New energy battery pack disassembling process

Abstract

The invention relates to the technical field of battery recovery, and provides a new energy battery pack disassembling process, which is implemented in a closed automatic workstation, performs battery pack entrance risk diagnosis and pretreatment through multi-source sensing fusion, and determines a new energy battery pack entrance risk according to a real-time three-dimensional state model and multi-sensor feedback based on a reinforcement learning intelligent decision-making module. And a fine disassembling action sequence is automatically generated and executed, so that the self-adaptive and flexible separation of the shell and the internal components is realized. According to the method, the safety, the working condition adaptability and the nondestructive recovery rate of core components in the disassembling process can be improved, and a reliable intelligent solution is provided for gradient utilization of batteries and efficient recovery of materials through continuous optimization of a data closed-loop driving process.

Description

Technical Field

[0001] This invention relates to the field of battery recycling technology, specifically a dismantling process for new energy battery packs. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the large-scale retirement of power battery packs has become an unavoidable reality. Efficient, safe, and environmentally friendly battery pack dismantling and recycling are crucial for achieving resource recycling and sustainable industrial development. However, current mainstream dismantling processes still face severe technical bottlenecks: First, due to the wide variety of brands, models, structures, and damage conditions of retired battery packs, traditional dismantling methods relying on fixed procedures or manual experience are poorly adaptable, inefficient, and unable to cope with unknown internal risks. Second, the dismantling process involves multiple safety hazards such as high voltage, high temperature, and electrolyte leakage; existing technologies lack systematic risk awareness and adaptive handling capabilities, resulting in high accident risks. Third, rough mechanical dismantling easily damages high-value components such as cells, modules, and copper busbars, severely reducing the material's reusability and economic value.

[0003] Therefore, a new energy battery pack dismantling process is proposed. First, by fusing infrared thermal imaging, millimeter-wave radar, electrical diagnostics, and cloud data, a multi-dimensional non-destructive assessment and risk classification of the battery pack's arrival status is achieved, followed by active safety discharge and targeted pre-processing. Subsequently, a precise work map is established using high-precision 3D scanning and machine vision. An intelligent decision-making module, integrating reinforcement learning algorithms and real-time feedback from multiple sensors, autonomously generates and executes a refined dismantling sequence adapted to the current component status (e.g., bolts, colloids, solder joints). The execution process employs flexible techniques such as force-position hybrid control, isothermal softening, ultrasonic cutting, and laser cutting to maximize component integrity. Finally, all process data is synchronized to a cloud-based digital twin platform, driving continuous iteration of the dismantling strategy and overall system optimization. This significantly improves dismantling efficiency and resource recovery value while ensuring operational safety. Summary of the Invention

[0004] To address the problems of poor adaptability, high safety risks, and easy damage to high-value components in existing disassembly methods, this invention provides a new energy battery pack disassembly process.

[0005] To achieve the above-mentioned objectives, the present invention provides the following technical solution: a new energy battery pack disassembly process, implemented in a closed automated disassembly workstation equipped with negative pressure suction and inert gas protection functions, the method comprising the following steps: Entry diagnostics and safety pretreatment steps: Obtain the baseline digital profile of the battery pack and perform a non-contact scan to identify areas of safety risk, followed by electrical isolation and graded discharge.

[0006] Adaptive disassembly of the outer shell: A model of the outer shell is constructed through high-precision 3D scanning. The connection features are identified by machine vision and a disassembly sequence is generated. The intelligent actuator, which integrates multiple sensors and a tool library, adaptively disassembles the outer shell connection based on real-time force feedback.

[0007] The intelligent separation step of internal components: After removing the outer shell and initializing the internal state, the actuator operation is controlled based on the reinforcement learning decision module; the reinforcement learning decision module outputs composite action commands to control the displacement, force and tool switching of the actuator according to the real-time system state that integrates visual features, force feedback and process context, so as to realize the adaptive and fine separation of internal components, while an independent safety guardian process performs real-time monitoring and intervention.

[0008] Material handling and data optimization steps: The disassembled parts are classified and recycled, and the data of the whole process is linked with the battery pack identification code and uploaded to the cloud digital twin platform for optimization of disassembly strategy.

[0009] Furthermore, in the entry diagnosis and safety pretreatment steps, the non-contact scanning includes: using an infrared thermal imager to scan and mark areas with temperatures 15°C or higher than the ambient temperature as Level 1 risk areas; and using millimeter-wave radar to scan and mark areas with structural collapse, module displacement, or liquid accumulation problems as Level 2 risk areas.

[0010] Furthermore, the graded discharge is as follows: discharge is performed through the high-voltage maintenance interface of the battery pack at a current not exceeding 0.05C until the total voltage of the battery pack drops below 60V, where C is the rated capacity of the cell; and the surface temperature rise rate of the battery pack is monitored in real time during the discharge process. When the rate is greater than 1℃ / min, the discharge is paused and the cooling program is started.

[0011] Furthermore, for battery packs marked as Level 1 risk areas or with a history of collisions or water immersion, a special pretreatment is performed after discharge: the battery pack is placed in a negative pressure pretreatment chamber, inert gas is introduced and reactive polymerizing agent is injected to form a solid polymer in situ at potential electrolyte leakage points in the battery pack for fixation.

[0012] Furthermore, in the adaptive release step of the outer shell, the generation of the disassembly sequence includes: comparing the three-dimensional scan model with the reference digital file, and marking it as a non-standard variation point when the actual position of the outer shell seam is offset from the nominal position recorded in the file by more than 5mm; and generating a disassembly action sequence containing operation priority and risk notes according to the principle of first releasing the electrical connection and then the mechanical connection, and first releasing the peripheral connection and then the core connection.

[0013] Furthermore, when the intelligent actuator disconnects the housing: for screw connections, disassembly is performed within a preset torque range. When the actual torque reaches the preset upper limit but the screw is not turned, the actuator first applies a micro-vibration with a frequency of 100Hz and an amplitude of 0.1mm in the axial direction for assistance; if this is ineffective, the tool head is replaced with a micro hollow drill bit to drill and remove the screw; for sealant connections, a constant temperature hot air knife is used to heat the joint at a temperature of 200±10℃ for 60 to 90 seconds to soften it, and then a flexible scraper with force feedback is used to apply a separation force of 5 to 10N to peel it off.

[0014] Furthermore, the reinforcement learning decision module adopts an architecture that combines imitation learning initialization with online reinforcement learning fine-tuning; the real-time system state is a high-dimensional vector fused and encoded from local visual point cloud features, normalized readings of six-dimensional force / torque sensors, tool head identifiers, actuator end pose, and historical action sequence embeddings; the composite action commands include millimeter-level displacement increments, Newton-level force components, and discrete commands for tool hold or switch in the tool coordinate system.

[0015] Furthermore, the optimization objective of the reinforcement learning decision module is defined by a reward function, which includes: a task completion reward for successfully disassembling the target component, an efficiency reward for smooth execution of actions, a safety penalty for exceeding the threshold force or unexpected deformation of the component, and a progress reward for improved tool alignment accuracy or increased connection-to-separation ratio.

[0016] Furthermore, in the intelligent separation step of the internal components: for battery modules fixed by structural adhesive, a linear ultrasonic scalpel with a frequency matching the characteristics of the adhesive layer is used to cut the adhesive layer; after the module frame is removed, an adaptive vacuum adsorption array gripper with independent air pressure control and micro-displacement sensor is used to extract individual battery cells at a speed of 5 to 10 mm / s; for the laser welding points of aluminum busbars, a pulsed fiber laser with a pulse energy of 20J and a pulse frequency of 100Hz is used for scanning cutting, and the damage depth of the electrode surface is controlled to be less than 0.1 mm.

[0017] Furthermore, in the material handling and data optimization steps, the cloud-based digital twin platform uses machine learning algorithms to analyze the uploaded full-process data, discover optimized dismantling strategies under different fault or damage modes, and distribute the updated strategies to each dismantling workstation to achieve continuous adaptive optimization of the dismantling system.

[0018] Compared with the prior art, the present invention provides a new energy battery pack disassembly process, which has the following beneficial effects: 1. In this solution, an autonomous action generation capability capable of responding to unknown structures and damage states is constructed through a reinforcement learning decision-making module. Combined with a multi-dimensional perception system of infrared, millimeter wave, electrical, and cloud data, it achieves advanced identification and precise handling of safety risks, solving the core problems of low safety and poor adaptability to working conditions in traditional dismantling.

[0019] 2. This solution achieves precise and non-destructive disassembly of high-value components by integrating force-position hybrid control, flexible separation technology, and real-time feedback. Simultaneously, it leverages a cloud-based digital twin platform to create a closed-loop data system and a continuous strategy optimization mechanism. This not only improves the recovery rate and reusability of key materials but also enables the entire disassembly system to learn autonomously and continuously evolve from single operations. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall process flow of the present invention; Figure 2 This is a schematic diagram of the entry diagnostic process of the present invention; Figure 3 This is a schematic diagram of the shell removal process of the present invention; Figure 4 This is a schematic diagram of the reinforcement learning module of the present invention; Figure 5 This is a schematic diagram of the internal separation process of the present invention; Figure 6 This is a schematic diagram of the material sorting process of the present invention; Figure 7 This is a schematic diagram of the security monitoring process of the present invention; Figure 8 This is a schematic diagram of the data closed-loop process of the present invention. Detailed Implementation

[0021] 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.

[0022] Example 1: Please see Figure 1 A new energy battery pack dismantling process is disclosed. This process is implemented in a closed, automated dismantling workstation. The workstation has negative pressure suction and inert gas protection functions, and integrates sensing, control, and execution systems, enabling adaptive dismantling of the battery pack from safety pretreatment to material recovery throughout the entire process. The method of this embodiment specifically includes the following steps: Step 1: Entry Inspection and Initial Safety Procedures Please see Figure 2 Before the disassembly operation begins, the battery pack must undergo on-site diagnostics and initial safety procedures to identify and mitigate potential risks.

[0023] First, scan the QR code or RFID tag on the battery pack casing to obtain the battery pack's unique identification code. Establish a communication connection between this code and a cloud database to retrieve the manufacturer's specifications, electrical connection diagrams, exploded structural diagrams, and known failure mode information, creating a baseline digital profile of the battery pack to provide data support for subsequent disassembly.

[0024] After establishing the baseline digital profile, a non-contact overall scan of the battery pack is performed. An infrared thermal imager is used to scan the battery pack from all angles, generating a temperature distribution cloud map. An algorithm is then used to identify temperature anomalies. Areas within the battery pack where the temperature exceeds 15°C above the ambient temperature are marked as Level 1 risk areas.

[0025] Simultaneously, millimeter-wave radar scanning equipment is used to perform transmission scanning on the battery pack to generate an internal density distribution image. Based on the image, it is determined whether there are problems such as structural collapse, module displacement, or liquid accumulation inside the battery pack. If so, it is marked as a secondary risk area.

[0026] After the non-contact overall scan is completed, electrical safety isolation and energy discharge operations are performed. The robotic arm locates and connects to the high-voltage maintenance interface of the battery pack, and attempts to communicate with the battery management system (BMS) through this interface to read the real-time voltage, temperature, and historical fault codes of the cells.

[0027] When communication is successful, the internal consistency risk is assessed based on the cell voltage imbalance. The cell voltage imbalance satisfies the formula... ,in , This is the highest voltage of the battery cell. This represents the minimum cell voltage, all in volts (V). A graded flexible active discharge procedure is initiated regardless of whether communication with the BMS is successful.

[0028] During discharge, a programmable electronic load is connected through the high-voltage maintenance interface, with a discharge rate not exceeding [a certain value]. The current discharges to the positive and negative terminals of the battery pack, among which... This refers to the rated capacity of the battery cell, measured in Ah. (For rated capacity...) The battery cell, discharge current The unit is amperes (A). The goal of discharging is to reduce the total voltage of the battery pack to below 60V.

[0029] During discharge, thermocouples are used to monitor the surface temperature of the battery pack in real time, and the temperature rise rate is monitored. Immediately pause the discharge and initiate the cooling process. Resume discharge only after the temperature returns to normal. The rate of change of surface temperature of the battery pack, in units of .

[0030] For battery packs marked as Level 1 risk areas, or those with a history of collisions or water immersion as shown in the database, a special pretreatment is performed after electrical safety isolation and energy release are completed. The battery pack is placed in a separate negative pressure pretreatment chamber, argon gas is introduced into the chamber, and a reactive polymer agent is injected simultaneously. This polymer agent can undergo a cross-linking reaction with carbonate solvents to form polymers around potential leak points in the battery pack, thereby achieving in-situ fixation of free electrolyte.

[0031] Step 2: Fine-grained disassembly of the outer shell and modeling of the internal state Please see Figure 3 This section removes the battery pack casing using non-destructive or low-destructive methods and constructs an internal dynamic 3D working map to provide data support for subsequent disassembly of internal components.

[0032] Two or more high-precision line laser scanners are used to scan the battery pack casing from different angles in all directions to generate a three-dimensional point cloud model. The accuracy of the three-dimensional point cloud model is at the sub-millimeter level.

[0033] Based on a 3D point cloud model, a machine vision system is used to identify and label feature points on the outer casing. These feature points include screws, clips, sealant seams, high-pressure interfaces, and labels. Screws include types such as Phillips head screws and hex socket screws. The machine vision system employs the YOLOv5 algorithm model based on deep learning.

[0034] After completing the shell feature recognition, the scan results are compared with the reference digital archive, and the differences between the physical object and the archive are marked. When the seam position offset... When, it is marked as a non-standard variant point, where This represents the offset between the actual position of the seam and the position of the file, expressed in mm.

[0035] Simultaneously, a disassembly action sequence is generated according to the order of first disconnecting electrical connections, then disconnecting mechanical connections, and first disconnecting the periphery, then disconnecting the core. The disassembly action sequence is in XML format and includes priority and risk notes.

[0036] After the disassembly sequence is generated, an adaptive shell separation operation is performed. The robotic arm is equipped with an intelligent end effector, which integrates a miniature high-definition camera, a six-dimensional force / torque sensor, and a tool head library, including electric screwdrivers, heat guns, ultrasonic scalpels, etc.

[0037] Preset torque range when removing screws ,in Torque, unit: The six-dimensional force / torque sensor provides real-time feedback on torque and axial force data. When the actual torque reaches the preset upper limit and the screw is not tightened, it is determined that the screw is corroded or stripped, and the rotation is stopped and the pre-set plan is executed.

[0038] The first contingency plan is a micro-vibration assisted mode, in which the actuator applies axial vibration at a frequency of... ,amplitude ,in Frequency, in Hz. The amplitude is measured in mm, while maintaining the rotational torque. If the first contingency plan fails, the second contingency plan is executed, replacing the tool head with a miniature hollow drill bit to drill out the screw at its center.

[0039] The sealant on the outer casing is softened by heating with a constant-temperature hot air knife at a temperature of [temperature value missing]. Heating time ,in Heating time, in seconds. The constant-temperature hot air knife moves at a uniform speed along the joint.

[0040] After the sealant softens, a flexible scraper with force feedback is used to apply a separating force perpendicular to the joint. ,in The force is the separation force, measured in N. It is achieved by gradually peeling off the outer shell with a flexible scraper, thus separating the outer shell from the internal components.

[0041] Step 3: Adaptive Fine-grained Separation of Internal Components Please see Figures 4-5 This section employs a reinforcement learning decision module to achieve adaptive and refined separation of internal components, addressing the problem that traditional rules cannot cover complex situations.

[0042] The reinforcement learning decision-making module employs a hybrid algorithm architecture that combines imitation learning initialization with online reinforcement learning fine-tuning. The internal component decomposition process is transformed into a Markov decision process, achieving decision-making through multi-dimensional data fusion and command control.

[0043] At the moment of decision The system status is ,in For time step identifier, The vector is a high-dimensional vector, fused and encoded from visual features, force features, and process context. Visual features are extracted from local point clouds using a lightweight 3D convolutional neural network; force features are normalized readings from a six-dimensional force / torque sensor; and process context includes the toolhead ID and the pose of the toolhead end effector. And the embedding representation of the most recent 10 executed action history sequences, where In a spatial rectangular coordinate system, This is the roll angle. The pitch angle, Yaw angle, units are all .

[0044] At the moment of decision Execute actions , These are fine-grained compound instructions used to control the actuator's actions in the next time step. This includes displacement motion, force control motion, and tool motion; positional movement is the displacement increment in the tool coordinate system. The displacement range is ±10mm, where These represent the displacement increments in the x, y, and z directions, respectively, in mm; force-controlled action is the applied force. ,in These represent the forces acting in the x, y, and z directions, respectively, with units of N; tool actions are discrete commands that maintain the current tool or switch tools. For example: This means moving 2mm in the x-direction, applying 5N of pressure in the z-direction, and maintaining the current tool position.

[0045] The reward function combines sparse and dense rewards, including task completion rewards, efficiency rewards, safety penalties, and progress rewards. Successfully disassembling a target component earns a task completion reward of +1000 points; each smooth time step of an action earns an efficiency reward of +0.1 points; situations such as exceeding force thresholds or unexpected component deformation incur a safety penalty of -50 to -200 points; progress improvements such as reduced tool alignment errors and increased colloid separation ratios earn a progress reward of +1 to +10 points.

[0046] The reinforcement learning algorithm employs either Deep Deterministic Policy Gradient (DDPG) or Soft Actor-Critic (SAC) algorithms, and includes two phases: offline pre-training and online decision-making and learning.

[0047] During the offline pre-training phase, in both the simulation environment and on-machine testing, engineers controlled the robotic arm via a master-slave teleoperation system to disassemble battery packs of different models and with varying degrees of damage, recording the status as they went. and actions To form an expert dataset ,in For the expert dataset, the initial policy network is trained using behavior cloning. ,in For policy networks, For network parameters, Indicates the initial state.

[0048] Policy Networks in the Online Decision-Making and Learning Phase Based on the current state Output Action And add exploratory noise. The actuator performs the action. Afterwards, the environment provided feedback on the new status. and instant rewards ,in The state after the action is performed. For immediate rewards. Experience tuples. Store in the experience replay buffer ,in This serves as a buffer for experience replay.

[0049] The learning thread runs in the background, from Sample data and update the policy network and value network ,in For value networks, For network parameters. An entropy regularization term is set during the update process to encourage exploration.

[0050] After the outer casing is removed, a second scan and status initialization are performed on the inside of the battery pack. The robotic arm is equipped with an endoscope and a structured light 3D scanning head to perform a comprehensive scan of the inside and update the 3D model. The 3D model displays the layout and physical status of the modules, wiring harnesses, cooling pipes, and BMS.

[0051] Meanwhile, an array of insulating probes is used to sequentially contact the positive and negative measurement points of each module to draw an internal electrical connection topology diagram. The topology diagram is accurate to the module level and identifies the charged module clusters.

[0052] After completing the internal scan and topology mapping, the task objectives and initial state will be determined. Input reinforcement learning decision module, where This represents the system state at the initial moment.

[0053] Data from multiple sensors is imported into the system in real time, updated, and encoded as the current state. Policy Network according to Calculate actions For example: [t adjusts the position to align with the screw center, switches to the electric screwdriver head, and applies a torque of T=8Nm], where Torque, unit: .

[0054] The system is configured with a security daemon process, which monitors in real time based on rules. and If a sudden surge in force, open flame, or other danger signals are detected, execution should be stopped immediately. It then performs an emergency stop or rollback operation, and feeds this situation back to the reinforcement learning algorithm as a negative reward.

[0055] Actuator execution Afterwards, the operation results and experience data are recorded. Under the control of the reinforcement learning strategy, the actuator adopts the corresponding separation method according to the fixing method of different components. For modules fixed with bolts, disassembly is performed according to the disassembly action sequence; for modules fixed with structural adhesive, a linear ultrasonic scalpel is used to cut the adhesive layer, and the frequency of the linear ultrasonic scalpel is matched with the characteristics of the adhesive layer.

[0056] After the module frame is removed, an adaptive vacuum adsorption array gripper is used to extract the battery cells. Each adsorption head of the adaptive vacuum adsorption array gripper has independent air pressure control and micro-displacement sensors. Battery cell extraction speed... ,in Extraction speed, in mm / s.

[0057] For laser welding joints on aluminum busbars, a pulsed fiber laser is used for scanning cutting, with laser pulse energy... pulse frequency ,in Energy, measured in J. Frequency, measured in Hz. Depth of laser cutting. ,in The depth of surface damage to the pole piece is measured in mm.

[0058] Once a single battery pack has been disassembled, or after the system has accumulated a preset amount of experience data, a background learning thread is started, using the newly collected data to refine the strategy network. Update.

[0059] Step 4: Material Handling and Data Closed Loop Please see Figures 6-8 After disassembly, the disassembled parts are processed, and the data from the disassembly process is collected and analyzed to optimize the process.

[0060] The disassembled components include plastic casings, aluminum frames, copper busbars, complete modules, individual battery cells, BMS boards, and wiring harnesses, which are transported to the sorting area via conveyor belt. The sorting area is equipped with a near-infrared spectroscopy (NIR) material recognition system and a vision recognition system to classify the components and temporarily store different types of components in anti-static or fireproof containers.

[0061] During the disassembly process, sensor data and decision execution data are collected. The sensor data includes images, point clouds, force, temperature, voltage, etc., and the decision execution data includes decision logs, execution results, and time consumption. The collected data is associated with the unique identification code of the battery pack, stored in a structured form in a local database, and simultaneously uploaded to a cloud-based digital twin platform for the disassembly process.

[0062] The cloud-based digital twin platform for disassembly processes uses machine learning algorithms to analyze uploaded data, uncover disassembly strategies under different failure modes, and update the optimized disassembly strategies to each disassembly workstation, thereby achieving continuous optimization of the disassembly system.

[0063] Through the above-described embodiments, this invention enables adaptive disassembly of battery packs, improving the rate of non-destructive acquisition of high-value components while ensuring disassembly safety, and providing technical support for the secondary use of batteries and material recycling.

[0064] Finally, it should be noted that the above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Any obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A disassembly process for a new energy battery pack, characterized in that: The process is implemented in a closed, automated dismantling workstation equipped with negative pressure suction and inert gas protection functions, and includes the following steps: Entry diagnostics and safety pretreatment steps: Obtain the baseline digital profile of the battery pack and perform non-contact scanning to identify areas of safety risk, followed by electrical isolation and graded discharge; Adaptive disassembly of the outer shell: A model of the outer shell is constructed by high-precision 3D scanning, and the connection features are identified by machine vision to generate a disassembly sequence. The connection of the outer shell is adaptively disassembled based on real-time force feedback using an intelligent actuator that integrates multiple sensors and tool libraries. The intelligent separation process of internal components involves the following steps: After removing the outer shell and initializing the internal state, the actuator operation is controlled based on a reinforcement learning decision module. This reinforcement learning decision module employs an architecture combining imitation learning initialization and online reinforcement learning fine-tuning. Its real-time system state is a high-dimensional vector fused from local visual point cloud features, normalized readings from a six-dimensional force / torque sensor, tool head identification, actuator end-effector pose, and historical action sequence embeddings. Based on this real-time system state, the reinforcement learning decision module outputs a composite action command, including millimeter-level displacement increments, Newton-level force components, and discrete commands for tool hold or switch in the tool coordinate system. This enables adaptive and precise separation of the internal components, while a separate safety guardian process provides real-time monitoring and intervention. Material handling and data optimization steps: The disassembled parts are classified and recycled, and the data of the whole process is linked with the battery pack identification code and uploaded to the cloud digital twin platform for optimization of disassembly strategy.

2. The new energy battery pack disassembly process according to claim 1, characterized in that: In the entry diagnosis and safety pretreatment steps, the non-contact scanning includes: using an infrared thermal imager to scan and mark areas with temperatures 15°C or higher than the ambient temperature as Level 1 risk areas; and using millimeter-wave radar to scan and mark areas with structural collapse, module displacement, or liquid accumulation problems as Level 2 risk areas.

3. The new energy battery pack disassembly process according to claim 1, characterized in that: The graded discharge is as follows: discharge at a current not exceeding 0.05C through the high-voltage maintenance interface of the battery pack until the total voltage of the battery pack drops below 60V, where C is the rated capacity of the cell; and monitor the surface temperature rise rate of the battery pack in real time during the discharge process. When the rate is greater than 1℃ / min, the discharge is paused and the cooling program is started.

4. The new energy battery pack disassembly process according to claim 3, characterized in that: For battery packs marked as Level 1 risk areas or with a history of collisions or water immersion, a special pretreatment is performed after discharge: the battery pack is placed in a negative pressure pretreatment chamber, inert gas is introduced and reactive polymerizing agent is injected to form a solid polymer in situ at potential electrolyte leakage points in the battery pack for fixation.

5. The new energy battery pack disassembly process according to claim 1, characterized in that: In the adaptive release step of the outer shell, the generation of the disassembly sequence includes: comparing the three-dimensional scanning model with the reference digital file, and marking it as a non-standard variation point when the actual position of the outer shell seam is offset from the nominal position recorded in the file by more than 5mm; and generating a disassembly action sequence containing operation priority and risk notes according to the principle of first releasing electrical connections and then releasing mechanical connections, and first releasing peripheral connections and then releasing core connections.

6. The new energy battery pack disassembly process according to claim 5, characterized in that: When the intelligent actuator disconnects from the housing: For screw connections, disassembly is performed within a preset torque range. When the actual torque reaches the preset upper limit but the screw is not turned, the actuator first applies a micro-vibration with a frequency of 100Hz and an amplitude of 0.1mm in the axial direction. If this is ineffective, the tool head is replaced with a miniature hollow drill bit to drill out the screw. For sealant joints, use a constant temperature hot air knife to heat along the joint at 200±10℃ for 60 to 90 seconds to soften it, and then use a flexible scraper with force feedback to apply a separation force of 5 to 10N to peel it off.

7. The new energy battery pack disassembly process according to claim 1, characterized in that: The reinforcement learning decision module adopts an architecture that combines imitation learning initialization with online reinforcement learning fine-tuning; the real-time system state is a high-dimensional vector fused and encoded from local visual point cloud features, normalized readings of six-dimensional force / torque sensors, tool head identifiers, actuator end pose, and historical action sequence embeddings; the composite action commands include millimeter-level displacement increments, Newton-level force components, and discrete commands for tool hold or switch in the tool coordinate system.

8. The new energy battery pack disassembly process according to claim 7, characterized in that: The optimization objective of the reinforcement learning decision module is defined by a reward function, which includes: a task completion reward for successfully disassembling the target component, an efficiency reward for smooth execution of actions, a safety penalty for exceeding the threshold force or unexpected deformation of the component, and a progress reward for improved tool alignment accuracy or increased connection-to-separation ratio.

9. The new energy battery pack disassembly process according to claim 1, characterized in that: In the intelligent separation step of the internal components: For battery modules fixed with structural adhesive, a linear ultrasonic scalpel with a frequency matching the adhesive layer characteristics is used to cut the adhesive layer. After the module frame is removed, an adaptive vacuum adsorption array gripper with independent air pressure control and micro-displacement sensor is used to extract individual battery cells at a speed of 5 to 10 mm / s. For the laser welding points of aluminum busbars, a pulsed fiber laser with a pulse energy of 20J and a pulse frequency of 100Hz is used for scanning cutting, and the damage depth of the electrode surface is controlled to be less than 0.1mm.

10. The new energy battery pack disassembly process according to claim 1, characterized in that: In the material handling and data optimization steps, the cloud-based digital twin platform uses machine learning algorithms to analyze the uploaded full-process data, discover optimized disassembly strategies under different fault or damage modes, and distribute the updated strategies to each disassembly workstation to achieve continuous adaptive optimization of the disassembly system.

Images