Intelligent wrapping film control system and device

Abstract

The application discloses an intelligent winding film wrapping control system and equipment, belongs to the technical field of logistics packaging, and aims to solve the problem that in a high-speed winding process, when a film passes through a fragile area such as a corner of goods, a large tension is easy to cause damage of the goods. Technical scheme points are as follows: including a data interface unit, a fragile degree analysis unit, a winding path planning unit and a dynamic tension regulation unit. The application combines a three-dimensional visual sensor with a neural network model, not only identifies a fragile area, but also gives the fragile area a dynamic fragile degree weight value, realizes quantitative evaluation of the fragility of goods, and lays a foundation for subsequent accurate control.

Description

Technical Field

[0001] This invention relates to the field of logistics packaging technology, and more specifically, to an intelligent wrapping control system and a wrapping device including the system, which is particularly suitable for adaptive flexible packaging of fragile items in high-speed assembly line scenarios. Background Technology

[0002] Stretch wrapping machines are common packaging equipment in modern logistics centers, used to wrap stretch film around palletized goods to secure them, prevent dust and moisture. With the rapid development of e-commerce and express logistics, the demand for packaging efficiency is increasing, making high-speed assembly line operations the mainstream.

[0003] However, when handling fragile items in high-speed assembly line scenarios, such as glassware, ceramics, electronic products, and irregularly shaped cakes, existing technologies have the following drawbacks: First, traditional wrapping machines use constant tension or simple mechanical tension control, which cannot adjust the tension in real time according to the fragility of different parts of the goods. During high-speed wrapping, when the film passes through vulnerable areas such as the edges and corners of the goods, the greater tension can easily cause damage.

[0004] Secondly, the packaging parameters of existing equipment, such as the number of wraps, tension, and film carriage lifting speed, are usually set manually based on experience, making it impossible to adaptively adjust for goods of different shapes and materials. In a multi-variety, small-batch production model, frequent manual adjustments lead to low efficiency and inconsistent packaging quality.

[0005] Third, the lack of a closed-loop feedback mechanism for packaging effectiveness makes it impossible to continuously optimize packaging parameters. Experience from each packaging session cannot be learned and reused by the equipment, resulting in film waste and fluctuations in packaging quality.

[0006] Therefore, a new solution is needed to address this problem. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent wrapping control system and equipment suitable for packaging fragile goods on high-speed production lines, so as to solve the technical problems in the prior art that cannot dynamically adjust the packaging tension according to the fragility of the goods, rely on human experience for packaging parameters, and lack self-optimization capabilities.

[0008] This invention achieves the above objective through the following technical solution: an intelligent wrapping control system, comprising: The data interface unit is configured to receive three-dimensional point cloud data from an external cargo sensing device, wherein the three-dimensional point cloud data includes the three-dimensional contour information and surface material characteristics of the cargo to be packaged. The vulnerability analysis unit is electrically connected to the data interface unit and is configured to analyze three-dimensional point cloud data through a neural network model, identify and mark the fragile feature areas on the surface of the cargo, and assign a dynamic vulnerability weight value to each fragile feature area to construct a three-dimensional cargo model with vulnerability weights. The winding path planning unit is electrically connected to the vulnerability analysis unit and is configured to generate an initial winding path containing a fast winding segment and a deceleration and avoidance segment based on the current tensile strength data of the film based on the three-dimensional cargo model and the pre-stored mechanical property database of film materials. The dynamic tension control unit is electrically connected to the winding path planning unit and is equipped with a tension signal input interface for connecting an external tension sensor. It is also configured to acquire real-time feedback data from the tension sensor of the external winding and packaging execution equipment during the winding operation, and generate a tension adjustment command through a model predictive control algorithm based on the fragility weight value in the three-dimensional cargo model and the feedback data. The tension adjustment command is sent to the servo drive tension control unit of the external winding and packaging execution equipment through a control signal output interface. The parameter self-optimization unit is electrically connected to the dynamic tension control unit, the fragility analysis unit, and the winding path planning unit, respectively. It is configured to record the tension curve, film consumption data, and packaging effect evaluation score provided by the packaging effect evaluation subunit for each packaging process, and feed them back to the fragility analysis unit and the winding path planning unit to periodically update the parameters of the neural network model and the reinforcement learning model.

[0009] In some embodiments, the vulnerability analysis unit includes: The convolutional neural network subunit is configured to extract geometric and spectral features from 3D point cloud data and identify fragile markings, angular areas, and suspended structures on the surface of goods based on a preset training sample library, so as to mark fragile feature areas.

[0010] In some embodiments, the winding path planning unit includes: The reinforcement learning subunit is configured to minimize packaging time and film consumption as optimization objectives. It generates the initial winding path through a deep deterministic policy gradient algorithm, combined with dynamic vulnerability weights.

[0011] In some embodiments, the dynamic tension control unit includes: The feedforward control subunit is configured to calculate the start time and adjustment range of the tension adjustment command in advance based on the distance and orientation of the next fragile feature area in the initial winding path; The feedback control subunit is configured to perform closed-loop correction on the tension adjustment command generated by the feedforward control subunit based on real-time feedback data from the tension sensor.

[0012] In some embodiments, the system further includes a corner protector installation control unit, configured to generate corner protector installation instructions based on the position of the corner areas in the three-dimensional cargo model, and send them to an external automatic corner protector installation device via a control signal output interface.

[0013] In some embodiments, the parameter self-optimization unit further includes: The packaging effect evaluation subunit is configured to receive images of the packaged goods from the downstream second vision sensor via the data interface unit, and analyze the film adhesion, damage, and looseness as packaging effect evaluation data.

[0014] In some embodiments: the thin film material mechanical property database contains stress-strain curve data of various thin films at different stretching rates; The dynamic tension control unit is also configured to dynamically match the elastic modulus of the current film from the stress-strain curve according to the stretching rate corresponding to the current packaging speed, and calculate the servo motor torque output value corresponding to the safety threshold range accordingly.

[0015] In some embodiments, a fault prediction and diagnosis unit is also included, configured to monitor the vibration, current and temperature data of the external servo-driven tension control unit, the rotating platform and the membrane breaking mechanism in real time through a signal input interface, and to predict potential faults through a time-series prediction model to generate preventive maintenance warnings.

[0016] In some embodiments, it also includes: The 5G communication unit is configured to upload the updated parameters generated by the parameter self-optimization unit to the cloud server in real time, and receive the optimized neural network model and reinforcement learning model parameters shared by other wrapping machines from the cloud server.

[0017] A smart wrapping device including the above-mentioned control system, characterized in that it further includes: The cargo sensing system, located upstream of the production line, includes a 3D vision sensor and is connected to the data interface unit of the control system. It is used to scan the cargo to be packaged and generate 3D point cloud data. An adaptive packaging execution system, located downstream of the production line, includes a rotating platform, a film holder with a servo-driven tension control unit, a film cutting mechanism, and an automatic corner protector installation device. The tension control unit is electrically connected to the control signal output interface of the dynamic tension control unit of the control system, and a tension sensor is provided on the membrane frame, which is electrically connected to the tension signal input interface of the dynamic tension control unit.

[0018] Compared with the prior art, the beneficial effects of the present invention are: By combining 3D vision with neural networks, a quantitative assessment of cargo vulnerability is achieved, laying the foundation for subsequent differentiated and precise control. Building upon this, reinforcement learning is used to autonomously plan variable-path, variable-speed winding strategies adapted to cargo characteristics. Combined with model predictive control algorithms, this enables millisecond-level flexible tension adjustment at the moment of contact with vulnerable areas, effectively balancing packaging efficiency and fragile item protection. Furthermore, the system possesses continuous self-learning capabilities; each packaging data point is recorded and used for model iteration, ensuring that packaging quality continuously improves over time. Attached Figure Description

[0019] Figure 1 The block diagram of the intelligent wrapping control system provided in the embodiment of the present invention shows the overall module composition and connection relationship of the independent intelligent wrapping control system of the present invention. Figure 2 A schematic diagram of the overall structure of the intelligent wrapping device including the control system provided in an embodiment of the present invention; Figure 3 A detailed structural block diagram of the dynamic tension control unit provided in the embodiments of the present invention is shown. Figure 1 Internal structure and working principle of the dynamic tension control unit; Figure 4 A time-series diagram of the vulnerability weight value curve in dynamic tension control provided in an embodiment of the present invention; Figure 5 A time-series diagram illustrating the comparison curve between target tension and real-time tension in dynamic tension control provided by an embodiment of the present invention; Figure 6 A timing diagram of the servo motor control command curve in dynamic tension control provided in an embodiment of the present invention; Figure 7 This is a feedback flowchart of the parameter self-optimization unit provided in an embodiment of the present invention; Figure 8 A flowchart illustrating the scoring calculation process for the packaging effect evaluation subunit provided in this embodiment of the invention.

[0020] Reference numerals: 210, measuring frame; 220, three-dimensional vision sensor; 230, hyperspectral imaging sensor; 310, rotating platform; 320, membrane holder; 321, tension sensor; 330, membrane breaking mechanism; 340, automatic corner protector installation device; 341, corner protector hopper; 342, handling robot; 343, visual positioning unit. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. It should be noted that these embodiments are for illustrative purposes only and do not constitute any limitation on the scope of the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principle of the present invention should be included within the scope of protection of the present invention.

[0022] Stretch wrapping machines are widely used automated packaging equipment in modern logistics centers. Their basic working principle involves a motor driving a stretch film to rotate around palletized goods, using the stretching and wrapping of the film to secure the goods, prevent dust and moisture. Compared to manual wrapping, automatic wrapping machines offer significant advantages in packaging efficiency, film consumption control, and packaging quality consistency.

[0023] In existing technologies, wrapping machines mainly include two types: rotary and swivel-arm. Rotary wrapping machines use a rotating platform to rotate the goods, while the film frame rises and falls to complete the wrapping process. Swivel-arm wrapping machines, on the other hand, keep the goods stationary, with the swivel arm rotating around the goods to complete the wrapping, making them particularly suitable for oversized, undersized, unstable, or overweight goods. Technically, some embodiments employ a PLC control system, equipped with a pre-stretched film frame (stretch ratio up to 150%-400%), and can set parameters such as the number of top and bottom wraps, the number of wrapping layers, and the lifting speed.

[0024] However, when handling fragile items (such as glassware, ceramics, electronic products, and irregularly shaped cakes) in high-speed assembly line scenarios, existing technologies have the following drawbacks: First, there is a contradiction between tension control methods and the need for fragile item protection. In some embodiments, the tension control of the equipment primarily serves two goals: "saving film" and "secure wrapping." Some embodiments employ a "pre-stretching + secondary stretching" technique to achieve a high stretching rate of 200%-300% to save film. During the winding process, some embodiments use a layered tension control strategy, such as reducing tension at the bottom, maintaining a set tension in the middle, increasing tension at the top, and automatically reducing tension by 10%-20% at corners to prevent film breakage. This control strategy aims to prevent film breakage due to stress concentration, and the magnitude and timing of tension reduction are mainly based on the mechanical properties of the film material, without considering the compressive strength of the goods themselves. Therefore, for fragile goods with surface compressive strength below the film tension threshold, even if the film does not break, this tension level may still cause crushing damage to vulnerable areas such as the edges of the goods. For fragile items, when the film passes through vulnerable areas such as the edges of the goods, even if the tension threshold for film breakage is not reached, it may still cause crushing damage. In some embodiments, the equipment lacks the ability to differentiate tension according to the fragility of different parts of the goods, failing to truly achieve the goal of "gently wrapping" fragile items.

[0025] Secondly, packaging parameter settings rely on manual experience and cannot achieve adaptive adjustment. Although some embodiments support multiple preset packaging parameter sets (typically 50-100 sets of parameters can be stored), the initial setting of parameters still depends on manual judgment based on experience. Operators need to manually select the corresponding parameter set according to the type of goods, or adjust parameters such as the number of wrapping turns, tension, and film carriage lifting speed via a touch screen. In a multi-variety, small-batch production mode, frequent manual adjustments lead to two problems: first, low efficiency, requiring machine stoppage for parameter adjustment every time production changes; second, unstable packaging quality, relying on the operator's experience and sense of responsibility, with the same batch of goods potentially showing differences in packaging effects due to human error or lack of experience. Some literature points out that insufficient wrapping tension will cause the wrapping film to loosen and fail to effectively secure the goods, while excessive tension may cause the film to break—this balance of "too much or too little" depends entirely on manual control, making it difficult to guarantee consistency. In addition, for irregularly shaped goods, even with a 3D contour scanning system in some embodiments, packaging strategies still need to be adjusted according to manually preset rules, lacking true autonomous decision-making capability.

[0026] Third, the lack of a closed-loop feedback mechanism prevents continuous optimization. In some embodiments, the control system is an "open-loop" system. While it can record operational data such as packaging quantity and film consumption, it cannot obtain quantitative evaluation information on packaging effectiveness. After each packaging cycle, the equipment cannot automatically assess packaging quality (such as film adhesion, wrapping tightness, and presence of damage), nor can it provide feedback to optimize subsequent packaging parameters. This means that even if a packaging cycle results in poor performance due to improper parameter settings, the equipment cannot "learn" from it and automatically adjust—the same error may repeat. Literature indicates that unsatisfactory wrapping effects may involve multiple factors such as tension settings, speed control, cargo size and shape, film quality, and equipment malfunction. However, in some embodiments, the equipment lacks the ability to comprehensively analyze and adaptively optimize these factors. The experience gained from each packaging cycle cannot be learned and reused by the equipment, leading not only to film waste and fluctuations in packaging quality but also hindering continuous improvement.

[0027] Fourth, equipment maintenance relies heavily on manual experience and lacks predictive capabilities. In some embodiments, equipment fault diagnosis primarily depends on operator observation and experience, or on simple built-in alarm functions (such as membrane breakage alarms). When problems such as non-produced film, film breakage, film feed deviation, or unstable wrapping occur, operators need to check possible causes one by one, such as speed controller failure, film delivery motor failure, film material mismatch, incoordination between turntable speed and film frame tension, or sharp edges on the surface of the goods. This "passive response" maintenance mode makes it difficult to control equipment downtime, especially in high-speed assembly line operations. A single unexpected downtime can lead to a large backlog of fragile items or production line interruption, resulting in significant losses.

[0028] In summary, in some embodiments, when the wrapping machine is used to process fragile items in a high-speed production line scenario, it has technical defects such as lack of targeted tension control, reliance on manual experience for parameter adjustment, lack of closed-loop feedback optimization mechanism, and passive response of maintenance mode. It is difficult to meet the comprehensive requirements of modern logistics for "high efficiency, high quality, and low damage" of fragile item packaging.

[0029] Based on this, one or more embodiments of this specification provide an intelligent wrapping control system suitable for packaging fragile items on high-speed production lines. It solves the technical problems in the prior art, such as the inability to dynamically adjust tension according to the fragility of goods, the reliance on human experience for packaging parameters, and the lack of self-optimization capabilities. It realizes adaptive flexible packaging of fragile items in high-speed production line scenarios, significantly reducing the breakage rate, improving packaging efficiency, and reducing film waste.

[0030] To facilitate understanding of this invention, some of the technical terms used in this document are explained as follows: 3D point cloud data: refers to a data set that records the spatial position and attribute information of an object's surface in the form of points, obtained through 3D scanning equipment. Each point contains 3D coordinates (X, Y, Z) and possible information such as color and reflection intensity.

[0031] Neural network model: refers to a computational model that simulates the structure and function of a biological neural network. It consists of a large number of interconnected neurons and learns patterns and features from data through training.

[0032] Reinforcement learning model: refers to a paradigm of machine learning in which an agent learns the optimal decision-making strategy by interacting with the environment and receiving reward signals.

[0033] Model predictive control: refers to an advanced process control method that uses a dynamic model of the process to predict future outputs, calculates control commands through online rolling optimization, and provides feedback correction for model mismatch and disturbances.

[0034] Dynamic vulnerability weight value: refers to the quantitative value assigned to different areas of the cargo surface in this invention to characterize their vulnerability. The value range is usually between 0 and 1. The larger the value, the more vulnerable the area is.

[0035] Reference Figures 1 to 8 This embodiment provides an intelligent wrapping control system suitable for packaging fragile items on high-speed production lines.

[0036] The intelligent wrapping control system exists as a standalone product. Its physical carrier can be an industrial control circuit board integrating all the aforementioned functional units, an embedded industrial computer, or a control module that can be installed in an existing equipment control cabinet. The intelligent wrapping control system has a built-in database of the mechanical properties of thin film materials and a database of cargo damage thresholds.

[0037] The cargo damage threshold database stores the maximum safe tension thresholds that different cargo materials can withstand during packaging. This database is obtained by conducting destructive stress tests on standard samples of typical materials (such as glass, ceramics, plastics, and paper) and is stored indexed by material labels. For example, the upper limit of the safe threshold for glass is 5N, and the upper limit for paper packaging boxes is 25N. When calculating the target safe tension, the dynamic tension control unit, based on the material category label output by the vulnerability analysis unit, queries the corresponding upper limit of the safe threshold from this database, combines it with the vulnerability weight value, and finally determines the target safe tension value for that area.

[0038] The intelligent wrapping control system consists of a data interface unit, a vulnerability analysis unit, a wrapping path planning unit, a dynamic tension control unit, a parameter self-optimization unit, a corner protector installation control unit, a fault prediction and diagnosis unit, and a 5G communication unit.

[0039] The data interface unit is equipped with multiple physical interfaces, including an Ethernet interface, an RS-485 serial interface, and a USB interface, for data communication with external devices. Its core function is to receive 3D point cloud data from an upstream external cargo sensing device. This data includes the 3D contour information and surface material characteristics (such as reflectivity and color) of the cargo to be packaged. The data interface unit is also configured to receive post-packaging cargo image data from a downstream second vision sensor for subsequent evaluation of the packaging effect.

[0040] The vulnerability analysis unit is electrically connected to the data interface unit, and its core processing module is a convolutional neural network (CNN) subunit. The CNN subunit contains a pre-trained CNN model, specifically a ResNet-50 architecture. The training sample library contains tens of thousands of images of goods labeled with fragile markings, angular areas, overhanging structures, and different materials (such as glass, ceramics, and plastics).

[0041] When 3D point cloud data is input, the convolutional neural network (CNN) subunits first perform data preprocessing, including denoising, registration, and feature extraction. Then, the CNN model extracts geometric features (such as curvature variations, abrupt changes in normal vectors, and edge detection) and spectral features (such as the reflectivity of different materials in specific wavelengths) from the point cloud data. Based on these features, the model outputs the fragility probability of each region and identifies the specific type of fragile feature region, such as: Fragile marking area: Identify warning labels such as "Fragile" and "Glass" printed on the surface of goods; Sharp corner areas: Identify sharp corners and protruding parts of the cargo's outer contour; Suspended structures: Identify unsupported, suspended parts in goods, such as decorations on top of cakes or protruding components of electronic products.

[0042] The vulnerability analysis unit assigns a dynamic vulnerability weight value between 0 and 1 to each identified fragile feature region based on the output of the CNN model. Specifically, the CNN model outputs a fragility probability value P (0 ≤ P ≤ 1) and a material category label for each point cloud region. The vulnerability analysis unit has a pre-defined material-basic vulnerability mapping table and geometric risk coefficient calculation rules. For each identified fragile feature region, its dynamic vulnerability weight value ω = α × P_material + β × P_geometry, where P_material is the basic vulnerability value obtained from the mapping table based on the material category, P_geometry is the geometric risk coefficient calculated based on the curvature abrupt change, suspension height, and other geometric features of the region, and α and β are weight coefficients with α + β = 1. For example, sharp glass edges: the material is glass (P_material=0.9), the edge curvature has a large abrupt change (P_geometry=1.0), so α=0.5 and β=0.5, then ω=0.95; the plane of ordinary paper packaging boxes: the material is cardboard (P_material=0.1), the geometric changes are gradual (P_geometry=0.1), then ω=0.1; suspended structures are weighted 0.6-0.8 according to their suspension height and support conditions. Finally, this unit constructs a three-dimensional cargo model containing vulnerability distribution information, where each spatial point in the model is associated with a vulnerability weight value.

[0043] The winding path planning unit is electrically connected to the vulnerability analysis unit, and its core is a reinforcement learning sub-unit. This sub-unit uses a deep reinforcement learning algorithm, preferably the Deep Deterministic Policy Gradient (DDPG) algorithm, to generate the optimal winding path.

[0044] The state space of reinforcement learning includes: Vulnerability distribution information in a 3D cargo model; Current winding position (indicated by angle and height); Current film tension value; Remaining film length.

[0045] The motion space includes: Turntable speed (or rotary arm speed, 0-20 rpm); Membrane frame lifting speed 320 (0-1m / s); Basic tension setting (10-50N).

[0046] The reward function is designed as follows: Negative packaging time (-t): Encourages rapid completion of packaging; Negative film consumption (-f): Encourages film conservation; Penalty: If the tension of the film exceeds the corresponding safety threshold when passing through a vulnerable area, a large negative reward (-100) is given to force the model to learn to protect the vulnerable area; Penalty: If the goods are loose or shifted after packaging, a negative reward will be given.

[0047] The reinforcement learning subunit also reads tensile strength data of the currently used films from a database of thin film material mechanical properties. For example, for a film with a fracture stress of 20N at 200% elongation, the system will ensure that the tension when passing through vulnerable areas is much lower than this value when planning the path.

[0048] After pre-training and online learning, the reinforcement learning sub-unit can generate an initial winding path. Pre-training is conducted in an offline virtual simulation environment built on a 3D cargo model and a database of thin-film material mechanical properties, enabling the generation of a large number of training samples without damaging the real cargo. After pre-training, the model is deployed to actual equipment for online learning during real packaging processes. During the online learning phase, the output of the policy network passes through a safety limiter to ensure that the output commands for speed, tension, and other actions remain within a preset safety range, preventing damage to the cargo or equipment malfunction due to exploratory actions.

[0049] Different strategies are used for the initial winding path in different regions: Rapid winding section: In the flat area of ​​the main body of the goods, where there are no vulnerable areas, a higher rotation speed (e.g., 15 rpm) and a faster lifting speed (e.g., 0.5 m / s) are used to ensure packaging efficiency; The deceleration and avoidance section automatically reduces the rotation speed (e.g., from 15 rpm to 8 rpm) and lifting speed (e.g., from 0.5 m / s to 0.2 m / s) when approaching the vulnerable area, providing dual protection: First, reducing the winding speed provides a longer response time window for the subsequent MPC tension controller, ensuring a smooth transition of tension to the target safe value before contacting the vulnerable area, avoiding overshoot due to system inertia; second, from a kinematic perspective, it reduces the relative linear velocity between the film and the fragile surface of the goods at the moment of contact, reducing the impact force and achieving "soft contact" with the fragile area. This "deceleration and avoidance section" does not mean physically avoiding or not winding around the area, but rather using reduced speed and tension to allow the film to adhere to the surface of the vulnerable area more gently. After passing the vulnerable area, the system automatically returns to the rotation speed and tension settings of the rapid winding section to continue completing the remaining packaging.

[0050] The dynamic tension control unit is the core of the execution process. It is electrically connected to the winding path planning unit and is equipped with a tension signal input interface for connecting an external tension sensor 321, as well as a control signal output interface for connecting an external servo drive tension control unit.

[0051] The dynamic tension control unit employs a model predictive control (MPC) algorithm. Its internal model describes the dynamic relationship between film tension and servo motor torque, membrane holder 320 speed, and turntable position. As an example, the system dynamics model can be described by the following simplified state-space equations: x(k+1) = A · x(k) + B · u(k) y(k) = C · x(k) The state vector x includes the film tension T and the membrane frame lifting linear velocity v, while the input vector u includes the servo motor torque command τ and the membrane frame speed command. The output y is the actual film tension. The parameters of matrices A, B, and C are predetermined and stored in the control system through system identification experiments (such as step response tests). At each sampling time, the MPC controller predicts the output in the next N time domains based on the current state x(k) and the model, and solves an online quadratic programming problem with tension tracking error and control quantity change as penalties to obtain the optimal control sequence u(k), u(k+1), ..., u(k+N-1). The first element u(k) is then output as the tension adjustment command for the current moment.

[0052] Specifically, the dynamic tension control unit includes a feedforward control subunit and a feedback control subunit.

[0053] The feedforward control subunit is configured to predict upcoming vulnerable areas based on the initial winding path provided by the winding path planning unit. For example, when the path information shows that the next vulnerable area is 100mm away and expected to arrive in 0.5 seconds, the feedforward control subunit calculates the tension value that needs to be adjusted in advance. This calculation is based on the stress-strain curve of the film: according to the stretching rate corresponding to the current packaging speed (e.g., 500mm / s), the elastic modulus E of the film at that rate is matched from the film material mechanical property database, and then combined with the vulnerability weight value ω, the target safe tension F_target = F_base × ω is calculated, where F_base is the base packaging tension and ω is the vulnerability weight value (e.g., 0.2 means that the area can only withstand 20% of the base tension). The final target tension value is ensured to be within the safe threshold range corresponding to the material in the cargo damage threshold database (e.g., the upper limit of the safe threshold for glass is 5N).

[0054] The feedback control subunit is configured to receive feedback data from the external tension sensor 321 in real time. When the actual tension deviates from the target tension calculated by the feedforward, the feedback control subunit uses a PID algorithm (proportional-integral-derivative control) to perform closed-loop correction on the feedforward command, ensuring that the tension converges accurately to the target value and eliminating the influence of model errors and external disturbances.

[0055] Reference Figures 4 to 6 When the film passes through the fragile area (t1-t2 period), the tension drops rapidly from 20N to 5N and recovers to 20N after passing through the fragile area. The response time of the whole process is controlled within 50 milliseconds, realizing "unobtrusive wrapping" of fragile items.

[0056] The parameter self-optimization unit is electrically connected to the dynamic tension control unit, the fragility analysis unit, and the winding path planning unit, respectively, and its core is the packaging effect evaluation sub-unit.

[0057] The packaging effect evaluation subunit receives images of the packaged goods from a downstream second vision sensor via a data interface unit. This unit uses image processing algorithms for analysis. Film adhesion: Calculate the ratio of the area of ​​the film to the surface of the goods to identify any bulges or loose areas; Damage status: By comparing images of the goods before and after packaging, identify any new damage or cracks; Looseness: Analyze the tightness of the film wrapping at the top and bottom of the goods to determine if there is a risk of displacement.

[0058] The parameter self-optimization unit records complete data for each packaging process, including: Tension curve (sampled 100 times per second); Membrane consumption (in meters or grams); Packaging time; Packaging effect evaluation score (0-100 points).

[0059] This data is periodically used (e.g., once a day or once every 100 wraps) to incrementally train the CNN model of the vulnerability analysis unit and the reinforcement learning model of the entanglement path planning unit. For example, for cases with high wrapping performance scores, their weight in the training samples is increased; for cases with low scores, the reasons are analyzed and the model parameters are adjusted. Through this continuous learning mechanism, the system's recognition accuracy and wrapping performance are continuously improved.

[0060] The corner protector installation control unit is electrically connected to the vulnerability analysis unit and is configured to generate a corner protector installation command based on the location of the corner areas in the 3D cargo model. When the vulnerability weight value exceeds a preset threshold (e.g., 0.8), the command is sent to the external automatic corner protector installation device 340 via a control signal output interface. This command includes parameters such as the corner protector material picking position, bonding position, and bonding angle, and is used to bond the corner protector material to the corners before the film comes into contact with the fragile feature areas, providing physical protection.

[0061] The fault prediction and diagnosis unit is equipped with multiple signal input interfaces for real-time monitoring of vibration, current, and temperature sensor data from the external servo-driven tension control unit, the rotary platform 310, and the membrane breaking mechanism 330. This unit incorporates a Long Short-Term Memory (LSTM) time-series prediction model, learning the normal operating modes of the equipment based on historical data.

[0062] When the real-time characteristics of a certain parameter are detected to deviate from the baseline by more than a preset threshold, such as the vibration characteristics of the main bearing gradually increasing, the LSTM model predicts the remaining service life and issues a preventive maintenance warning 72 hours in advance, suggesting that the bearing be replaced during planned shutdowns to avoid production interruptions caused by unplanned shutdowns.

[0063] The 5G communication unit is configured to upload updated parameters (including CNN model weights, reinforcement learning model parameters, optimal tension curves, etc.) generated by the parameter self-optimization unit to the cloud server in real time. Simultaneously, this unit receives optimized model parameters shared by five other wrapping machines within the same industrial park from the cloud server. Through this group learning mechanism, the learning results of a special packaging case encountered by one machine (such as a novel irregularly shaped container) can be quickly shared by other machines, accelerating the overall optimization process.

[0064] Reference Figure 2 This embodiment provides an intelligent wrapping device that includes the above-described control system.

[0065] The equipment is equipped with a cargo sensing system, a control system, and an adaptive packaging execution system sequentially along the production line.

[0066] The cargo sensing system is installed on a measuring frame 210 upstream of the production line, and includes a 3D vision sensor 220 (specifically, a Keyence LJ-V7000 series laser profile scanner) and a hyperspectral imaging sensor 230. When the cargo to be packaged moves along the conveyor belt to below the measuring frame 210, the encoder triggers the 3D vision sensor 220 and the hyperspectral imaging sensor 230 to synchronously collect data, generating 3D point cloud data containing 3D contour information and surface material spectral characteristics. The cargo sensing system communicates with the control system's data interface unit via an Ethernet cable, transmitting the point cloud data to the control system in real time.

[0067] The adaptive packaging execution system is located downstream of the production line and includes a rotary platform 310, a film holder 320 with a servo-driven tension control unit, a tension sensor 321, a film breaking mechanism 330, and an automatic corner protector installation device 340.

[0068] The rotating platform 310 is driven by a servo motor and carries the goods to rotate. The rotation speed is controlled by the dynamic tension control unit of the control system through control signals.

[0069] A servo-driven tension control unit is installed on the membrane holder 320. This unit is electrically connected to the control signal output interface of the dynamic tension regulation unit of the control system, receives tension adjustment commands, and precisely controls the output tension of the film.

[0070] Tension sensor 321 is installed on the film exit path of film holder 320 to monitor film tension in real time. It is electrically connected to the tension signal input interface of dynamic tension control unit to provide feedback data.

[0071] The film cutting mechanism 330 automatically cuts the film at the end of the packaging process.

[0072] The automatic corner protector installation device 340 includes a corner protector hopper 341, a picking and placing robot 342, and a vision positioning unit 343. It is electrically connected to the control signal output interface of the corner protector installation control unit and is used to execute corner protector installation commands.

[0073] The equipment's workflow is as follows: Step 1: The goods enter the measurement area, and the 3D vision sensor 220 and hyperspectral imaging sensor 230 collect 3D point cloud data and send the data to the control system.

[0074] Step 2: The vulnerability analysis unit of the control system constructs a three-dimensional cargo model with vulnerability weights, the winding path planning unit generates the initial winding path, and sends the path data and tension control parameters to the dynamic tension control unit.

[0075] Step 3: The goods are moved to the rotating platform 310. The corner protector installation control unit controls the automatic corner protector installation device 340 to install paper corner protectors at the vulnerable corners according to the position of the corners in the three-dimensional goods model.

[0076] Step 4: The rotating platform 310 begins to rotate, the film holder 320 rises and falls, and the dynamic tension control unit, based on real-time sensor feedback and preset path, precisely controls the tension at the millisecond level through a combination of feedforward and feedback adjustment to complete the wrapping and packaging.

[0077] Step 5: After packaging is completed, the goods are moved out of the packaging area. The second vision sensor collects images of the packaged goods and sends them to the parameter self-optimization unit of the control system for effect evaluation.

[0078] Step 6: The parameter self-optimization unit records data and updates the model periodically to achieve continuous optimization.

[0079] Through the aforementioned technical means, this embodiment achieves precise protection of fragile items under high-speed production line conditions (processing efficiency can reach 60-80 pallets / hour). According to simulation tests and preliminary experimental data, after adopting the technical solution of this invention, the breakage rate of fragile item packaging can be reduced by 60%-80%, film consumption can be reduced by 15%-20%, changeover adjustment time can be reduced by more than 90%, and packaging quality consistency can be significantly improved.

[0080] To further illustrate the internal processing logic of the control system, we will now describe it step by step using a cargo with sharp glass edges as an example: Step A: Vulnerability Analysis. After receiving 3D point cloud data from the cargo perception system, the vulnerability analysis unit uses a CNN model to extract geometric and spectral features from the point cloud, identifying four glass corner regions located at the four corners of the cargo. Each corner region is assigned a vulnerability weight value ω=0.95, and the material type is labeled as "glass". The cargo damage threshold database is consulted to obtain the upper limit of the safety threshold for glass material, F_max=5N. Based on this, the system determines the target safety tension for these corner regions as F_target = F_max × ω = 5N × 0.95 ≈ 4.75N.

[0081] Step B: Winding Path Planning. After reading the 3D cargo model with vulnerability weights, the reinforcement learning sub-unit plans an optimized path: in the non-vulnerable planar region of the cargo (ω≤0.2), a fast winding section of 15rpm is used, with the basic tension set at 20N; about 100mm before the corner area, a deceleration and avoidance section is automatically planned, reducing the speed to 8rpm, providing a time window of about 0.3 seconds for tension adjustment.

[0082] Step C: Dynamic Tension Control. During the winding process, when the turntable position sensor indicates 100ms remaining until the next corner area, the feedforward control subunit calculates the required servo motor torque change to reduce the tension from 20N to 4.75N based on the system dynamics model and outputs an initial adjustment command. Simultaneously, the feedback control subunit continuously receives real-time feedback data from the tension sensor 321 and uses a PID algorithm to perform closed-loop correction of the feedforward command. For example, if the actual tension only drops to 5.2N due to batch differences in the film material, deviating from the target of 4.75N, the feedback control subunit immediately calculates the compensation torque and adds it to the servo drive command, ensuring that the tension is precisely converged to a safe range of 4.75N ± 0.2N the instant the film contacts the corner. The response time of the entire dynamic adjustment process is controlled within 50 milliseconds. After passing the corner area, the system smoothly restores the tension to 20N and resumes rapid winding.

[0083] Step D: Parameter Self-Optimization. After packaging is completed, the downstream vision sensor captures images of the packaged product. The packaging effect evaluation subunit analyzes the images and obtains an evaluation score of 98 points, indicating a fit of 95%, no damage, and a looseness of 1%. The parameter self-optimization unit records and stores the complete tension curve, membrane consumption data, and the evaluation score of 98 points as high-weight positive samples for subsequent incremental training of the reinforcement learning model.

[0084] The control system in this embodiment is a standalone product that can be used to intelligently upgrade existing traditional wrapping equipment. By adding it to existing equipment, along with a 3D vision sensor 220 and a tension sensor 321, and connecting it to the existing servo driver, all the aforementioned intelligent functions can be achieved.

[0085] The control system receives data from the newly installed 3D vision sensor 220 through its data interface unit, generates commands to drive the existing servo system through the dynamic tension control unit, and continuously optimizes packaging parameters through the parameter self-optimization unit. This transformation method requires little investment, yields quick results, revitalizes old equipment, and significantly improves packaging quality and efficiency.

[0086] This invention constructs a complete intelligent closed loop of "perception-decision-execution-optimization". From obtaining cargo vulnerability information through three-dimensional perception, to quantitative assessment by neural networks, to path planning through reinforcement learning, to millisecond-level tension adjustment by model predictive control, and then to continuous learning and improvement by the self-optimization module, each link is interconnected and indispensable, forming a complete technical solution.

[0087] In addition, by setting up a fault prediction and diagnosis unit, this solution can perform real-time status monitoring and fault prediction for key actuators such as the servo drive tension control unit, rotary platform and membrane breaking mechanism, and issue preventive maintenance warnings before potential faults occur, effectively reducing unplanned downtime and further ensuring the continuity of high-speed assembly line operations.

[0088] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0089] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. An intelligent wrapping control system, characterized in that, include: The data interface unit is configured to receive three-dimensional point cloud data from an external cargo sensing device, wherein the three-dimensional point cloud data includes the three-dimensional contour information and surface material characteristics of the cargo to be packaged. The vulnerability analysis unit is electrically connected to the data interface unit and is configured to analyze three-dimensional point cloud data through a neural network model, identify and mark the fragile feature areas on the surface of the cargo, and assign a dynamic vulnerability weight value to each fragile feature area to construct a three-dimensional cargo model with vulnerability weights. The winding path planning unit is electrically connected to the vulnerability analysis unit. It is configured to generate an initial winding path and corresponding initial tension setting curve that includes a fast winding section and a deceleration and avoidance section through a reinforcement learning model based on the current tensile strength data of the film based on the three-dimensional cargo model and the pre-stored mechanical property database of the film material. The dynamic tension control unit is electrically connected to the winding path planning unit and is equipped with a tension signal input interface for connecting an external tension sensor. The dynamic tension control unit has a built-in system dynamics model and is also configured to acquire feedback data from the tension sensor of the external winding packaging execution equipment in real time during the winding operation. Based on the vulnerability weight value in the three-dimensional cargo model and the feedback data, and based on the pre-stored system dynamics model, it generates a tension adjustment command through a model predictive control algorithm so that the actual film tension converges to the safe target tension determined by the vulnerability weight value. The system dynamics model describes the dynamic relationship between film tension and servo motor torque, film carriage speed, and turntable position; the tension adjustment command is sent to the servo drive tension control unit of the external wrapping and packaging equipment through the control signal output interface; The parameter self-optimization unit is electrically connected to the dynamic tension control unit, the vulnerability analysis unit, and the winding path planning unit, respectively. It is configured to record the tension curve and film consumption data of each packaging process and feed them back to the vulnerability analysis unit and the winding path planning unit to periodically update the parameters of the neural network model and the reinforcement learning model.

2. The intelligent wrapping control system according to claim 1, characterized in that, The vulnerability analysis unit includes: The convolutional neural network subunit is configured to extract geometric and spectral features from 3D point cloud data and identify fragile markings, angular areas, and suspended structures on the surface of goods based on a preset training sample library, so as to mark fragile feature areas.

3. The intelligent wrapping control system according to claim 1, characterized in that, The winding path planning unit includes: The reinforcement learning subunit is configured to minimize packaging time and film consumption as optimization objectives. It generates the initial winding path through a deep deterministic policy gradient algorithm, combined with dynamic vulnerability weights.

4. The intelligent wrapping control system according to claim 1, characterized in that, The dynamic tension control unit includes: The feedforward control subunit is configured to calculate the start time and adjustment range of the tension adjustment command in advance based on the distance and orientation of the next fragile feature area in the initial winding path; The feedback control subunit is configured to perform closed-loop correction on the tension adjustment command generated by the feedforward control subunit based on real-time feedback data from the tension sensor.

5. The intelligent wrapping control system according to claim 1, characterized in that, It also includes a corner protector installation control unit, configured to generate corner protector installation instructions based on the position of the corner areas in the three-dimensional cargo model, and send them to the external automatic corner protector installation device through a control signal output interface.

6. The intelligent wrapping control system according to claim 1, characterized in that, The data interface unit is further configured to receive image data of the goods after packaging is completed, and the parameter self-optimization unit further includes: The packaging effect evaluation subunit is configured to receive images of the packaged goods from the downstream second vision sensor via the data interface unit, and analyze the film adhesion, damage, and looseness as packaging effect evaluation data.

7. The intelligent wrapping control system according to claim 1, characterized in that, The vulnerability analysis unit is configured to assign a dynamic vulnerability weight value and a corresponding tension safety threshold range to each fragile feature region. The thin film material mechanical property database contains stress-strain curve data of various thin films at different stretching rates. The dynamic tension control unit is also configured to dynamically match the elastic modulus of the current film from the stress-strain curve according to the stretching rate corresponding to the current packaging speed, and calculate the servo motor torque output value corresponding to the tension safety threshold range corresponding to the fragility weight value of the current fragile feature area.

8. The intelligent wrapping control system according to claim 1, characterized in that, It also includes a fault prediction and diagnosis unit, which is configured to monitor the vibration, current and temperature data of the external servo drive tension control unit, rotating platform and membrane breaking mechanism in real time through the signal input interface, and predict potential faults through a time-series prediction model to generate preventive maintenance warnings.

9. The intelligent wrapping control system according to claim 1, characterized in that, Also includes: The 5G communication unit is configured to upload the updated parameters generated by the parameter self-optimization unit to the cloud server in real time, and receive the optimized neural network model and reinforcement learning model parameters shared by other wrapping machines from the cloud server.

10. An intelligent wrapping device comprising the control system described in any one of claims 1-9, characterized in that, Also includes: The cargo sensing system, located upstream of the production line, includes a 3D vision sensor and is connected to the data interface unit of the control system. It is used to scan the cargo to be packaged and generate 3D point cloud data. An adaptive packaging execution system, located downstream of the production line, includes a rotating platform, a film holder with a servo-driven tension control unit, a film cutting mechanism, and an automatic corner protector installation device. The tension control unit is electrically connected to the control signal output interface of the dynamic tension control unit of the control system, and a tension sensor is provided on the membrane frame, which is electrically connected to the tension signal input interface of the dynamic tension control unit.

Images