , For the training set Middle A specific historical ontology data, using features Indicates one of high barrier properties, heat resistance, low temperature resistance, environmental protection, oil resistance, antistatic properties, tear resistance and light weight. Features Information gain rate : For each feature The corresponding information gain rate Calculate and select the maximum information gain rate The corresponding features As a node, and according to the maximum information gain rate The corresponding features The training set Divide into decision subsets so that the number of historical ontology data in each decision subset is the same; According to the maximum information gain rate Divide the decision subsets and repeat the algorithm recursive process until the decision subsets cannot be divided into smaller decision subsets; the trained decision tree model is tested on the validation set. The performance evaluation is carried out on the network, and finally the construction of the first decision tree model is completed.

5. A production line real-time data analysis and energy-saving optimization method according to claim 4, characterized in that: Random forest construction methods include: Collect M pieces of historical ontology data, which include ontology performance data and ontology production data corresponding to the ontology performance data; sample N tree subsets from the M pieces of historical ontology data, and the N tree subsets will be used to train each decision tree; Use the other N-1 tree subsets in turn to train N-1 decision trees; the generation process of each decision tree is independent and based on the CART algorithm; finally, a random forest of N decision trees is constructed.

6. A production line real-time data analysis and energy-saving optimization method according to claim 5, characterized in that: The second machine learning model training process: Calculate the loss function value on all N decision trees; input the training data into each decision tree and record the decision output, which is the ontology production data corresponding to the ontology performance data; calculate the mean square loss of the error between the average output of the N decision trees and the true value; reduce the mean square loss through parameter tuning, and select the model parameters with the smallest mean square loss as the model parameters of the second machine learning model.

7. A production line real-time data analysis and energy-saving optimization method according to claim 1, characterized in that: The fourth machine learning model training method comprises: Preliminarily collecting a correction data set, wherein the correction data includes conflict characteristic data and corrected production data corresponding to the conflict characteristic data; The corrected data set is divided into a training set and a test set, and a regression network is constructed. The conflict feature data in the training set is used as the input data of the regression network, and the corrected production data in the training set is used as the output data of the regression network. The regression network is trained to obtain an initial regression network for generating corrected production data. The initial regression network is tested with the test set, and the output regression network that satisfies a preset error value is used as the fourth machine learning model.

8. A production line real-time data analysis and energy-saving optimization system, characterized in that: Implementing a production line real-time data analysis and energy-saving optimization method as described in any one of claims 1 to 7, the system comprises: Collection module, used to obtain all product orders from the database; The first analysis module is used to analyze all product orders to check whether there is a printing pattern requirement in the order. If there is a printing pattern requirement in the product order, the printing pattern is sorted; The second analysis module inputs the printing pattern into the pre-built first machine learning model to output product printing data; the first machine learning model is a convolutional neural network model; the product printing data includes printing ink type, ink ratio parameters, printing effect, and printing control parameters; if there is no printing pattern demand, the body performance data in the product order is input into the pre-built second machine learning model to output the body production data, which includes the body raw material ratio, mixing parameters, and molding parameters; the body performance data includes high barrier properties, heat resistance, low temperature resistance, environmental protection, oil resistance, antistatic properties, tear resistance, and lightweight; the second machine learning model is a random forest regression model; A third analysis module inputs the product printing data and the ontology performance data in the product order into a pre-built third machine learning model to output a conflict identification result, wherein the conflict identification result includes no conflict and conflict; the third machine learning model is a regression network model; The fourth analysis module inputs the conflicting product printing data and entity performance data into the pre-trained fourth machine learning model to output corrected production data, wherein the corrected production data includes corrected product printing data and corrected entity production data; the fourth machine learning model is a regression network model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the production line real-time data analysis and energy-saving optimization method described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, the production line real-time data analysis and energy-saving optimization method described in any one of claims 1 to 7 is implemented.

Images