A method and system for one-step folding of flexible packaging sides using reinforcement learning

Abstract

This invention discloses a method and system for one-step folding and forming of flexible packaging sides using reinforcement learning, relating to the field of packaging forming technology. The method includes: acquiring multi-layer composite material information and interlayer adhesive structure information of the target flexible packaging material; using reinforcement learning to learn the transition folding angle based on a preset interlayer alignment index, generating transition folding angle parameters; after placing the target flexible packaging material into the forming unit, performing one-step folding and forming control, and activating a status sensor for real-time monitoring to generate real-time folding status data; and learning adjustable parameters for the folding process to generate updated transition folding angle parameters. This invention solves the technical problem of low interlayer alignment accuracy in the existing flexible packaging side folding and forming process, leading to low flexible packaging quality, and achieves high-precision one-step folding and forming of flexible packaging sides, thus improving the technical effect of flexible packaging quality.

Description

Technical Field

[0001] This invention relates to the field of packaging forming technology, and more specifically to a method and system for one-time folding forming of flexible packaging sides using reinforcement learning. Background Technology

[0002] In the field of flexible packaging production, the quality of the side folding process is crucial. Current technologies for side folding of flexible packaging suffer from low layer alignment accuracy and the inability to fold in one step, resulting in large residual creases with a high rebound angle, layer misalignment, and a high probability of peeling, thus affecting the quality and usability of the flexible packaging.

[0003] Existing technologies suffer from low layer alignment accuracy during side-folding forming of flexible packaging, resulting in poor packaging quality. Summary of the Invention

[0004] This application provides a method and system for one-time folding of flexible packaging sides using reinforcement learning, which addresses the technical problem of low interlayer alignment accuracy in the existing flexible packaging side folding process, resulting in low quality of flexible packaging.

[0005] In view of the above problems, this application provides a method and system for one-time folding of flexible packaging side using reinforcement learning.

[0006] The first aspect of this application provides a method for one-time folding of the side of flexible packaging using reinforcement learning, the method comprising:

[0007] Information on the multilayer composite material and interlayer adhesive structure of the target flexible packaging material is collected. Based on the multilayer composite material and interlayer adhesive structure information, reinforcement learning is used to learn the transition folding angle with a preset interlayer alignment index, generating transition folding angle parameters. After the target flexible packaging material is placed into the forming unit, the folding device is controlled to perform a single folding forming control using the transition folding angle parameters, and a status sensor is activated for real-time monitoring, generating real-time folding status data. Adjustable parameters for the folding process are learned using the real-time folding status data, and the transition folding angle parameters are adjusted to generate updated transition folding angle parameters.

[0008] A second aspect of this application provides a one-step folding system for flexible packaging sides using reinforcement learning, the system comprising:

[0009] The system includes a material information acquisition module for acquiring information on the multi-layer composite material and interlayer adhesive structure of the target flexible packaging material; a transition folding angle learning module for learning the transition folding angle based on the multi-layer composite material and interlayer adhesive structure information using reinforcement learning with a preset interlayer alignment index, generating transition folding angle parameters; a real-time monitoring module for controlling the folding device to perform a single folding forming control using the transition folding angle parameters after the target flexible packaging material is placed into the forming unit, and activating a status sensor for real-time monitoring, generating real-time folding status data; and a parameter adjustment module for learning adjustable parameters of the folding process using the real-time folding status data, adjusting the transition folding angle parameters, and generating updated transition folding angle parameters.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] Information on the multi-layer composite material and interlayer adhesive structure of the target flexible packaging material is collected. Reinforcement learning is used to learn the transition folding angle based on a preset interlayer alignment index, generating transition folding angle parameters. After the target flexible packaging material is placed into the forming unit, a single folding forming control is performed, and a status sensor is activated for real-time monitoring, generating real-time folding status data. Adjustable parameters for the folding process are learned using the real-time folding status data, and the transition folding angle parameters are adjusted to generate updated transition folding angle parameters. This achieves high-precision single-step folding forming of the flexible packaging side, improving the technical effect of flexible packaging quality. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A schematic flowchart of a method for one-time folding and forming of the side of flexible packaging using reinforcement learning, provided in an embodiment of this application;

[0014] Figure 2 This is a schematic diagram of a one-time folding system for flexible packaging sides using reinforcement learning, provided in an embodiment of this application.

[0015] Figure labeling: Material information acquisition module 10, transition rail angle learning module 20, real-time monitoring module 30, parameter adjustment module 40. Detailed Implementation

[0016] This application provides a method and system for one-time folding of flexible packaging sides using reinforcement learning, which addresses the technical problem of low interlayer alignment accuracy in the existing flexible packaging side folding process, resulting in low quality of flexible packaging.

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] Example 1, as Figure 1 As shown, this application provides a method for one-time folding of the side of flexible packaging using reinforcement learning, the method comprising:

[0019] Step S100: Collect information on the multilayer composite material and interlayer bonding structure of the target flexible packaging material.

[0020] Specifically, sensor devices are used to comprehensively collect information on the multi-layer composite materials of the target flexible packaging material, covering the material type, thickness specifications, tensile strength, elastic modulus, and other physical and mechanical properties of each layer, as well as interlayer bonding structure information, including the material composition of the adhesive layer, bonding process (such as hot pressing or adhesive bonding), bonding strength index, and microstructure distribution of the bonding interface. This information will serve as the basic input for subsequent reinforcement learning, used to construct the mapping relationship between material properties and folding angles, ensuring that the generation of subsequent transition folding angle parameters accurately adapts to the actual characteristics of the material.

[0021] Step S200: Based on the information of the multilayer composite material and the interlayer bonding structure, use reinforcement learning to learn the transition folding angle with a preset interlayer alignment index to generate transition folding angle parameters.

[0022] Specifically, the target folding angle is first determined and a transition space for the folding angle is constructed. Using multilayer composite material information and interlayer bonding structure information as retrieval factors, folding adaptation evaluation samples are collected according to preset interlayer alignment indicators (including at least the residual crease springback angle, interlayer misalignment, and interlayer delamination probability). Reinforcement learning is then used to train the interlayer folding adaptation analysis model. When this model is invoked, the empirical value of the transition folding angle (the angle most frequently used in historical records) is used as the starting point. Through transition smoothing optimization, the residual crease springback angle, interlayer misalignment, and interlayer delamination probability are iteratively decreased. The stopping condition is that the preset minimum transition unit or the fitness index set obtained from the iterations meets the interlayer alignment indicator threshold. This generates a termination fitness index set, termination transition folding angle parameters, and termination condition identifier. If the termination condition is identified as the fitness index set meeting the threshold, then the transition angle parameter is generated using the termination transition angle parameter; if it is a preset minimum transition unit, then the termination fitness index set is compared with the threshold, the index items that meet and do not meet the threshold are extracted, the preheating influence relationship corresponding to each index under the termination transition angle parameter is constructed based on the material information, the preheating parameter is generated through collaborative analysis, and then the transition angle parameter is generated by combining the termination transition angle parameter.

[0023] Step S300: After the target flexible packaging material is placed into the forming unit, the folding device is controlled to perform a folding forming control by the transition folding rail angle parameter, and the status sensor is activated for real-time monitoring to generate real-time folding status data.

[0024] Specifically, after the target flexible packaging material is placed into the forming unit, the folding device is precisely controlled based on the generated transition folding angle parameters to perform a single folding operation on the side of the flexible packaging. During this process, status sensors such as thickness sensors, infrared thermometers, displacement encoders, and stress sensors are simultaneously activated to collect data on local thickness, temperature, displacement, and tension fluctuations in the folding area in real time. This generates real-time folding status data containing multi-dimensional physical quantities, providing real-time dynamic monitoring information for subsequent parameter adjustments in the folding process, ensuring precise control and status feedback during the folding process.

[0025] Step S400: Learn adjustable parameters of the folding process using the real-time folding status data, adjust the transition folding angle parameters, and generate updated transition folding angle parameters.

[0026] Specifically, the transition folding angle parameter is first used as a benchmark, and its offset is compared with real-time folding state data (including local thickness, temperature, displacement, and tension fluctuation data of the folding area) to accurately locate the real-time state offset vector. Then, the folding forming knowledge base is connected. This knowledge base stores folding forming samples based on a ternary structure of folding state-correction action-correction reward, where the correction action is the parameter correction behavior for the folding state, and the correction reward is the corresponding correction effect. Using the sample data in the knowledge base, the real-time state offset vector is corrected. By analyzing the causes of the offset and matching corresponding correction actions, the transition folding angle parameter is dynamically adjusted, ultimately generating updated transition folding angle parameters, achieving adaptive optimization of the folding process.

[0027] In one possible implementation, step S200 further includes:

[0028] The preset interlayer alignment indicators include at least the crease residual springback angle, interlayer misalignment amount, and interlayer peeling probability.

[0029] Specifically, the preset interlayer alignment index covers at least the residual crease springback angle, interlayer misalignment, and interlayer peeling probability. These three indexes will be evaluated independently in the subsequent fitness analysis. The residual crease springback angle measures the angular deviation caused by material springback at the crease after folding; the interlayer misalignment characterizes the relative displacement between layers during the folding process of multi-layer materials; and the interlayer peeling probability assesses the likelihood of interlayer adhesive failure and separation caused by the folding operation. By quantifying these three indexes separately, the impact of the transition folding angle parameter on the folding quality of flexible packaging materials can be comprehensively and independently evaluated.

[0030] In one possible implementation, step S200 further includes:

[0031] Step S210: Determine the target folding angle and construct the transition space for the folding angle.

[0032] Step S220: Using the information of the multilayer composite material and the interlayer bonding structure as retrieval factors, collect track-folding adaptation evaluation samples based on the preset interlayer alignment index, and train the interlayer track-folding adaptation analysis model using reinforcement learning.

[0033] Step S230: Invoke the inter-layer track bending adaptation analysis model, filter the transition track bending angles that meet the inter-layer alignment index threshold based on the track bending angle transition space, and generate the transition track bending angle parameters.

[0034] Specifically, based on the design standards and actual process requirements for side folding of flexible packaging, a target folding angle is defined. This angle represents the ideal angle value that the flexible packaging side needs to achieve after folding. Using the target folding angle as the center, and combining material characteristics and production experience, a reasonable angle fluctuation range is set, thereby constructing a transition space for the folding angle. This provides a specific parameter search range for subsequent use of reinforcement learning algorithms to search and optimize the transition folding angle, ensuring that adjustments to the folding angle can be made within the range that meets the forming requirements.

[0035] The collected information on the multilayer composite materials of the target flexible packaging material (including physical and mechanical properties such as material, thickness, and tensile strength of each layer) and interlayer bonding structure information (such as adhesive layer material composition, bonding process, and bonding strength) are used as retrieval factors. Based on preset interlayer alignment indicators (crease residual springback angle, interlayer misalignment, and interlayer peeling probability), folding adaptation evaluation samples matching the current material characteristics are retrieved and collected from historical production data. These samples contain actual test data of each indicator under different folding angles. The samples are iteratively trained using a reinforcement learning algorithm to construct an interlayer folding adaptation analysis model. This model can learn the mapping relationship between material properties and folding angles, thereby enabling it to predict the optimal folding angle parameters based on the input material information.

[0036] First, the most frequently used transition folding angle in historical records of single-folding of flexible packaging sides is collected as an empirical value. Starting from this empirical value, the interlayer folding adaptation analysis model is invoked. Within the folding angle transition space, iterative optimization is performed to decrease the residual crease springback angle, interlayer misalignment, and interlayer peeling probability. During the iteration process, the angle parameters are adjusted through a transition smoothing optimization algorithm. The iteration stops when the fitness index set obtained by the iteration meets the preset interlayer alignment index threshold or reaches the preset minimum transition unit, generating a termination fitness index set, termination transition folding angle parameters, and termination condition identifier. If the termination condition is identified as the fitness index set meeting a threshold, the target transition angle parameter is directly generated using the termination transition angle parameter. If the termination condition is identified as a preset minimum transition unit, the termination fitness index set is compared with the interlayer alignment index threshold, and the satisfied and unsatisfied index items are extracted. Based on the multilayer composite material information and interlayer bonding structure information, the preheating influence relationship corresponding to each index under the termination transition angle parameter is constructed. The first preheating temperature range of the unsatisfied index items is determined through collaborative analysis, and the temperature value that makes the index influence meet the preset threshold is selected as the preheating parameter in combination with the preheating influence relationship of the satisfied index items. Finally, the transition angle parameter is generated using the preheating parameter and the termination transition angle parameter.

[0037] In one possible implementation, step S230 further includes:

[0038] Step S231: Collect empirical values ​​of transition folding angles. The empirical values ​​of transition folding angles are the transition folding angles that are most frequently used in the historical records of one-time folding of the side of flexible packaging.

[0039] Step S232: Invoke the interlayer folding adaptation analysis model, starting from the empirical value of the transition folding angle, and perform a decreasing iteration of the residual crease springback angle, interlayer misalignment amount and interlayer peeling probability through transition smoothing optimization. The iteration stops when the preset minimum transition unit or the fitness index set obtained by the iteration satisfies the interlayer alignment index threshold. Generate the termination fitness index set, termination transition folding angle parameter and termination condition identifier when the iteration stops.

[0040] Step S233: If the termination condition is identified as the fitness index set obtained by iteration satisfies the inter-layer alignment index threshold, the transition angle parameter is generated using the termination transition angle parameter.

[0041] Step S234: If the termination condition is identified as a preset minimum transition unit, perform preheating parameter analysis during folding based on the termination fitness index set and the termination transition folding angle parameter, and generate the transition folding angle parameter by combining the preheating parameter.

[0042] Specifically, by statistically analyzing historical production records of flexible packaging side folding, the most frequently used transition folding angle in past production practices was extracted and determined as the empirical value for the transition folding angle. This empirical value is a representative angle parameter selected from a large number of historical forming cases, reflecting the commonly used angle settings when folding similar flexible packaging materials. It can provide an initial reference benchmark for subsequent optimization of the transition folding angle, reduce the search space of algorithm iteration, and improve optimization efficiency.

[0043] The trained interlayer folding adaptation analysis model is invoked, using the empirical value of the transition folding angle as the starting point for iterative optimization. An iterative process is executed within the folding angle transition space using a transition smoothing optimization algorithm. During iteration, the residual crease springback angle, interlayer misalignment, and interlayer peeling probability are continuously optimized, causing each indicator value to gradually decrease. When the preset minimum transition unit (i.e., the minimum step size for angle adjustment) is reached, or the fitness indicator set obtained through iteration meets the preset interlayer alignment indicator threshold, the iteration stops. The process generates the termination fitness indicator set, the termination transition folding angle parameter, and a termination condition identifier to indicate the reason for iteration termination.

[0044] When the termination condition indicates that the fitness index set obtained through iteration has met the preset interlayer alignment index threshold (i.e., the residual crease springback angle, interlayer misalignment amount, and interlayer peeling probability all meet the process requirements), it means that the termination transition folding angle parameter obtained through iterative optimization can ensure the quality of the side folding of the flexible packaging. At this time, the termination transition folding angle parameter is directly output as the final transition folding angle parameter to control the subsequent folding device to perform the folding operation.

[0045] If the termination condition is marked as a preset minimum transition unit, it indicates that the iterative optimization has reached the minimum step size for angle adjustment but still has not made the fitness index set meet the interlayer alignment index threshold. At this point, a preheating parameter analysis is performed based on the termination fitness index set and the termination transition folding angle parameter: First, the termination fitness index set is compared with the index threshold to distinguish between satisfied and unsatisfied index items. Then, based on the multilayer composite material information and interlayer bonding structure information, a preheating influence relationship model is constructed for the crease residual springback angle, interlayer misalignment, and interlayer peeling probability under this termination angle parameter. This model is used to collaboratively analyze satisfied and unsatisfied index items. For unsatisfied index items, the corresponding first preheating temperature range is determined. Simultaneously, combined with the preheating influence relationship of satisfied index items, temperature values ​​within this temperature range that allow the index influence change to meet the preset threshold are selected, generating preheating parameters. Finally, the preheating parameters are combined with the termination transition folding angle parameter to generate a transition folding angle parameter that comprehensively considers the material preheating effect, compensating for index deviations caused by limited angle iteration and ensuring the folding forming quality.

[0046] In one possible implementation, step S234 further includes:

[0047] Step S2341: Compare the termination fitness index set with the inter-layer alignment index threshold, and extract the index items that meet the criteria and the index items that do not meet the criteria.

[0048] Step S2342: Based on the information of the multilayer composite material and the interlayer bonding structure, construct the preheating influence relationships corresponding to the residual springback angle of the crease, the amount of interlayer misalignment, and the probability of interlayer peeling under the termination transition folding angle parameter.

[0049] Step S2343: Based on the preheating influence relationship, perform a collaborative analysis of the satisfying and non-satisfied index items to generate the preheating parameters.

[0050] Step S2344: Generate the transition angle parameters using the preheating parameters and the termination transition angle parameters.

[0051] Specifically, the termination fitness index set obtained when the iteration stops (including specific values ​​of crease residual springback angle, interlayer misalignment, and interlayer peeling probability) is compared one by one with the preset interlayer alignment index threshold. By comparing the values, the index items that meet or exceed the threshold requirements are accurately identified, as well as the index items that do not meet the threshold requirements. This clarifies the index directions that need to be optimized in the future, providing a precise index selection basis for subsequent construction of preheating influence relationships and collaborative analysis based on material properties.

[0052] Based on the collected information on multilayer composite materials (including thermophysical properties such as the coefficient of thermal expansion, elastic modulus, and thickness of each layer) and interlayer bonding structure information (including the heat resistance temperature range of the bonding layer material and the temperature-dependent bonding strength curve), this study constructs the corresponding influence relationships between the residual crease springback angle, interlayer misalignment, and interlayer delamination probability and the preheating temperature, focusing on the termination transition folding angle parameter. By analyzing the changes in the mechanical properties of the material under different preheating temperatures, a mathematical model is established to quantify the influence of preheating temperature on each indicator. For example, it determines the change in the residual crease springback angle, the fluctuation range of interlayer misalignment, and the decrease in the interlayer delamination probability for every 1°C increase in preheating temperature, providing a quantitative basis for subsequent collaborative analysis of preheating parameters.

[0053] Based on the established preheating influence relationships, a collaborative analysis is conducted on both satisfactory and unsatisfactory indicators. First, based on the preheating influence relationships corresponding to unsatisfactory indicators (such as interlayer delamination probability exceeding a threshold), a first preheating temperature range is determined that can optimize the indicator value to within the threshold. For example, by analyzing the negative correlation between preheating temperature and interlayer delamination probability, it is concluded that the temperature needs to be increased to 60-70℃ to reduce the delamination risk. Simultaneously, combined with the preheating influence relationships of satisfactory indicators (such as crease residual springback angle meeting the standard), the changes in the indicators of satisfactory indicators within this temperature range are evaluated. For example, it is found that the crease springback angle increases by 0.5° at 60-70℃ due to material softening; this increment must be ensured not to exceed the preset threshold of 0.8°. Finally, within the temperature range, a temperature value (such as 65℃) is selected that can optimize unsatisfactory indicators to acceptable levels while keeping the changes in satisfactory indicators within the allowable range. Preheating parameters containing specific temperature parameters are generated to achieve a collaborative balance in indicator optimization.

[0054] By establishing a mapping model between preheating parameters and changes in material mechanical properties, parameters such as preheating temperature and preheating time are transformed into changes in parameters such as the material's elastic modulus and yield strength. Then, based on the relationship between changes in material mechanical properties and the transition angle, the compensation value for the termination transition angle parameter is calculated. For example, when the preheating temperature is 65℃, the material's elastic modulus decreases by 10%. Calculations show that the termination transition angle parameter needs to be adjusted from 30° to 30.5°, resulting in a final transition angle parameter of 30.5°. This achieves the fusion of preheating parameters and the termination transition angle parameter, generating a transition angle parameter adapted to the material's preheating state.

[0055] In one possible implementation, step S2343 further includes:

[0056] Step S23431: Based on the preheating influence relationship corresponding to the residual crease rebound angle, interlayer misalignment, and interlayer peeling probability, perform index optimization for the unsatisfactory index items and determine the first preheating temperature range for the unsatisfactory index items.

[0057] Step S23432: Combining the preheating influence relationship of satisfying the index items, determine the index influence update of satisfying the index items under the first preheating temperature range, screen the temperature values ​​where the index influence meets the preset threshold, and generate the preheating parameters.

[0058] Specifically, a mathematical model is established to quantitatively analyze the relationship between the residual crease springback angle, interlayer misalignment, and interlayer delamination probability and the preheating temperature. Historical data is used to fit the curve equations of each index changing with temperature. For index items that do not meet the requirements, such as the interlayer delamination probability exceeding the threshold, the corresponding preheating influence equation is solved simultaneously with the index threshold to calculate the lower and upper bounds of the temperature range that allows the index to meet the requirements. At the same time, considering the material's heat resistance limit and process feasibility, temperature regions that lead to material degradation or excessive energy consumption are excluded, and the first preheating temperature range that does not meet the index items is finally determined. For example, calculations show that when the preheating temperature is between 65℃ and 75℃, the interlayer delamination probability can be reduced from 0.25 to below 0.1, and the material's thermal deformation is within the allowable range.

[0059] After obtaining the first preheating temperature range that does not meet the target criteria, the preheating influence relationship data corresponding to the target criteria that do meet the criteria (such as the function curve of the residual crease rebound angle as a function of temperature) is retrieved, and this temperature range is substituted into the relationship model to calculate the change in the target criteria. For example, if the first preheating temperature range is 60-70℃, the model calculation shows that the residual crease rebound angle increases with temperature in this range, with an increase of 0.05° for every 1℃ increase. The preset threshold is that the angle increment does not exceed 0.5°, so the upper limit of the temperature can be determined as 60 + (0.5 / 0.05) = 70℃. At the same time, the stability of the target criteria at the temperature boundary points is verified by combining historical molding data, and temperature values ​​such as 65℃, which make the change in the target criteria (0.25°) much smaller than the threshold and do not meet the target criteria optimization target, are selected. Finally, a preheating scheme containing accurate temperature parameters is generated.

[0060] In one possible implementation, step S300 further includes:

[0061] Step S310: The status sensors include a thickness sensor, an infrared thermometer, a displacement encoder, and a stress sensor; the real-time folding status data includes local thickness, temperature, displacement, and tension fluctuation data of the folding area.

[0062] Specifically, the status sensors include a thickness sensor, an infrared thermometer, a displacement encoder, and a stress sensor. The thickness sensor measures the local thickness changes in the folding area in real time; the infrared thermometer monitors the temperature distribution in the folding area; the displacement encoder records the displacement during the folding process; and the stress sensor detects tension fluctuations in the folding area. These sensors work together to generate real-time folding status data covering the local thickness, temperature, displacement, and tension fluctuations in the folding area. This provides comprehensive and accurate data support for learning adjustable parameters in the folding process and adjusting the transition folding angle parameters, ensuring real-time monitoring and precise control of the folding status.

[0063] In one possible implementation, step S400 further includes:

[0064] Step S410: Use the transition folding angle parameter to perform offset comparison on the real-time folding state data to locate the real-time state offset vector.

[0065] Step S420: Connect to the folding and forming knowledge base, perform the correction of the real-time state offset vector, and generate the updated transition folding angle parameters.

[0066] Specifically, the transition folding angle parameter is used as a standard reference value. It is compared and analyzed with the folding state data such as local thickness, temperature, displacement and tension fluctuation of the folding area collected in real time by the state sensor. The deviation between the actual data and the reference value is calculated, thereby locating the real-time state offset vector that can reflect the degree and direction of the current folding state deviating from the expectation, providing a precise deviation basis for subsequent parameter adjustment of the folding process.

[0067] The system connects to a folding and forming knowledge base via an API interface. This knowledge base uses a triplet structure (folding state, correction action, correction reward) to store historical forming samples. The folding state includes multi-dimensional feature vectors such as local thickness deviation and abnormal temperature range. After obtaining the real-time state offset vector, a cosine similarity algorithm is used to retrieve the historical folding state with the highest matching degree from the knowledge base. For example, if the current offset vector is thickness +0.2mm, temperature -5℃, and displacement +1.5mm, then a record with similar offset features in the historical samples will be matched. The corresponding correction actions (such as increasing the folding angle by 0.5° or raising the preheating temperature by 10°C) are extracted, and the optimal correction strategy is determined by combining the Q-value evaluation mechanism of reinforcement learning with correction rewards (such as increasing the target achievement rate by 20%). Finally, the real-time offset vector and the correction actions are linearly superimposed to calculate the optimal correction strategy. The real-time state offset vector (such as the deviation of local thickness, temperature, displacement, and tension fluctuations in the folding area) is calculated with the corresponding correction actions (such as adjusting the folding angle or changing the preheating temperature) retrieved from the folding forming knowledge base using a linear superposition mathematical method. For example, if the real-time state offset vector shows that the temperature in the folding area is 5°C lower than expected, the corresponding correction action is to increase the preheating temperature by 10°C and increase the folding angle by 0.5°. These offsets and corrections are then superimposed according to a linear relationship to generate updated transition folding angle parameters, thereby achieving precise control of the folding process.

[0068] In one possible implementation, step S420 further includes:

[0069] Step S421: The folding forming knowledge base stores folding forming samples based on a triplet structure, wherein the triplet structure includes folding state, correction action, and correction reward.

[0070] Step S422: The correction action is a parameter correction behavior performed on the folded edge state, and the correction reward is the corresponding correction effect.

[0071] Specifically, the folding forming knowledge base stores folding forming samples organized in a ternary structure. The folding state is a comprehensive representation of real-time data such as local thickness, temperature, displacement, and tension fluctuations in the folding area during the folding process, forming a multi-dimensional feature vector. The correction action refers to the parameter adjustment measures taken to address specific folding state deviations, such as modifying the transition folding angle or adjusting the preheating temperature. The correction reward is an indicator that quantifies the effectiveness of the correction action, such as the reduction ratio of interlayer misalignment and the reduction value of the residual springback angle of the fold. This structured storage method provides data support for the real-time folding state offset correction.

[0072] Corrective actions refer to parameter adjustments taken when the folding condition (such as abnormal local thickness in the folded area, temperature fluctuations, displacement deviations, or tension imbalances) deviates from expectations. Examples include increasing the transition folding angle by 0.5°, raising the preheating temperature by 10°C, or adjusting tension control parameters. Corrective rewards are a quantitative assessment of the effectiveness of corrective actions. They are measured by calculating indicators such as the reduction in interlayer misalignment, the decrease in the residual springback angle of the fold, or the decrease in the probability of interlayer peeling. For example, if a correction reduces interlayer misalignment from 0.3mm to 0.1mm, this difference is the corresponding corrective reward.

[0073] Example 2, based on the same inventive concept as the one-time folding method for flexible packaging using reinforcement learning in the previous examples, such as... Figure 2 As shown, this application provides a one-step folding system for flexible packaging sides using reinforcement learning. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0074] The material information acquisition module 10 is used to acquire information on the multilayer composite material and interlayer bonding structure of the target flexible packaging material.

[0075] The transition folding angle learning module 20 is used to learn the transition folding angle based on the multilayer composite material information and the interlayer bonding structure information, using reinforcement learning with a preset interlayer alignment index, and to generate transition folding angle parameters.

[0076] The real-time monitoring module 30 is used to control the folding device to perform a folding forming control with the transition folding rail angle parameter after the target soft packaging material is placed into the forming unit, and to activate the status sensor for real-time monitoring and generate real-time folding status data.

[0077] The parameter adjustment module 40 is used to learn adjustable parameters of the folding process based on the real-time folding status data, adjust the transition folding angle parameters, and generate updated transition folding angle parameters.

[0078] Furthermore, the system is also used to implement the following functions:

[0079] The preset interlayer alignment indicators include at least the crease residual springback angle, interlayer misalignment amount, and interlayer peeling probability.

[0080] Furthermore, the system is also used to implement the following functions:

[0081] Determine the target folding angle and construct a folding angle transition space; using the multilayer composite material information and the interlayer bonding structure information as retrieval factors, collect folding adaptation evaluation samples based on the preset interlayer alignment index, and train the interlayer folding adaptation analysis model using reinforcement learning; call the interlayer folding adaptation analysis model, and filter the transition folding angles that meet the interlayer alignment index threshold based on the folding angle transition space to generate the transition folding angle parameters.

[0082] Furthermore, the system is also used to implement the following functions:

[0083] Empirical values ​​for transition folding angles are collected, which are the most frequently used transition folding angles in historical records of single-fold forming of flexible packaging sides. The interlayer folding adaptation analysis model is invoked, starting with these empirical values. Through transition smoothing optimization, iterative decreases are performed on the residual crease springback angle, interlayer misalignment, and interlayer peeling probability. The iteration stops when a preset minimum transition unit or the fitness index set obtained through iteration satisfies the interlayer alignment index threshold. A termination fitness index set, a termination transition folding angle parameter, and a termination condition identifier are generated when the iteration stops. If the termination condition identifier indicates that the fitness index set obtained through iteration satisfies the interlayer alignment index threshold, the transition folding angle parameter is generated using the termination transition folding angle parameter. If the termination condition identifier indicates a preset minimum transition unit, preheating parameter analysis during folding is performed based on the termination fitness index set and the termination transition folding angle parameter, and the transition folding angle parameter is generated by combining the preheating parameters.

[0084] Furthermore, the system is also used to implement the following functions:

[0085] The termination fitness index set is compared with the interlayer alignment index threshold to extract the index items that meet the criteria and those that do not. Based on the multilayer composite material information and the interlayer bonding structure information, the preheating influence relationships corresponding to the residual springback angle of the crease, the amount of interlayer misalignment, and the probability of interlayer delamination under the termination transition folding angle parameter are constructed respectively. Based on the preheating influence relationship, a collaborative analysis of the index items that meet the criteria and those that do not meet the criteria is performed to generate the preheating parameters. The transition folding angle parameters are generated using the preheating parameters and the termination transition folding angle parameters.

[0086] Furthermore, the system is also used to implement the following functions:

[0087] Based on the preheating influence relationships corresponding to the residual crease rebound angle, interlayer misalignment, and interlayer delamination probability, the index optimization for unsatisfactory index items is performed to determine the first preheating temperature range for unsatisfactory index items. Combining the preheating influence relationships for satisfactory index items, the index influence updates for satisfactory index items within the first preheating temperature range are determined. Temperature values ​​whose index influences meet preset thresholds are selected to generate the preheating parameters.

[0088] Furthermore, the system is also used to implement the following functions:

[0089] The status sensors include a thickness sensor, an infrared thermometer, a displacement encoder, and a stress sensor; the real-time folding status data includes local thickness, temperature, displacement, and tension fluctuation data of the folding area.

[0090] Furthermore, the system is also used to implement the following functions:

[0091] The real-time folding state data is offset by the transition folding angle parameter to locate the real-time state offset vector; the folding forming knowledge base is connected to perform the correction of the real-time state offset vector to generate the updated transition folding angle parameter.

[0092] Furthermore, the system is also used to implement the following functions:

[0093] The folding forming knowledge base stores folding forming samples based on a triplet structure, where the triplet structure includes a folding state, a correction action, and a correction reward; the correction action is a parameter correction behavior performed on the folding state, and the correction reward is the corresponding correction effect.

[0094] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0095] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0096] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A soft-wrapped side-edge once-folded forming method using reinforcement learning, characterized in that, The method comprises the following steps: Collecting information of a multi-layer composite material of a target soft packaging material and information of an interlayer adhesion structure; Based on the multi-layer composite material information and the interlayer adhesion structure information, using reinforcement learning to learn a transition folding rail angle with preset interlayer alignment indicators, and generating a transition folding rail angle parameter, wherein the preset interlayer alignment indicators at least include a crease residual rebound angle, an interlayer misalignment amount, and an interlayer peeling probability; After the target soft packaging material is put into a forming unit, a first folding forming control is performed on a folding device with the transition folding rail angle parameter, and a state sensor is activated for real-time monitoring to generate real-time folding state data; Adjustable parameter learning of a folding process is performed on the real-time folding state data, the transition folding rail angle parameter is adjusted, and an updated transition folding rail angle parameter is generated; Based on the multi-layer composite material information and the interlayer adhesion structure information, using reinforcement learning to learn a transition folding rail angle with preset interlayer alignment indicators, and generating a transition folding rail angle parameter, comprises: Determining a target folding angle, and constructing a folding rail angle transition space; Using the multi-layer composite material information and the interlayer adhesion structure information as retrieval factors, collecting folding rail adaptation evaluation samples based on the preset interlayer alignment indicators, and using reinforcement learning to train an interlayer folding rail adaptation analysis model; Calling the interlayer folding rail adaptation analysis model, screening a transition folding rail angle that meets an interlayer alignment indicator threshold based on the folding rail angle transition space, and generating the transition folding rail angle parameter; Based on the folding rail angle transition space, screening a transition folding rail angle that meets an interlayer alignment indicator threshold, and generating the transition folding rail angle parameter, comprises: Collecting a transition folding rail angle experience value, which is the transition folding rail angle with the highest frequency of use in historical soft packaging side first folding forming records; Calling the interlayer folding rail adaptation analysis model, taking the transition folding rail angle experience value as a starting point, performing a descending iteration of a crease residual rebound angle, an interlayer misalignment amount, and an interlayer peeling probability through transition smoothing optimization, taking a preset minimum transition unit or an adaptability indicator set obtained through iteration meeting the interlayer alignment indicator threshold as an iteration stop condition, generating a terminal adaptability indicator set and a terminal transition folding rail angle parameter at the iteration stop condition, and a terminal condition identifier; If the terminal condition identifier is that the adaptability indicator set obtained through iteration meets the interlayer alignment indicator threshold, generating the transition folding rail angle parameter with the terminal transition folding rail angle parameter; If the terminal condition identifier is a preset minimum transition unit, performing a preheating parameter analysis of folding when based on the terminal adaptability indicator set and the terminal transition folding rail angle parameter, and generating the transition folding rail angle parameter in combination with the preheating parameter; Adjustable parameter learning of a folding process is performed on the real-time folding state data, the transition folding rail angle parameter is adjusted, and an updated transition folding rail angle parameter is generated, which comprises: Performing offset comparison on the real-time folding state data with the transition folding rail angle parameter, and positioning a real-time state offset vector; The connection flange forming knowledge base is formed, the correction of the real-time state offset vector is performed, and the updated transition flange rail angle parameter is generated.

2. The soft-pack side-edge once-fold forming method using reinforcement learning of claim 1, wherein, Based on the set of termination fitness indicators and the termination transition flange rail angle parameter, preheating parameter analysis during folding is performed, and the transition flange rail angle parameter is generated in combination with the preheating parameter, including: The set of termination fitness indicators is compared with the interlayer alignment indicator threshold, and the indicators that meet the indicators and do not meet the indicators are extracted; According to the multi-layer composite material information and the interlayer bonding structure information, the preheating influence relationship respectively corresponding to the fold residual rebound angle, the interlayer misregistration amount and the interlayer peeling probability under the termination transition flange rail angle parameter is constructed; Based on the preheating influence relationship, the collaborative analysis of the indicators that meet the indicators and do not meet the indicators is performed, and the preheating parameter is generated. The preheating parameter and the termination transition flange rail angle parameter are used to generate the transition flange rail angle parameter.

3. The soft-pack side-edge once-fold forming method using reinforcement learning according to claim 2, wherein, Based on the preheating influence relationship, the collaborative analysis of the indicators that meet the indicators and do not meet the indicators is performed, and the preheating parameter is generated, including: According to the preheating influence relationship respectively corresponding to the fold residual rebound angle, the interlayer misregistration amount and the interlayer peeling probability, the index optimization of the indicators that do not meet the indicators is performed, and the first preheating temperature range of the indicators that do not meet the indicators is determined; In combination with the preheating influence relationship of the indicators that meet the indicators, the index influence update of the indicators that meet the indicators under the first preheating temperature range is determined, the temperature value of the index influence meeting the preset threshold is screened, and the preheating parameter is generated.

4. The soft-pack side-edge once-fold forming method using reinforcement learning of claim 1, wherein, The state sensor includes a thickness sensor, an infrared temperature measuring instrument, a displacement encoder and a stress sensor; the real-time flange state data includes local thickness, temperature, displacement and tension fluctuation data of the flange area.

5. The soft-pack side-edge once-fold forming method using reinforcement learning of claim 1, wherein, The flange forming knowledge base stores flange forming samples based on a triple structure, wherein the triple structure includes a flange state, a correction action and a correction reward; The correction action is a parameter correction behavior for the flange state, and the correction reward is a corresponding correction effect.

6. A soft-wrapped side-edge once-fold forming system utilizing reinforcement learning, characterized in that, The system is used to implement the soft package side one-fold forming method using reinforcement learning according to any one of claims 1-5, and the system comprises: A material information acquisition module is configured to acquire multi-layer composite material information and interlayer bonding structure information of a target soft package material. A transition flange rail angle learning module is configured to learn a transition flange rail angle based on the multi-layer composite material information and the interlayer bonding structure information using reinforcement learning with a preset interlayer alignment indicator to generate a transition flange rail angle parameter. A real-time monitoring module is configured to control a flange device to perform one-fold forming control using the transition flange rail angle parameter after the target soft package material is placed in a forming unit, and activate a state sensor to perform real-time monitoring to generate real-time flange state data. A parameter adjustment module is configured to learn adjustable parameters of a flange process using the real-time flange state data, adjust the transition flange rail angle parameter, and generate an updated transition flange rail angle parameter.

Images