Multi-angle cooperative control method and system for battery cell module tab bending equipment

Abstract

The invention relates to the technical field of battery manufacturing equipment, in particular to a multi-angle cooperative control method and system of battery cell module tab bending equipment, data acquisition: acquiring geometric characteristics, material attributes, real-time positions, bending stress and equipment operation state data of tabs through vision, laser and force sensors, optimizing parameters, and determining the bending state of the tabs. A deep reinforcement learning algorithm is utilized, a bending strategy model is trained based on collected data, bending angle, strength and speed parameters are adjusted in real time, bending errors and material damage risks are minimized, trajectory planning is carried out, a cross-scale multi-physics field simulation engine is utilized to integrate electrochemical, mechanical and heat conduction models, a bending forbidden zone map is generated, and the bending forbidden zone map is obtained. Global optimization and strategy generation are carried out on a multi-axis motion track by combining a particle swarm optimization algorithm, a state space and an action space are defined through a DRL dynamic strategy generation layer, a bending strategy is generated by combining a reward function and adopting a cloud edge cooperative training mechanism, and accurate control is carried out.

Description

Technical Field

[0001] This invention relates to the field of battery manufacturing equipment technology, specifically to a multi-angle collaborative control method and system for cell module tab bending equipment. Background Technology

[0002] As we all know, in the current booming development of the new energy industry, the quality of battery manufacturing processes, as a core energy storage component, directly affects the development level of the entire industry. As a key component of the battery system, the tab bending process plays an irreplaceable role in ensuring the reliability of the electrical connections of the battery cell module. However, the tab bending equipment and technology currently used in the industry have many problems that urgently need to be solved.

[0003] From a structural perspective, most traditional tab bending equipment uses a fixed-angle mechanical design. This "one-size-fits-all" approach severely limits the equipment's flexibility when dealing with battery cell modules of different specifications. Taking prismatic and cylindrical cells as examples, prismatic cells have diverse tab positions, sizes, and shapes, while cylindrical cells have specific requirements for tab bending angles and methods. Traditional equipment struggles to quickly adapt to these differences. To meet diverse production needs, some companies have had to purchase multiple bending machines of different types. This not only significantly increases equipment procurement costs but also occupies a large amount of production space, while the equipment idle rate remains high, resulting in a substantial waste of resources.

[0004] In terms of control precision, traditional bending processes lack effective real-time feedback and dynamic adjustment mechanisms. Existing equipment typically relies solely on simple preset programs or operator experience to control bending parameters, failing to make precise adjustments based on the actual material properties of the tabs, such as differences in elastic modulus and hardness between different materials, as well as real-time positional changes during production. For example, aluminum and copper tabs require significantly different forces and angles during bending; using the same control parameters can easily lead to quality problems such as tab breakage and deformation. Industry statistics show that in battery cell modules produced using traditional equipment, the defect rate due to poor tab bending is as high as 8%-12%, which not only increases production costs for companies but also seriously affects the product's market competitiveness.

[0005] From the perspective of intelligence, traditional equipment has an extremely low level of intelligence. When faced with complex and changing operating conditions, such as the impact of ambient temperature and humidity variations on the performance of the tab material and subtle differences in material properties, the equipment cannot automatically identify and adjust its control strategies. For example, in high-temperature environments, the flexibility of the tab material changes. If traditional equipment cannot adjust bending parameters in time, it will lead to increased bending angle deviations, affecting the performance of the battery cell module. Furthermore, the equipment lacks self-learning and self-diagnostic functions. When a fault occurs, troubleshooting and repair must be done manually, which not only consumes a lot of time and manpower but also leads to prolonged equipment downtime, severely impacting production schedules. Therefore, it is necessary to propose solutions to this technical problem. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a multi-angle collaborative control method and system for battery cell module tab bending equipment.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the present invention provides the following technical solution: a multi-angle collaborative control method for a cell module tab bending equipment, comprising the following steps:

[0010] Step 1: Data acquisition. Collect data on the geometric features, material properties, real-time position, bending stress, and equipment operating status of the tabs using vision, laser, and force sensors.

[0011] Step 2: Parameter optimization. Using deep reinforcement learning algorithms, a bending strategy model is trained based on the collected data. The bending angle, force, and speed parameters are adjusted in real time to minimize bending error and material damage risk.

[0012] Step 3: Trajectory planning. Using a multi-scale multiphysics simulation engine, electrochemical, mechanical, and thermal conduction models are integrated to generate a bending restricted area map. The particle swarm optimization algorithm is then used to perform global optimization of the multi-axis motion trajectory.

[0013] Step 4: Strategy generation. The state space and action space are defined through the DRL dynamic strategy generation layer. Combined with the reward function, the bending strategy is generated using a cloud-edge collaborative training mechanism.

[0014] Step 5: Precise control. PID control parameters are dynamically generated using a neural differential equation controller, and combined with the characteristics of a biomimetic flexible bending actuator, to achieve precise control of the tab bending.

[0015] Furthermore, the present invention is improved in that, in step 1, a multimodal perception fusion layer is used to process the collected data. The multimodal perception fusion layer adopts the Transformer architecture, performs sparse convolutional encoding on 3D point cloud data, performs position encoding on force-displacement time series data, achieves deep fusion of different modal data through cross-modal attention layer, and dynamically adjusts the weights of visual and force data according to actual working conditions.

[0016] Furthermore, the present invention is improved in that, in step 4, the DRL dynamic policy generation layer defines a 12-dimensional state space and a continuous action space, and the reward function is... Where e represents the natural constant, Δθ represents the bending angle deviation, ΔF represents the bending force deviation, and P crack The probability of electrode cracking is represented by a cloud-edge collaborative training mechanism. The cloud simulator generates complex working conditions for model training, and the edge device uploads strategy performance data to achieve daily iterative updates of the model.

[0017] Furthermore, the present invention is improved in that, in step 5, the neural differential equation controller uses equations Dynamically generate PID control parameters, where d represents a small change, t is time, and K is... p K i K d These are the proportional, integral, and differential coefficients, respectively, and θ. error It's an angular error. It is the rate of change of force, T represents temperature, and f is a function parameterized by a neural network.

[0018] Furthermore, the present invention includes a predictive maintenance step: predicting the lifespan of key components of the equipment based on digital twins and LSTM models, and automatically triggering a recommended maintenance plan when an anomaly is detected.

[0019] This invention also provides a multi-angle collaborative control system for a battery cell module tab bending device, including a data acquisition module, a parameter optimization module, a trajectory planning module, a strategy generation module, and a precision control module. The data acquisition module is used to collect data on the geometric features, material properties, real-time position, bending force, and equipment operating status of the tabs through vision, laser, and force sensors. The parameter optimization module uses a deep reinforcement learning algorithm to train a bending strategy model based on the collected data, and adjusts the bending angle, force, and speed parameters in real time to minimize bending errors and material damage risks. The trajectory planning module uses a cross-scale multiphysics simulation engine to integrate electrochemical, mechanical, and thermal conduction models to generate a bending no-go zone map, and combines a particle swarm optimization algorithm to globally optimize the multi-axis motion trajectory. The strategy generation module defines the state space and action space through a DRL dynamic strategy generation layer, combines a reward function, and uses a cloud-edge collaborative training mechanism to generate bending strategies. The precision control module uses a neural differential equation controller to dynamically generate PID control parameters, and combines the characteristics of a biomimetic flexible bending actuator to achieve precise control of tab bending.

[0020] Furthermore, the present invention includes a multimodal perception fusion module, which adopts a Transformer architecture to perform sparse convolutional encoding on 3D point cloud data, implement position encoding on force-displacement temporal data, achieve deep fusion of different modal data through a cross-modal attention layer, and dynamically adjust the weights of visual and force data according to actual working conditions to process the data collected by the data acquisition module.

[0021] Furthermore, the present invention includes a biomimetic flexible bending actuator, which is driven by a McKibben-type pneumatic muscle array and is equipped with an SMA microneedle array that can automatically expand and contract according to the contact temperature and a pressure-sensitive film that generates stress cloud maps in real time.

[0022] Furthermore, the present invention includes a modular adaptive bending mechanism, which supports automatic tool adaptation and zero-adjustment tool change, is equipped with a laser calibration module to detect the initial position error of the bending arm and generate compensation parameters, and can automatically select the appropriate tool according to the tab thickness.

[0023] Furthermore, the present invention includes a cloud collaboration and remote optimization module. This module builds an industrial internet platform to realize data sharing and model iteration among multiple factories. Each factory's equipment trains its model locally and uploads encrypted parameters to the cloud for aggregation. The cloud model periodically sends optimized bending strategies to the local equipment.

[0024] (III) Beneficial Effects

[0025] Compared with the prior art, the present invention provides a multi-angle collaborative control method and system for battery cell module tab bending equipment, which has the following beneficial effects:

[0026] The multi-angle collaborative control method and system for the battery cell module tab bending equipment achieves precise control of parameters such as bending angle and force through multi-modal perception fusion, DRL dynamic strategy generation and neural differential equation controller, reducing the tab bending angle error from ±0.5° of traditional equipment to ±0.1°, and the standard deviation of the 10-tab synchronous bending angle is as low as ±0.08°, significantly improving bending accuracy and reducing product defect rate;

[0027] The modular adaptive bending mechanism enables rapid model changeover, reducing model changeover time from 25 minutes to 38 seconds; combined with optimized trajectory planning and control strategies, production cycle time is shortened, supporting high-speed continuous bending >10 times / second, significantly improving production efficiency;

[0028] The system has self-learning, self-diagnosis and remote collaboration capabilities. The DRL dynamic strategy generation layer realizes adaptive parameter adjustment without human intervention. The predictive maintenance system provides early warning of equipment failures based on digital twins and LSTM models, reducing human intervention. The cloud-based collaborative optimization mechanism promotes the continuous evolution and global standardization of bending processes, reducing reliance on operators' technical skills.

[0029] High-precision bending control reduces product scrap losses, predictive maintenance reduces equipment maintenance costs and downtime, and automated control reduces labor costs, resulting in a significant reduction in overall production costs.

[0030] The biomimetic flexible bending actuator and modular adaptive bending mechanism enable the equipment to adapt to the bending requirements of battery cell modules of different specifications and materials, thereby enhancing the equipment's versatility and market competitiveness. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the steps of the multi-angle collaborative control method of the present invention;

[0032] Figure 2 This is a schematic diagram of the multimodal sensing fusion algorithm of the present invention;

[0033] Figure 3 This is a schematic diagram of the DRL dynamic strategy generation layer architecture of the present invention;

[0034] Figure 4 This is a schematic diagram of the biomimetic flexible bending actuator structure of the present invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Please see Figures 1-4 This invention provides a multi-angle coordinated control method for a battery cell module tab bending equipment, comprising the following steps:

[0037] Step 1: Data acquisition. Collect data on the geometric features, material properties, real-time position, bending stress, and equipment operating status of the tabs using vision, laser, and force sensors.

[0038] Step 2: Parameter optimization. Using deep reinforcement learning algorithms, a bending strategy model is trained based on the collected data. The bending angle, force, and speed parameters are adjusted in real time to minimize bending error and material damage risk.

[0039] Step 3: Trajectory planning. Using a multi-scale multiphysics simulation engine, electrochemical, mechanical, and thermal conduction models are integrated to generate a bending restricted area map. The particle swarm optimization algorithm is then used to perform global optimization of the multi-axis motion trajectory.

[0040] Step 4: Strategy generation. The state space and action space are defined through the DRL dynamic strategy generation layer. Combined with the reward function, the bending strategy is generated using a cloud-edge collaborative training mechanism.

[0041] Step 5: Precise control. PID control parameters are dynamically generated using a neural differential equation controller, and combined with the characteristics of a biomimetic flexible bending actuator, to achieve precise control of the tab bending.

[0042] In this scheme, in step 1, a multimodal perception fusion layer is used to process the acquired data. The multimodal perception fusion layer adopts the Transformer architecture, performs sparse convolutional encoding on 3D point cloud data, implements position encoding on force-displacement time series data, achieves deep fusion of different modal data through a cross-modal attention layer, and dynamically adjusts the weights of visual and force data according to the actual working conditions. This refines the data acquisition and processing steps. By introducing the multimodal perception fusion layer and the Transformer architecture, the acquired data is processed in depth, achieving effective fusion and intelligent analysis of multi-source heterogeneous data such as visual and force data. The data weights are dynamically adjusted according to the actual working conditions, enhancing data reliability, avoiding the influence of single data deviations, providing a more accurate basis for subsequent control strategies, and improving the system's adaptability to complex working conditions.

[0043] In this scheme, in step 4, the DRL dynamic policy generation layer defines a 12-dimensional state space and a continuous action space, and the reward function is... Where e represents the natural constant, Δθ represents the bending angle deviation, ΔF represents the bending force deviation, and P crack This paper describes the probability of tab cracking and employs a cloud-edge collaborative training mechanism. A cloud simulator generates complex working conditions for model training, while edge devices upload strategy performance data to achieve daily iterative updates of the model. The paper details the working mechanism of the DRL dynamic strategy generation layer, including the definition of state space and action space, the setting of reward function, and the cloud-edge collaborative training mechanism. By scientifically setting the state and action spaces and combining them with a reasonable reward function, the cloud-edge collaborative training enables the model to quickly iterate and optimize. This allows the system to automatically learn and generate the optimal bending strategy according to different working conditions without frequent manual intervention, thereby improving the system's intelligence level and the flexibility and adaptability of the bending strategy.

[0044] In this scheme, in step 5, the neural differential equation controller uses equations Dynamically generate PID control parameters, where d represents a small change, t is time, and K is... p K i K d These are the proportional, integral, and differential coefficients, respectively, and θ. error It's an angular error. It represents the rate of change of force, T represents temperature, and f is a parameterized function of the neural network. This explains the principle and method of generating PID control parameters by the neural differential equation controller. It breaks through the limitations of traditional PID parameters being fixed or manually adjusted, and dynamically generates PID parameters according to real-time working conditions to achieve precise control of the tab bending process. Especially when facing complex working conditions such as sudden changes in material thickness and temperature, it effectively reduces overshoot and ensures the stability and accuracy of the bending process.

[0045] This solution also includes predictive maintenance steps: lifespan prediction of key equipment components based on digital twins and LSTM models, automatic triggering of maintenance plan recommendations when an anomaly is detected, introducing digital twins and LSTM models for equipment lifespan prediction and fault warning, changing the traditional passive maintenance mode after equipment failure, realizing proactive maintenance, discovering potential equipment failures in advance, reducing downtime, lowering maintenance costs, improving equipment reliability and production continuity, and ensuring production efficiency and product quality.

[0046] This invention also provides a multi-angle collaborative control system for a battery cell module tab bending device, including a data acquisition module, a parameter optimization module, a trajectory planning module, a strategy generation module, and a precision control module. The data acquisition module collects data on the geometric features, material properties, real-time position, bending force, and equipment operating status of the tabs using vision, laser, and force sensors. The parameter optimization module utilizes a deep reinforcement learning algorithm to train a bending strategy model based on the collected data, adjusting the bending angle, force, and speed parameters in real time to minimize bending errors and material damage risks. The trajectory planning module integrates electrochemical, mechanical, and thermal conduction models using a cross-scale multiphysics simulation engine. The system generates a bending restricted area map and uses a particle swarm optimization algorithm to globally optimize the multi-axis motion trajectory. The strategy generation module defines the state space and action space through a dynamic strategy generation layer (DRL), and generates bending strategies using a cloud-edge collaborative training mechanism combined with a reward function. The precision control module dynamically generates PID control parameters using a neural differential equation controller, and combines the characteristics of a biomimetic flexible bending actuator to achieve precise control of the tab bending. Through the collaborative work of various functional modules, the system achieves automated and intelligent control of the entire tab bending process, ensuring the implementation of the control method at the hardware system level, providing support for the efficient and stable operation of the equipment, and improving the overall performance and competitiveness of the equipment.

[0047] This solution also includes a multimodal perception fusion module. This module employs a Transformer architecture, performing sparse convolutional encoding on 3D point cloud data and position encoding on force-displacement time-series data. It achieves deep fusion of different modal data through a cross-modal attention layer and dynamically adjusts the weights of visual and force data according to actual working conditions. This module processes the data acquired by the data acquisition module, supplementing the control system with the multimodal perception fusion module. Corresponding to the data processing stage in the control method, it enhances the system's ability to process multi-source data, further improving data processing accuracy and reliability. This provides the control system with higher-quality data input, enabling the system to more accurately perceive changes in working conditions, make more reasonable control decisions, and improve system stability and bending quality.

[0048] This solution also includes a biomimetic flexible bending actuator, which is driven by a McKibben-type pneumatic muscle array. It is equipped with an SMA microneedle array that can automatically expand and contract according to the contact temperature and a pressure-sensitive film that generates stress cloud maps in real time. Adding a biomimetic flexible bending actuator to the control system highlights the hardware innovation of the equipment. The pneumatic muscle array enables dynamic stiffness adjustment, the SMA microneedle array solves the problem of hidden cracks in the current collector, and the pressure-sensitive film optimizes the force application trajectory. From the hardware level, the bending quality is improved, the tab damage is reduced, the product qualification rate is increased, and the adaptability of the equipment to different tab materials and shapes is enhanced.

[0049] This solution also includes a modular adaptive bending mechanism. This mechanism supports automatic tool adaptation and zero-adjustment tool changeover. It is equipped with a laser calibration module to detect the initial position error of the bending arm and generate compensation parameters. It can also automatically select the appropriate tool based on the tab thickness. Introducing a modular adaptive bending mechanism into the control system improves the flexibility of the equipment's mechanical structure, realizes automatic tool adaptation and zero-adjustment tool changeover, ensures accuracy through laser calibration, and automatically selects the tool based on the tab thickness. This shortens the equipment model switching time, improves production efficiency, reduces manual operation requirements, and enhances the equipment's versatility and production flexibility.

[0050] This solution also includes a cloud-based collaboration and remote optimization module. This module builds an industrial internet platform to enable data sharing and model iteration among multiple factories. Each factory's equipment trains its model locally and uploads encrypted parameters to the cloud for aggregation. The cloud model periodically sends optimized bending strategies to the local equipment. This adds a cloud-based collaboration and remote optimization module to the control system, constructs an industrial internet collaborative architecture, enables data sharing and model iteration among multiple factories, promotes production process optimization and standardization, and improves the consistency and advancement of each factory's equipment through cloud-based update strategies. This reduces R&D costs and drives technological progress and production level improvement across the entire industry.

[0051] Example 1, conventional battery cell module tab bending:

[0052] Data acquisition and preparation: Standard-sized battery cell modules are placed on the transmission mechanism, and the data acquisition module begins operation. A vision sensor uses a high-resolution camera to generate point cloud data of the electrode tabs, a laser sensor measures the distance between the electrode tabs and the bending arm in real time, a force sensor collects the clamping force of the positioning mechanism on the battery cell module and the force data during the bending process, and a temperature and humidity sensor acquires information such as ambient temperature and humidity. This data is transmitted to a multimodal perception fusion module for processing. The 3D point cloud data is sparsely convolutionally encoded using a Transformer architecture, and the force-displacement time-series data is positionally encoded. Deep data fusion is achieved through a cross-modal attention layer, and data weights are dynamically adjusted based on environmental factors. Part of the processed data is used to build a digital twin model, and the other part is transmitted to the control center. The control center, combined with preset bending process parameters, uses deep reinforcement learning algorithms and adaptive fuzzy control algorithms to calculate the initial control parameters for each actuator.

[0053] The control strategy is executed. The transmission mechanism, based on commands from the control center and position and attitude data processed using a Kalman filter algorithm, transmits the battery cell module to the positioning mechanism with precise speed and path. The positioning mechanism's visual servo system, based on the battery cell module's position information acquired by an image sensor, drives a servo motor through a PID control algorithm, ensuring the positioning fixture precisely fixes the battery cell module. Pressure sensors automatically adjust the clamping force according to the battery cell module's material. The pneumatic muscle joints of the biomimetic flexible bending actuator adjust air pressure according to initial control parameters to achieve axial stiffness adjustment; the SMA microneedle array preheats and prepares, automatically expanding and contracting to form a conformal contact upon contact with the electrode tab; the pressure-sensitive film begins real-time monitoring of stress distribution. The bending mechanism's servo electric cylinder and pressure adjustment device also complete parameter settings, preparing for electrode tab bending.

[0054] During the tab bending process, after bending begins, the multimodal perception fusion module continuously collects multi-source data such as visual and force perception, which is then processed and input into the strategy generation module. The DRL dynamic strategy generation layer generates the optimal bending control strategy in real time based on 12-dimensional state parameters including position error, material properties, and stress conditions, along with a reward function. It outputs adjusted control signals such as angle, force, and speed to the precision control module. The precision control module utilizes a neural differential equation controller to apply equations based on real-time operating conditions.

[0055] Dynamically generated PID control parameters drive the actions of actuators such as servo electric cylinders and pressure regulating devices. Simultaneously, a cross-scale multiphysics simulation engine simulates real-time changes in electrochemical, mechanical, and thermal conductivity physical fields during the bending process, generating a bending exclusion zone map. If potential defects such as localized stress concentration or excessively high temperatures leading to changes in material properties are predicted, the information is promptly fed back to the control center, which adjusts the bending trajectory and parameters to prevent defects. Acoustic emission sensors monitor changes in electrode bending stress in real time; when anomalies occur, the control center responds quickly and further optimizes control parameters to ensure that the electrode bending angle and quality meet requirements.

[0056] After bending and transfer, once the tabs are bent, the positioning mechanism releases the cell module, and the transfer mechanism transports the bent cell module to the next process. During this process, the transfer mechanism continues to use sensors to detect position and orientation, ensuring accuracy and stability. Simultaneously, relevant data from this bending process, such as control parameters, bending effect, and equipment operating status, are stored in the system database for further model training and optimization. Testing showed that the tab bending angle error of this batch of conventional cell modules was controlled within ±0.1°, with a yield rate of 99.8%.

[0057] Example 2, bending of electrode tabs in special specification battery cell modules:

[0058] For battery cell modules with special specifications, the position, shape, and material of the tabs differ significantly from those of standard modules. The data acquisition module utilizes higher-precision image sensors and material sensors to collect detailed data on the tab position, size, and material properties of the battery cell module, such as the effects of special alloy composition and surface treatment processes on hardness and elasticity, as well as the overall shape. The multimodal perception fusion module uses a complex neural network model to deeply mine the collected data and extract feature information of the battery cell modules with special specifications. Based on this data and preset special bending process requirements, the control center uses deep reinforcement learning algorithms, adaptive fuzzy control algorithms, and a quantum-classical hybrid optimization architecture to formulate a specialized multi-angle collaborative control strategy and determine the specific control parameters for each actuator. Specifically, the quantum-classical hybrid optimization architecture transforms the bending collaboration problem into a QUBO model, which is solved using a quantum annealing optimizer to obtain the optimal bending parameters, such as the specific bending angle and force combination for tabs made of special materials.

[0059] In the equipment adjustment and preparation phase, the transmission mechanism, based on instructions from the control center and combined with position and attitude data processed by path planning and Kalman filtering algorithms, transmits the battery cell module to the positioning mechanism at a specific speed and path. The positioning mechanism, based on the unique shape information of the battery cell module obtained from multimodal perception fusion analysis, uses visual servo technology and PID control algorithms to drive the positioning fixture for adaptive adjustment. Pressure sensors automatically adjust the clamping force according to the special material, ensuring the battery cell module is stable and undamaged. The pneumatic muscle joints of the biomimetic flexible bending actuator, the SMA microneedle array, and the servo electric cylinders and pressure adjustment devices of the bending mechanism are all reset according to specific control parameters. Simultaneously, digital twin technology is used to simulate the bending process in virtual space, and a cross-scale multiphysics simulation engine is used to optimize control parameters, ensuring the accuracy of the actual bending process. For example, the extension threshold of the SMA microneedle array and the stiffness adjustment range of the pneumatic muscle joint are adjusted according to the thermal expansion coefficient and elastic modulus of the special material.

[0060] During the bending operation and monitoring of the electrode tabs, the data acquisition module collects operational status data, electrode tab bending data, and stress change data of each actuator at a higher frequency. Based on real-time data, the control center frequently adjusts the control parameters of each actuator using model predictive control algorithms, deep learning algorithms, and neural differential equation controllers. For example, if the electrode tab material of a special-specification battery cell module is brittle, and the acoustic emission sensor detects abnormal stress, the control center reduces the bending pressure of the bending mechanism in real time, while simultaneously optimizing the motion trajectory and speed of the servo electric cylinder to prevent electrode tab breakage. The DRL dynamic strategy generation layer continuously optimizes the bending strategy based on complex operating conditions, ensuring that the electrode tab is accurately bent to the required angle and shape. The cross-scale multiphysics simulation engine updates the bending restricted area map in real time, providing real-time guidance for the bending operation.

[0061] In subsequent processing and verification, after the tab bending is completed, the transmission mechanism transfers the battery module to the quality inspection station for testing. The inspection results are fed back to the control center, which uses deep learning algorithms to analyze the results, determine the reasons for bending quality defects such as angle deviation and surface damage, and optimize control strategies and parameters accordingly. The optimized parameters and experience data are then stored in the system database to provide a reference for the bending operation of similar special-specification battery modules in the future. Testing showed that the average tab bending angle error of this special-specification battery module was ±0.08°, with a yield rate of 99.5%, effectively meeting the production requirements of special-specification products.

[0062] Example 3: Predictive equipment maintenance and cloud-based collaborative optimization:

[0063] Predictive maintenance involves a data acquisition module that continuously collects operational data from critical components such as bending arms, servo motors, and pneumatic joints during routine equipment operation. This data includes vibration frequency, temperature changes, current and voltage fluctuations, and component wear levels. This data is transmitted to the predictive maintenance and self-diagnosis module, which uses digital twins and LSTM models to predict the lifespan of critical components. For example, by analyzing the current variation curve of a servo motor and bearing vibration frequency data, the LSTM model predicts that a servo motor bearing will experience severe wear risk in three months. When an anomaly is detected, the system automatically triggers recommended repair solutions, such as suggesting the specific time to replace the bearing, recommending suitable repair personnel and parts suppliers, and generating detailed repair procedure guides to help maintenance personnel perform repairs quickly and accurately, avoiding downtime losses due to sudden equipment failures.

[0064] Cloud-based collaborative optimization involves each factory's cell module tab bending equipment uploading encrypted operational data, such as production quantity, yield rate, and control parameter adjustment records, to an industrial internet platform built on a cloud-based collaborative and remote optimization module via local networks. The cloud model integrates data uploaded from factories worldwide, using big data analytics and machine learning algorithms for model iteration to optimize bending strategies and process parameters. For example, analyzing bending data from multiple factories producing the same specifications of cell modules revealed that adjusting the initial bending speed and pressure combination for a specific material tab can improve the yield rate by 2%. The cloud model periodically distributes optimized bending strategies to local equipment. Upon receiving the updates, each factory's equipment automatically adjusts its control parameters, achieving consistent improvement and continuous optimization of the production process. Statistics show that within six months of implementing cloud-based collaborative optimization, the average yield rate of each factory increased by 3.5%, and production efficiency increased by 20%, effectively driving technological progress and improving production levels across the entire industry.

[0065] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-angle coordinated control method for a cell module tab bending equipment, characterized in that, Includes the following steps: Step 1: Data acquisition. Collect data on the geometric features, material properties, real-time position, bending stress, and equipment operating status of the tabs using vision, laser, and force sensors. Step 2: Parameter optimization. Using deep reinforcement learning algorithms, a bending strategy model is trained based on the collected data. The bending angle, force, and speed parameters are adjusted in real time to minimize bending error and material damage risk. Step 3: Trajectory planning. Using a multi-scale multiphysics simulation engine, electrochemical, mechanical, and thermal conduction models are integrated to generate a bending restricted area map. The particle swarm optimization algorithm is then used to perform global optimization of the multi-axis motion trajectory. Step 4: Strategy generation. The state space and action space are defined through the DRL dynamic strategy generation layer. Combined with the reward function, the bending strategy is generated using a cloud-edge collaborative training mechanism. Step 5: Precise control. PID control parameters are dynamically generated using a neural differential equation controller, and combined with the characteristics of a biomimetic flexible bending actuator, to achieve precise control of the tab bending.

2. The multi-angle collaborative control method for the cell module tab bending equipment according to claim 1, characterized in that, In step 1, a multimodal perception fusion layer is used to process the collected data. The multimodal perception fusion layer adopts the Transformer architecture, performs sparse convolutional coding on 3D point cloud data, performs position coding on force-displacement time series data, achieves deep fusion of different modal data through cross-modal attention layer, and dynamically adjusts the weights of visual and force data according to actual working conditions.

3. The multi-angle collaborative control method for the cell module tab bending equipment according to claim 1, characterized in that, In step 4, the DRL dynamic policy generation layer defines a 12-dimensional state space and a continuous action space, and the reward function is... Where e represents the natural constant, Δθ represents the bending angle deviation, ΔF represents the bending force deviation, and P crack The probability of electrode cracking is represented by a cloud-edge collaborative training mechanism. The cloud simulator generates complex working conditions for model training, and the edge device uploads strategy performance data to achieve daily iterative updates of the model.

4. The multi-angle collaborative control method for the battery cell module tab bending equipment according to claim 1, characterized in that, In step 5, the neural differential equation controller uses equations Dynamically generate PID control parameters, where d represents a small change, t is time, and K is... p K i K d These are the proportional, integral, and differential coefficients, respectively, and θ. error It's an angular error. It is the rate of change of force, T represents temperature, and f is a function parameterized by a neural network.

5. The multi-angle coordinated control method for the cell module tab bending equipment according to claim 1, characterized in that, It also includes predictive maintenance steps: predicting the lifespan of key equipment components based on digital twins and LSTM models, and automatically triggering maintenance plan recommendations when an anomaly is detected.

6. A multi-angle collaborative control system for a cell module tab bending equipment, characterized in that, The system includes a data acquisition module, a parameter optimization module, a trajectory planning module, a strategy generation module, and a precision control module. The data acquisition module collects data on the geometric features, material properties, real-time position, bending force, and equipment operating status of the tab using vision, laser, and force sensors. The parameter optimization module uses a deep reinforcement learning algorithm to train a bending strategy model based on the collected data, adjusting the bending angle, force, and speed parameters in real time to minimize bending errors and material damage risks. The trajectory planning module uses a cross-scale multiphysics simulation engine to integrate electrochemical, mechanical, and thermal conduction models to generate a bending no-go map, and combines a particle swarm optimization algorithm to globally optimize the multi-axis motion trajectory. The strategy generation module defines the state space and action space through a DRL dynamic strategy generation layer, combines a reward function, and uses a cloud-edge collaborative training mechanism to generate bending strategies. The precision control module uses a neural differential equation controller to dynamically generate PID control parameters, and combines the characteristics of a biomimetic flexible bending actuator to achieve precise control of the tab bending.

7. The multi-angle collaborative control system for the battery cell module tab bending equipment according to claim 6, characterized in that, It also includes a multimodal perception fusion module, which adopts a Transformer architecture to perform sparse convolutional encoding on 3D point cloud data, implement position encoding on force-displacement time series data, achieve deep fusion of different modal data through cross-modal attention layers, and dynamically adjust the weights of visual and force data according to actual working conditions to process the data collected by the data acquisition module.

8. The multi-angle collaborative control system for the battery cell module tab bending equipment according to claim 6, characterized in that, It also includes a biomimetic flexible bending actuator, which is driven by a McKibben-type pneumatic muscle array and is equipped with an SMA microneedle array that can automatically expand and contract according to the contact temperature and a pressure-sensitive film that generates stress cloud maps in real time.

9. The multi-angle collaborative control system for the battery cell module tab bending equipment according to claim 6, characterized in that, It also includes a modular adaptive bending mechanism that supports automatic tool adaptation and zero-adjustment tool change, is equipped with a laser calibration module to detect the initial position error of the bending arm and generate compensation parameters, and can automatically select the appropriate tool according to the tab thickness.

10. The multi-angle collaborative control system for the battery cell module tab bending equipment according to claim 6, characterized in that, It also includes a cloud collaboration and remote optimization module. This module builds an industrial internet platform to realize data sharing and model iteration among multiple factories. Each factory's equipment trains the model locally and uploads encrypted parameters to the cloud for aggregation. The cloud model periodically sends the optimized bending strategy to the local equipment.

Images