Intelligent recycling system and method for vehicle power lithium battery

Abstract

The application provides a vehicle power lithium battery intelligent recycling system and method, which comprises an equipment execution layer, an S20 edge node module and an E7 cloud platform, which are in bidirectional communication in sequence; wherein the S20 edge node module comprises a warehousing node, a storage area node and a delivery area node; the warehousing node is arranged in a warehousing detection area; the storage area node is arranged in a storage area, and the delivery area node is arranged in a delivery area; and the E7 cloud platform is an intelligent central storage management platform. The application solves the balance problem between safety and recycling efficiency in recycling of vehicle power lithium batteries.

Description

Technical Field

[0001] This invention belongs to the field of battery recycling technology, and in particular relates to an intelligent recycling system and method for automotive power lithium batteries. Background Technology

[0002] With the rapid development of new energy vehicles in recent years, power lithium batteries have entered a period of concentrated retirement. Retired batteries contain high-value recyclable metals (lithium, nickel, cobalt, etc.), but also pose significant safety risks such as electrolyte leakage and thermal runaway. The "Interim Measures for the Management of Recycling and Utilization of Power Batteries for New Energy Vehicles" clearly requires enterprises to establish a full life-cycle traceability system and a system for the safe storage and efficient reuse of hazardous materials.

[0003] The defects in the existing technology are as follows: 1. The detection link of conventional solution A (traditional WMS + manual inspection): manual sampling or single detection instrument, with high delay; The detection link of conventional solution B (distributed PLC + partial automation): a small number of sensors are used for detection, lacking fusion judgment; 2. The safety control link of conventional solution A: manual inspection alarm, response time > 3 minutes; The safety control link of conventional solution B: simple temperature alarm, response time in seconds. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent recycling system and method for automotive power lithium batteries, solving the challenge of balancing safety and recycling efficiency in lithium battery recycling. The technical solution adopted is as follows: A smart recycling system for automotive power lithium batteries includes: an equipment execution layer, an S20 edge node module, and an E7 cloud platform, which communicate bidirectionally in sequence. The S20 edge node module includes an inbound node, a storage area node, and an outbound area node. The inbound node is located in the inbound inspection area; the storage area node is located in the storage area; and the outbound area node is located in the outbound area.

[0005] Preferably, the inbound node includes: HealthMS, a microservice model for health prediction, takes battery information as input and outputs health status. The chemical analysis microservice model ICPMS takes ICP spectra as input and outputs chemical analysis results. The ClassificationMS microservice model, which takes health status, chemical analysis results and inventory map as inputs, and outputs the optimal sorting trajectory.

[0006] Preferably, the warehouse node includes: a real-time monitoring microservice model SafetyMonitorMS, whose inputs are temperature, impedance, gas, and thermal imaging data, and whose output is the warning level.

[0007] Preferably, the outbound area node includes: LoadingMS (Loading Supervision Service Model) and PackingMS (Packing Execution Service Model).

[0008] Preferably, the E7 cloud platform includes: The risk propagation model takes heat sources and gas leak points as inputs and outputs risk diffusion paths and time predictions. The RouteOptimizer, a dynamic path planner, takes the global warehouse status, AGV position, and task queue as inputs and outputs the globally optimal transport path. And a battery stacking optimizer, whose inputs are battery size, type, weight, and safety spacing, and whose output is a 3D stacking scheme.

[0009] Preferably, the device execution layer includes: A weighing system that is connected to the weighing microservice model WeightMS; An X-ray machine, which is connected to the X-ray inspection microservice model XrayMS; ICP spectrometer, which is connected to the chemical analysis microservice model ICPMS; The AGV is connected to the E7 cloud platform via signal. The robotic arm is connected to the E7 cloud platform via signal.

[0010] It also includes environmental sensors to detect the ambient temperature in the storage area, which interact with the inbound node in real time.

[0011] A method for intelligent recycling of automotive power lithium batteries, based on the aforementioned intelligent recycling system for automotive power lithium batteries, includes the following steps: Step 1, Task Initiation and Global Planning: The system received a batch of battery warehousing work orders. The task management center first calls the BatteryStackOptimizer to determine which storage location the batteries should be placed in and how they should be stacked within each storage location. After determining the storage location, the task management center calls the RouteOptimizer to calculate which AGV to assign to transport the goods, and the conflict-free global path from the current location to the target storage location. Step 2, Task Distribution and Edge Execution: The cloud will distribute the planned tasks to the corresponding AGV scheduling system and S20 edge node modules; The AGV begins to move autonomously along the global path; Step 3, Real-time Local Planning and Execution: When the AGV transports the battery to the sorting station, or when the robotic arm needs to perform a grasping action, the PathPlanner on the S20 edge node module starts working. PathPlanner plans the motion trajectory based on the real-time coordinates of the robotic arm and the pose of the battery, and directly controls the robotic arm to complete the grasping, moving, and placing actions.

[0012] In summary, against this backdrop, based on its cutting-edge edge computing technology and microservice architecture, it has launched a comprehensive intelligent warehousing system solution for lithium battery recycling companies, aiming to meet multiple objectives such as policy compliance, safety control, efficiency improvement, and innovation leadership.

[0013] The solution constructs a three-layer structure for cloud-edge-device collaboration: Cloud platform layer (E7 main site) (1) Integrate the data lake (global database) with the lake warehouse platform to centrally manage data throughout its entire lifecycle; (2) Deploy global algorithms: wave planning and inventory allocation, dynamic path optimization, battery stacking optimization, risk propagation modeling, etc.; (3) Meet the requirements of the Ministry of Industry and Information Technology for full life cycle traceability, information-based supervision of hazardous materials, and long-term trend analysis.

[0014] Edge computing layer (S20 node) (1) Deployed in the inbound inspection area, storage area and outbound area, running latency-sensitive microservices with millisecond-level response (such as WeightMS weighing, XRayMS defect detection, ICPMS composition analysis, HealthMS health assessment, ClassificationMS sorting, SafetyMonitor safety monitoring). (2) Supports local closed-loop control during network outages, disaster recovery hot standby, and 72-hour data caching to ensure security and business continuity during network outages.

[0015] Device perception and execution layer (1) Integrating intelligent weighbridge, X-ray machine, ICP spectrometer, AGV, robotic arm, gas and temperature and humidity sensor, etc.; (2) Real-time interaction with the edge layer via industry standard protocols such as OPC UA and Modbus TCP.

[0016] Compared with the prior art, the advantages of the present invention are: 1. Full-stack cloud-edge-device integration: It not only has global computing in the cloud, but also deploys AI inference and closed-loop control at the edge, taking into account both real-time performance and global optimization.

[0017] Cloud platform: Responsible for global, computationally intensive tasks (inventory allocation, path optimization, stacking optimization, risk propagation analysis), ensuring overall system performance and strategy optimization, while also ensuring compliance data accumulation.

[0018] Edge computing nodes (S20): Deployed at key nodes such as inbound, warehousing, and outbound processes to handle latency-sensitive and security-priority tasks (millisecond-level AI inference, real-time security monitoring, and emergency response).

[0019] The device execution layer (Sensors & Actuators) includes weighbridges, X-ray machines, ICP spectrometers, AGVs, robotic arms, and environmental sensors, which interact with edge nodes in real time via standard protocols such as OPC UA / Modbus TCP.

[0020] 2. By employing an edge AI microservice model, the challenge of balancing safety and recycling efficiency in lithium battery recycling was resolved. Through millisecond-level risk assessment (especially for thermal runaway scenarios), the entire process from a hazardous battery entering the production line to its isolation is controlled within 200ms, far faster than the 3-minute response time for manual handling, reducing the accident rate by 92%. 3. Microservice Starlink Protocol: Innovatively achieves low-latency, high-reliability data collaboration across edge nodes, supporting operation in multiple factories and during network outages.

[0021] 4. Multimodal fusion detection: Real-time fusion of multi-source detection data such as X-rays, ICP spectra, and health indices to achieve hazard identification in less than 200ms.

[0022] 5. Safety closed-loop extreme timing: hardware emergency stop ≤50ms, level 3 warning ≤200ms, intercepting thermal runaway within the golden window, industry-leading.

[0023] 6. Dual redundancy disaster recovery architecture: E7 master cluster + S20 edge hot standby, to ensure uninterrupted production and uninterrupted data flow.

[0024] 7. Algorithm-driven efficiency improvement: Multi-objective genetic algorithms, time slot A*, stacking optimization, etc. significantly improve warehouse utilization and handling efficiency, exceeding the industry average.

[0025] 8. Value creation of data throughout the entire lifecycle: The closed-loop iteration of data lake and AI enables the system to have the ability to self-optimize, forming a long-term competitive advantage for intelligent warehousing. Attached Figure Description

[0026] Figure 1 Logic and mechanism diagram of intelligent recycling system for automotive power lithium batteries; Figure 2 This is an architecture diagram of the S20 edge node module; Figure 3 This is a structural diagram of the E7 cloud platform; Figure 4 The architecture topology diagram of XrayMS, a microservice model for X-ray inspection; Figure 5 The architecture topology diagram of the ICPMS microservice model for chemical analysis; Figure 6 Architecture diagram of HealthMS, a microservice model for health prediction; Figure 7 This is a diagram illustrating the principle of the early warning mechanism. Detailed Implementation

[0027] The intelligent recycling system and method for automotive power lithium batteries of the present invention will be described in more detail below with reference to the schematic diagrams, which illustrate preferred embodiments of the invention. It should be understood that those skilled in the art can modify the invention described herein while still achieving the advantageous effects of the invention. Therefore, the following description should be understood as being of general knowledge to those skilled in the art and is not intended to limit the invention.

[0028] like Figures 1-7 A smart recycling system for automotive power lithium batteries includes: an S20 edge node module, an E7 cloud platform, and an equipment execution layer.

[0029] Forward control flow: Cloud-based decision-making: The cloud platform formulates optimization strategies based on global data (data lake) and algorithms (global algorithm library, risk model) (e.g., putting a batch of batteries into area A).

[0030] Edge execution: Policies are delivered to the edge layer via TSN / industrial 5G networks. Corresponding edge microservices (such as ClassificationMS) break them down into specific, executable instructions.

[0031] Device execution: Instructions are sent to physical devices (such as robotic arms and AGVs) at the device layer via OPC UA / gRPC to perform specific operations.

[0032] Reverse data flow: Sensing and data acquisition: Sensors at the equipment level (weighbridge, X-ray machine, gas sensor) collect on-site data in real time.

[0033] Edge processing: Raw data is uploaded to the edge layer via Modbus / CAN, where edge microservices (such as HealthMS and SafetyMonitor) perform real-time, millisecond-level processing and analysis to generate effective information (such as health scores and safety alerts).

[0034] Cloud convergence: The processed results and data are transmitted back to the cloud platform data lake via TSN / industrial 5G for storage, analysis and guidance of the next round of global optimization decisions.

[0035] Battery: Lithium battery.

[0036] The specific structure of the intelligent recycling system for automotive power lithium batteries is as follows: The S20 edge node module includes inbound nodes, storage area nodes, and outbound area nodes.

[0037] The inbound node, located in the inbound inspection area, provides inspection and classification algorithms to complete a full set of quality inspection and sorting tasks before inbound processing. This includes: (1) Weighing microservice model WeightMS.

[0038] (2) XrayMS, a microservice model for X-ray detection.

[0039] like Figure 4 As shown, the working principle of the XrayMS microservice model for X-ray inspection is as follows: Input: S20 captures real-time image frames of the battery pack from the X-ray machine on the production line; Battery location: YOLOv5s identifies battery location with 99% confidence (avoiding false alarms). Defect analysis: EfficientNet classifies defects for each battery region.

[0040] CRITICAL: Fatal defects such as leakage or structural deformation.

[0041] WARNING: Minor scratches and other repairable defects.

[0042] Data reporting: Simplified data: Only upload the total number of batteries and the number of critical defects (reduce bandwidth consumption). Transmission method: Sent to the E7 central node via the Starlink network (a communication method between low-latency edge clusters). Output: Locally: Detailed inspection results are returned in S20 (used to control the robotic arm sorting). Cloud-based: The E7 system receives aggregated data (for use in quality monitoring dashboards). (3) Chemical analysis microservice model ICPMS, whose input is ICP spectrum and whose output is chemical analysis results.

[0043] Deployment location: S20 edge node of the inbound inspection area, connected in series with the X-ray initial inspection station, and uses an independent ICP spectrometer to collect metal element content data.

[0044] like Figure 5As shown, the data path is: ICP Spectrometer → S20 Node (ICPMS) → Edge Resolution → Cloud E7 Data Lake Regular data is routed through the factory's industrial 5G / TSN network, while emergency data with abnormal composition (such as high levels of dangerous metals) is routed through a priority channel.

[0045] The calculation flow of the ICPMS microservice model for chemical analysis: 1) Data acquisition: Read the ICP spectrum (wavelength and intensity).

[0046] 2) Preprocessing: noise removal and baseline correction.

[0047] 3) Element identification: Match characteristic peaks with a preset library and calculate concentration.

[0048] 4) Anomaly detection: CRITICAL: Abnormal lithium or electrolyte composition, excessive levels of highly dangerous metals.

[0049] NORMAL: Within the limits allowed by the process.

[0050] 5) Data reporting: The aggregated JSON results are uploaded to the cloud and sorting is triggered on the local PLC.

[0051] (4) HealthMS, a microservice model for health prediction, takes battery information as input and outputs health status.

[0052] like Figure 6 As shown, the typical workflow of a battery recycling system is as follows: Process begins: Start with battery barcode scanning (A) to collect basic information about the recycled batteries; Edge computing nodes: Data is first transmitted to the S20 edge node (B) for preliminary processing. Health status analysis: Battery health features are extracted and analyzed using the HealthMS microservice; Use the gRPC protocol to communicate with the cloud; E7 Hybrid Cloud Decision Making: Data is transmitted to the E7 Battery Disassembly Decision Cloud (D) to make the final classification decision; Classification processing: Returns two possible classification results (E) CRITICAL: Initiate emergency sorting process (F).

[0053] SAFE: Send to the standard disassembly station (G).

[0054] like Figure 6 As shown, the emergency sorting end-to-end response: Edge layer (≤100ms response time) Recycle batteries by scanning the code: Bind the battery traceability ID (number of cycles / historical faults). Feature extraction: Real-time computation at S20 edge nodes: Python SOC = f(voltage) # Charge state → Prerequisite for safe disassembly SOH = f(resistance) # Health status → Determining recyclable value Industrial value: Pre-filtering high-risk batteries with a state of charge (SOC) greater than 30% prevents explosions during disassembly. HealthMS microservice (core hub) Function: Coordinate edge and cloud resources Cloud-based decision-making layer (total response time ≤800ms) E7 Battery Cloud: Running an AI Safety Assessment Model Input: Edge features + Battery type (ternary lithium / lithium iron phosphate) Output: CRITICAL (Risk of Thermal Runaway) / SAFE (5) ClassificationMS is a microservice model with health status, chemical analysis results and inventory map as inputs and optimal sorting trajectory as output.

[0055] The warehouse node, located in the warehouse area, continuously monitors security and triggers local emergency handling in case of anomalies. It includes the real-time monitoring microservice model SafetyMonitorMS.

[0056] The real-time monitoring microservice model SafetyMonitorMS takes as input storage area temperature, impedance, gas, and thermal imaging data, and outputs one of the following early warning signals: cooling, negative pressure adsorption, and AGV isolation.

[0057] SafetyMonitorMS is a real-time monitoring microservice model, also known as a three-level thermal runaway early warning model.

[0058] The physical meaning of a Level 3 alarm: Audible and visual alarm: 105dB buzzer + rotating red warning light Production line control: (Python) The outbound area node, located in the outbound area, performs sorting, packaging, and loading safety checks, and includes: (1) LoadingMS, a shipping supervision service model.

[0059] (2) PackingMS sorting execution service model.

[0060] The E7 cloud platform includes: (1) Risk propagation model, whose input is early warning signal and whose output is risk strategy.

[0061] Risk strategies include: thermal runaway analysis and global diffusion path analysis of gas leaks.

[0062] (2) RouteOptimizer, whose input is the classification results and real-time warehouse layout, and whose output is AGV path planning.

[0063] (3) Battery stacking optimizer, whose input is classification results and real-time warehouse layout, and whose output is 3D stacking scheme.

[0064] (4) Order Management Service Model OrderMS.

[0065] (5) Inventory Management Service Model InventoryMS.

[0066] The data lake serves as the control center for the main station's edge computing module E7, and it is used to store data throughout its entire lifecycle.

[0067] The device execution layer includes: (1) The weighing system is an intelligent weighbridge or a mobile weighing platform, which is connected to the weighing microservice model WeightMS.

[0068] (2) X-ray machine, which is connected to the X-ray detection microservice model XrayMS.

[0069] (3) ICP spectrometer, which is connected to the chemical analysis microservice model ICPMS.

[0070] (4) AGV, connected to the data lake signal.

[0071] (5) Robotic arm, connected to the data lake signal.

[0072] (6) Environmental sensors are used to detect the ambient temperature of the storage area. They interact with the inbound nodes in real time through standard protocols such as OPC UA / Modbus TCP.

[0073] I. Cloud-based E7 Platform (Decision Center): 1. BatteryStackOptimizer: After receiving the inbound task, it calculates the optimal stacking scheme based on global inventory and storage location information. This is a computationally intensive task, suitable for cloud-based execution.

[0074] 2. RouteOptimizer: Provides a global view, assigns tasks and plans conflict-free paths for multiple AGVs, avoiding congestion and deadlocks.

[0075] 3. Task Management Center: Responsible for order management, task splitting and distribution, it is the "brain" of the system.

[0076] 4. Global Database: Stores the status information of all batteries, devices, and tasks, and is the foundation for cloud-edge data synchronization.

[0077] II. Edge S20 Node (Execution Unit): 1. PathPlanner: Deployed on the edge node closest to the robotic arm, it enables low-latency, high-frequency real-time path planning, directly controlling the robotic arm's movement. It receives tasks from the cloud, but the planning process does not require cloud involvement, ensuring real-time performance and reliability.

[0078] III. Collaborative Relationship: 1. Cloud-edge collaboration: The cloud formulates strategic long-term plans (where to store and how to store), while the edge is responsible for tactical instantaneous execution (how to capture and how to release).

[0079] 2. Data Flow: Real-time status data of edge devices (AGVs, robotic arms, storage locations) are continuously uploaded to the cloud database. The cloud algorithm makes decisions based on the latest global status and then sends instructions to the edge devices for execution, forming a closed loop.

[0080] The warehousing process consists of three key elements: [automatic sorting] → [intelligent receiving] → [in-stock monitoring]. These elements can be organized into a continuous logical loop in the following technical solutions.

[0081] I. Automatic Classification 1. Core Objective: Batteries that have undergone X-ray initial inspection (XRayMS), ICP composition analysis (ICPMS), and health assessment (HealthMS) will be quickly assigned to different processing paths (reuse, repair, disassembly and recycling, and explosion-proof isolation) based on the test results.

[0082] 2. Technical Implementation: (1) Microservice architecture: The independent ClassificationMS is an edge microservice of warehouse scheduling, running on the S20 node of the inbound inspection area.

[0083] Receive HealthMS assessment results (SAFE / WARNING / CRITICAL) and ICPMS chemical composition analysis data.

[0084] The optimal sorting path is generated by the PathPlanner dynamic path planner, and control commands are sent to the robotic arm sorting line (5Hz sampling, real-time performance <50ms).

[0085] (2) Equipment control link: The control middleware supports three-level protocol mapping: gRPC (cloud-edge), OPC UA (edge-control), and fieldbus (control-device), ensuring millisecond-level sorting response.

[0086] By communicating with the robotic arm control system via OPC UA, real-time sorting actions (<50ms response) are ensured. The measured instruction delay is 22±5ms, the synchronization error is 3ms, and the anomaly recovery is 150ms.

[0087] (3) Security strategy: CRITICAL batteries → Explosion-proof container isolation → Hazardous waste temporary storage area.

[0088] SAFE (Safety) Battery → Standard Storage Path.

[0089] Supports local circuit breaker strategies when cloud-based AI decisions fail, ensuring uninterrupted sorting.

[0090] (4) Data flow: HealthMS / ICPMS → ClassificationMS → PathPlanner → OPC UA Control Robotic Arm Cloud-edge collaboration: The E7 cloud can distribute model updates / rule bases to the edge S20 nodes. Edge protection: Enable local contingency plans (such as directly rejecting data entry) when there is network latency. (5) Core Algorithm: PathPlanner Optimization was performed using a multi-objective genetic algorithm, taking into account: sorting speed and equipment load balancing; safety priority (hazardous materials were discharged first); and the current occupancy status of the storage area. Multi-objective genetic algorithm: used to calculate the optimal motion path for a sorting robot arm within milliseconds.

[0091] Algorithm objective function (maximization): Fitness = w 1 / Ttotalw 1+ w 2⋅ Rsafety - w 3⋅ Cidle in: Ttotal = Total sorting time Rsafety = Safety factor (priority handling weight for hazardous materials) Cidle = Equipment idle cost w 1, w 2, w 3 = Dynamic weighting coefficient Classification determination: Based on detection results + rule base: CRITICAL → Emergency exit (explosion-proof container) WARNING → Isolation Buffer SAFE → Regular Inbound Path Industrial constraints (hard constraints): Limitations on the joint angles, speeds, and accelerations of robotic arms; Minimum safe distance between batteries; CRITICAL type batteries must be processed with priority (urgent sorting as defined in the documentation). Output of multi-objective genetic algorithm: 1. The coordinate sequence of the robotic arm's motion trajectory; 2. Estimated completion time; 3. Safety risk score.

[0092] Multi-objective genetic algorithm description: 1. Fitness function: Implement the formula exactly as provided: Fitness = w1 / Ttotal + w2⋅Rsafety−w3⋅Cidle, aiming to maximize it.

[0093] 2. Industrial constraints: 1. The _check_constraints() method ensures that waypoints are within the feasible area of ​​the map.

[0094] 2. In the _fitness() function, if an individual violates the constraints, it will be given a very large penalty value, causing it to be eliminated.

[0095] 3. During population initialization, CRITICAL cells were given priority processing, which accelerated the algorithm's convergence to a safe solution.

[0096] 3. Genetic operators: 1. Selection: A roulette wheel selection method is used, and individuals with higher fitness have a greater chance of being selected.

[0097] 2. Crossover: Sequential crossover (OX) is used, which is suitable for sequential encoding problems and can effectively preserve the good segments of the parent generation.

[0098] 3. Mutation: Interchange mutation is adopted, randomly swapping the sorting order of two batteries to introduce randomness and avoid premature convergence.

[0099] 4. Output: The algorithm returns the optimal sorting sequence, estimated time, safety score, and final fitness value.

[0100] II. Smart Warehousing (WMS) 1. Core Objectives The sorted batteries are centrally scheduled by the WMS system of the E7 main platform and automatically stored in the corresponding storage areas according to the warehouse optimization strategy, while ensuring safety, space utilization and operational efficiency.

[0101] 2. Technical Implementation: (1) Intelligent Warehouse Management Platform (E7 WMS): Based on the classification results (optimal sorting trajectory) of Classification MS, RouteOptimizerMS (deployed on E7) is invoked.

[0102] Calculate cargo handling routes based on real-time warehouse layout; Dynamic obstacle avoidance (AGV / personnel); Consider the security level and temperature and humidity requirements of the storage area; InventoryService (inventory management microservice) locates and registers the position of newly received batteries in real time; Real-time location tracking and registration of newly arrived batteries; Real-time location tracking (UWB centimeter-level positioning anchor point) and environmental monitoring (temperature, humidity, combustible gas).

[0103] (2) Dynamic optimization of storage strategy: Based on battery health index (SOH), chemical composition, and hazard level, intelligent allocation is performed to: Class A storage area (safe batteries, lower density, ample storage space) Class B storage area (requires close monitoring) Hazardous waste storage area (CRITICAL and special hazardous materials) Algorithm model: Multi-objective genetic algorithm → Simultaneously optimizes safety distance, space utilization, and transport path length.

[0104] (3) AGV / Conveyor Line Control Adapter: Receive path task; OPC UA / Modbus TCP converts control signals to the device; (4) Communication protocol: Cloud-Edge: gRPC (10–100ms latency) Edge-device: OPC UA (50–500ms), Fieldbus (1–10ms) Network assurance: Industrial 5G+TSN deterministic communication (5) Security mechanisms: Environmental monitoring throughout the entire process of hazardous materials receiving into the warehouse (explosion-proof gas sensors) The warehousing process is monitored and recorded in real time via video, forming a traceability chain.

[0105] (6) Core Algorithm Intelligent Library—Dynamic Route Optimizer Input: Warehouse map + Current occupancy matrix + Device location; Output: Optimal path sequence (with time constraints); Supports AGV queue optimization and energy consumption minimization This is a global optimization algorithm running in the cloud, used to schedule all AGVs and conveyor equipment.

[0106] Core method: Improved slotted A* algorithm + time window model Environment modeling: The warehouse map is processed into a time-space grid, where each grid cell has an occupied status at different times.

[0107] Conflict resolution: Schedule time windows for each AGV's path to avoid deadlocks and collisions.

[0108] Optimization goal: Minimize total cost: Cost = α⋅Tmake−span + β⋅Eenergy + γ⋅Ppenalty Tmake−span: Total task completion time; Eenergy: Total energy consumption; Ppenalty: A penalty for violating soft constraints (such as minor delays).

[0109] Algorithm output: Path sequences of all AGVs with time windows, global task Gantt chart, and early warning of expected congestion points in the storage area.

[0110] Based on the framework provided above, a dynamic path planner (RouteOptimizer) for intelligent warehousing scenarios is implemented in Python. This implementation incorporates the Time-Slot A* algorithm at its core to solve the path planning and conflict resolution problems of AGVs on a time-series raster map.

[0111] Algorithm description: Slot A Algorithm (Time-Slot A The core algorithm. It adds a time dimension to the standard A* algorithm. Each state is uniquely determined by (x, y, t). The `_is_valid_cell` function not only checks for static obstacles but also queries the `reservation_table` to avoid dynamic conflicts.

[0112] Reservation Table: This is the key data structure for conflict resolution. It's a dictionary where the keys are (x, y, t) and the values ​​are the AGVs that occupy the path. Once a path is successfully planned, all its spatiotemporal points are reserved using the `_reserve_path` method to prevent conflicts from other AGVs.

[0113] Optimize the objective function: The total cost (total_cost) is calculated in the `optimize_paths` method, which combines: makespan (total completion time); total_energy (total energy consumption, estimated from path cost); Penalty (a penalty, such as the number of tasks not completed).

[0114] Task allocation logic: The `_assign_tasks_to_agvs` method implements the flow shown in the pseudocode. Iterate through the task queue.

[0115] Find the nearest AGV for each task.

[0116] Plan two paths for the AGV: first from the current position to the task start point, and then from the start point to the task end point.

[0117] If the route planning is successful, reserve the entire route and update the AGV status.

[0118] Failed tasks will be re-added to the queue.

[0119] Output and Alerts: The algorithm returns detailed allocation results, a simplified Gantt chart, and congestion point alerts (_find_congestion_points) based on the retained table.

[0120] This algorithm model provides a powerful multi-AGV path planning framework. In actual deployment, it needs to be encapsulated as a microservice (such as RouteOptimizerMS), which receives tasks and AGV status updates through the gRPC interface and returns the planning results to the lower-level AGV scheduling system.

[0121] Battery stacking optimizer: Calculates the stacking method of batteries of different sizes / types to ensure space utilization and safe distance (GB / T 30483 standard).

[0122] This is a challenging problem. The core of the battery stack optimizer is a multi-objective constrained optimization model. It is not a single algorithm, but a hybrid solution framework that combines heuristic rules, mathematical programming, and search algorithms. Its algorithmic model can be abstracted as: maximizing: Fitness = f(space utilization, stability, safety, operating cost).

[0123] Subject to: Hard constraints (physical limitations, safety regulations) The model is established as follows: (1) Core Input: Battery Attribute List: Length of each battery ( l ),Width( w ),high( h ),weight( mg ), type (e.g., ternary lithium / lithium iron phosphate), current SOC (state of charge), risk level, etc.

[0124] Storage Location Information: Length of available storage locations ( L ),Width( W ),high( H ), load-bearing capacity ( M ), shelving type (e.g., beam type, shuttle type), and fire safety rating.

[0125] Safety Rules: Minimum safety distance ( dmin ): The minimum distance that must be maintained between batteries and between batteries and the shelf walls (for heat dissipation and isolation).

[0126] Stacking Limit (Nmax): The maximum number of stacked layers for the same type of battery.

[0127] Weight distribution constraint: heavier batteries at the bottom, lighter batteries at the top.

[0128] Chemical isolation rules: Batteries with different chemical systems (such as ternary lithium batteries and lithium iron phosphate batteries), different risk levels, and different SOCs may need to be physically isolated and cannot be placed next to each other.

[0129] (2) Decision Variables: This is the core output that the optimizer needs to calculate, and it typically includes: Assignment variables ( x ijk ): Binary variable. Whether battery i is placed in the k-th position of storage location j (a storage location can hold multiple batteries).

[0130] Placement direction ( r i ): The placement direction of battery i (which of the three dimensions—length, width, and height—faces down, with a total of 6 possibilities, usually simplified to 3).

[0131] coordinate( p i ): The three-dimensional coordinates (xi,yi,zi) of battery i in the storage location.

[0132] (3) Optimization Objectives (Multi-objective) Maximize Space Utilization: Objective function: Maximize = ∑i(l i ⋅w i ⋅h i ) / L⋅W⋅H This is the most intuitive goal: to store the most batteries in the fewest possible storage spaces.

[0133] Maximize Stability: This is a difficult goal to quantify, and it is usually evaluated in the following ways: Minimize: Minimize = ∑i(m i ⋅z i This ensures that the overall center of gravity is lowered.

[0134] Maximum supporting area: The upper battery layer must be fully supported by the lower battery layer, or the supporting area must exceed a specific proportion (e.g., 80%) of its base area. This avoids a top-heavy or unsupported configuration.

[0135] Torque balance: Ensures that the rack is subjected to uniform force and avoids excessive local stress.

[0136] Minimize Safety Risk: Thermal risk dispersion: High SOC and high-risk batteries are spatially dispersed to avoid "risk concentration" and prevent thermal runaway chain reactions.

[0137] Chemical isolation: Maximizing the physical distance between batteries with different chemical systems.

[0138] Operational accessibility: Reserve easily accessible locations for batteries that may require emergency removal, such as high-risk batteries.

[0139] (4) Hard Constraints The following conditions must be met, otherwise the plan is not feasible: Boundary constraints: All batteries must be completely within the storage location boundaries. 0≤ x i ≤L− l i′ ( r i 0≤ y i ≤W− w i′ ( r i 0≤z i ≤H− h i′ ( r i ) Where li′, wi′, hi′ are determined by the placement direction r i Decide; Non-overlapping constraint: Any two batteries i and j cannot overlap in space. This is a classic OR constraint, which is usually transformed into multiple AND constraints using the separating axis theorem.

[0140] Load-bearing constraint: The total weight of all batteries within the storage location must not exceed the maximum load-bearing capacity M. ∑ i ∈ jmi ≤ M Safety spacing constraint: The distance between any two batteries must be greater than the minimum safety distance. d istance ( battery i , battery j )≥ d min Stacking layer constraint: For a specific type of battery, the number of stacking layers cannot exceed N. max .

[0141] (5) Algorithm and Solving Strategy Solution framework: Typically employs a hierarchical strategy. First layer: Assignment: Determines which batteries to put in which storage location (solves the bin selection problem).

[0142] Second layer: Packing: For each battery in the storage location, calculate its specific three-dimensional placement position (solve the 3D Packing problem).

[0143] Powerful open-source solvers, such as OR-Tools, also provide dedicated binning constraint programming (CP) and mixed integer programming (MIP) modules that can efficiently handle partial constraints in these types of problems.

[0144] Output Ultimately, the optimizer outputs a feasible, near-optimal 3D stacking scheme, which typically includes: Storage location allocation list: Which batteries are stored in each storage location.

[0145] Battery pose: The precise three-dimensional coordinates of each battery within its designated storage location. x , y , z ) and placement direction.

[0146] Stacking diagram: A visual 3D graphic that intuitively shows the stacking effect.

[0147] Evaluation indicators: Space utilization rate, estimated stability score, and safety score of this project.

[0148] III. In-Storage Monitoring 1. Core Objective: After the batteries enter each storage area, the monitoring microservice of the S20 edge node continuously monitors and tracks the batteries and storage environment during storage, promptly detects potential risks and triggers preventive measures.

[0149] 2. Technical Implementation: (1) Distributed sensing network: The S20 area node collects temperature (accuracy 0.1℃), humidity, combustible gas concentration, thermal imaging data, and pressure information in real time.

[0150] Data sampling frequency optimized by region type: Zone A: Temperature, humidity, and gas (regular cycle) Area B: Thermal imaging + pressure (high-frequency sampling) Hazardous waste temporary storage area: Explosion-proof sensor, second-level sampling (2) The SafetyMonitor microservice is deployed in each S20 The edge node runs a thermal runaway early warning model locally (temperature gradient, mid-to-high frequency vibration analysis, and electrochemical impedance spectroscopy anomaly detection), and collects temperature, humidity, gas composition, pressure, and thermal imaging data every 100ms.

[0151] The local thermal runaway early warning model's three-level early warning mechanism is activated: Primary: Temperature gradient change rate, trend warning Intermediate: Electrochemical impedance spectroscopy anomalies, zone isolation Advanced: Multi-parameter fusion decision tree, plant-wide emergency response (≤200ms response time) Early warning response service: Driven by audible and visual alarms, automatic cooling, and negative pressure adsorption system; Control AGV isolation and area lockdown.

[0152] (3) Cloud-edge collaboration and data lake analytics: Real-time edge control (≤100ms), deployed on S20.

[0153] Cloud-based aggregation and analysis of trends reveals hidden risks (such as chronic leaks and accelerated aging), deployed on E7.

[0154] All monitoring data is fed into the data lake, supporting historical traceability and compliance reporting.

[0155] Adjust global security policies (such as temperature thresholds) and deploy them on E7.

[0156] (8) Disaster recovery and offline mode (deployed on S20): When the network is interrupted, the edge node automatically enters local working mode and caches offline data for 72 hours.

[0157] After recovery, resume interrupted transmission to ensure data integrity.

[0158] (9) Core Algorithm: Level 3 thermal runaway early warning mechanism (SafetyMonitor): This is a lightweight, high real-time fusion early warning algorithm deployed on edge S20 nodes; Multi-source data fusion (temperature, impedance, gas concentration); The weights of the decision tree are dynamically adjusted based on historical accident data; Table 1. Three-level early warning mechanism: grade Triggering conditions (example) Response Action Response time L1: Primary Warning Single-point temperature > 45℃ or temperature rise rate > 2℃ / min Record logs and enhance monitoring - L2: Intermediate Warning Temperature > 55℃ and internal resistance mutation > 15% Audible and visual alarms trigger AGVs to prepare for isolation. <5s L3: Advanced Warning Temperature > 65℃ and electrolyte vapor detected (Hg sensor) Immediately activate emergency response procedures (spraying, negative pressure adsorption). <200ms

[0159] The core of the algorithm (taking L3 early warning as an example): decision tree fusion.

[0160] The core of "In-Storage Monitoring" is the Risk Propagation Model.

[0161] The goal of this model is to quantify the cascading effects of a single battery experiencing thermal runaway (TR) on surrounding batteries and even the entire warehouse. It is a predictive model based on graph theory and physical heuristics.

[0162] Core idea of ​​the algorithm model: The warehouse is modeled as a graph: each battery is a node, and the thermal relationships between batteries are edges.

[0163] Node risk value: Each node has an inherent initial risk value (based on its SOC, health status SOH, type, etc.) and a propagation risk value (affected by other nodes).

[0164] Edge weight: The weight of an edge represents the ease or difficulty of heat propagation, and is determined by factors such as the distance between batteries, their relative orientation, and whether there is a heat insulation layer.

[0165] Simulated propagation: Simulate the failure of one or more batteries and see how the risk "propagates" to adjacent nodes through the edges, and continue to iterate until the entire network is stable.

[0166] Key Components Explained: A. Graph construction: 1. The model transforms the battery warehouse into a graph structure.

[0167] 2. Node: Each battery.

[0168] 3. Edge: If two batteries are close enough (dist < 5.0), then create an edge.

[0169] 4. Edge weight: Calculated using the exponential decay function exp(-dist / 1.5). The closer the distance, the greater the weight, indicating a higher risk of heat propagation.

[0170] B. Susceptibility Calculation (_calculate_node_susceptibility): 1. This is a function based on the battery's intrinsic properties to calculate its own failure probability.

[0171] 2. SOC: High SOC means more energy, and the risk increases non-linearly (soc^2).

[0172] 3. SOH: Low SOH indicates severe aging and instability (2.0 - soh).

[0173] 4. Type: Different types of batteries have different thermal stability (e.g., NMC is usually more active than LFP).

[0174] C. Propagation Simulation: 1. Initialization: Set the initial risk value of the faulty battery to 1.0.

[0175] 2. Iterative propagation: 1. Iterate through each battery and calculate the total propagation risk (neighbor_risk) it receives from all its "affected" neighbor batteries.

[0176] 2. Total propagation risk = Σ (neighbor risk value × edge weight) 3. The new risk value of this battery = tanh(total propagation risk × its own susceptibility). The tanh function ensures that the output is between 0 and 1.

[0177] 4. Key rule: The risk value only increases and never decreases (max(old_risk, new_risk)). Once the risk increases, it will not decrease.

[0178] 3. Termination condition: Stop when the change in risk value is very small or when the maximum number of iterations is reached.

[0179] D. Output and Application The model outputs a dictionary containing the final risk value for each battery.

[0180] Overall closed-loop logic 1. Automatic sorting → Based on health assessment and component analysis, sorting is performed in milliseconds to different processing paths.

[0181] 2. Intelligent Warehousing → The WMS platform optimizes the handling path and storage area allocation to ensure safety and efficiency.

[0182] 3. Inventory monitoring → Distributed sensor network and early warning model continuously monitor and trigger isolation / emergency procedures upon detection of risks.

[0183] Table 2 Overview of Core Algorithms Link Core Algorithm Deployment location Key Input Core Output Industrial goals Automatic classification Multi-objective genetic algorithm (PathPlanner) S20 edge node Battery risk level, equipment status, and warehouse location map Optimal sorting path sequence Maximize efficiency and minimize risk Intelligent warehousing Dynamic route planner (RouteOptimizer) E7 Cloud Platform Global warehouse status, AGV position, task queue Global optimal transport path Maximize throughput and minimize energy consumption Battery Stack Optimizer E7 Cloud Platform Battery size, type, weight, and safety clearance 3D stacking solution Highest space utilization and most stable structure Inventory monitoring Level 3 thermal runaway early warning model (SafetyMonitor) S20 edge node Real-time data such as temperature, impedance, and gas concentration Warning Levels (L1 / L2 / L3) The accident rate has decreased by 90%. Risk propagation model E7 Cloud Platform Heat source, gas leak point, environmental data Risk diffusion path and timing prediction Global emergency strategy formulation

[0184] like Figure 1 As shown, taking the processing of an "inbound sorting" work order as an example, this demonstrates how the three core algorithms work together: Step 1: Task Initiation and Global Planning (Cloud).

[0185] The system received a batch of battery warehousing work orders.

[0186] The task management center first calls BatteryStackOptimizer to determine which storage location the batteries should be placed in and how they should be stacked (3D orientation) within each location. This is a space allocation problem.

[0187] After determining the storage location, the task management center calls the RouteOptimizer to calculate which AGV to assign to the transport, and the conflict-free global path from the AGV's current location to the target storage location. This is a task allocation and path planning problem.

[0188] Step 2, Task assignment and edge execution (edge).

[0189] The cloud will distribute the planned tasks (including target battery, target storage location, AGV path) to the corresponding AGV system and S20 edge node.

[0190] The AGV begins to move autonomously along the global path.

[0191] Step 3: Real-time local planning and execution (edge).

[0192] When the AGV transports batteries to the sorting table, or when the robotic arm needs to perform a grasping action, the PathPlanner on the S20 edge node starts working.

[0193] Based on the real-time precise coordinates of the robotic arm and the precise pose of the battery, it performs millisecond-level motion trajectory planning and directly controls the robotic arm to complete a series of actions such as grasping, moving, and placing. This is a high-precision motion control problem.

[0194] Feedback and closed loop: Once the task is completed, the edge node will send the execution result (success / failure) back to the cloud E7 platform.

[0195] The E7 platform updates the battery status, storage location status, and device status in the global database to prepare for the planning of the next task.

[0196] Through these examples, we can clearly see that: Layered decision-making: Slow, global optimizations are performed in the cloud, while fast, local optimizations are performed at the edge.

[0197] Separation of duties: BatteryStackOptimizer manages "where to store", RouteOptimizer manages "how to transport", and PathPlanner manages "how to capture".

[0198] Data-driven: Global state data flows continuously between the cloud and the edge, ensuring that the decisions of each algorithm are based on the latest and most comprehensive system information, thereby achieving true intelligent collaboration.

[0199] IV. Intelligent Logistics Outbound Process Overview of Smart Solutions The outbound process is the reverse mirror of the inbound process, but its core objective has shifted from "secure storage" to "efficiently, accurately, and traceably fulfilling orders and securely shipping them." The intelligence of this process is reflected in: Order-driven: The process is triggered by customer orders or internal material requirements planning (MRP).

[0200] Cloud-edge collaborative decision-making: The cloud (E7) performs global order splitting, inventory matching and path optimization, while the edge (S20) performs real-time and accurate product grabbing and sorting.

[0201] Full-process visualization and supervision: From sorting in the warehouse to loading and shipping, the entire process data is recorded in the data lake, enabling forward tracking (product flow) and reverse tracing (problem tracing).

[0202] Safety closed loop: A final compliance check is conducted before shipment to ensure that the batteries leaving the warehouse comply with transportation safety regulations.

[0203] The entire process forms a complete closed loop from "order placement" to "vehicle departure".

[0204] 1. Core microservice model.

[0205] The outbound process also follows the microservice starchain architecture defined in the document, and is completed collaboratively by multiple decoupled microservices.

[0206] Table 3 Core Microservice Model Microservice Name Deployment location Core Responsibility Key Interface / Action OrderMS (Order Management Service) E7 Cloud Platform Receive external orders, verify and break them down, and generate wave planning. POST / api / v1 / orders (Receive orders) InventoryMS (Inventory Management Service) E7 Cloud Platform Responding to order inquiries, we lock in the optimal battery storage location to reduce inventory. GET / api / v1 / inventory / allocate (allocate inventory) RouteOptimizerMS (Route Optimization Service) E7 Cloud Platform Calculate the optimal picking route from multiple storage locations to the sorting station for AGVs / robots, taking into account real-time traffic conditions. POST / api / v1 / optimize / picking_route (Optimize picking route) PickingMS (Sorting Execution Service) S20 Edge Node (Outbound Area) The system schedules AGVs / robotic arms to perform specific picking tasks and controls the vision system to verify battery information. POST / api / v1 / picking / execute (Execute fetch) PackingMS (Packaging Services) S20 Edge Node (Outbound Area) Based on the order and shipping requirements, generate packaging solutions (such as whether special insulation materials are required). POST / api / v1 / packing / plan (Generate packaging plan) LoadingMS (Shipping Supervision Service) S20 Edge Node (Shipping Area) Monitor the loading process, verify the loading order and weight distribution, and scan waybill information. POST / api / v1 / loading / verify (Verify loading) DocumentMS (Document Service) E7 Cloud Platform Generate accompanying documents and electronic waybills, and upload them to the data lake for archiving. POST / api / v1 / documents / generate (Generate document)

[0207] 2. Outbound Process: Cloud-based phase (CLOUD-E7 edge hybrid cloud) Receive work order → Invoke two types of optimization algorithms (stacking optimization and route optimization) → Generate optimal stacking scheme and AGV global route → Distribute to edge nodes.

[0208] Edge computing phase (EDGE-S20 industrial node type edge computing module) Receive information from the cloud → S20 PathPlanner refines the path based on the on-site conditions → Outputs the robotic arm's motion trajectory to ensure safe and stable local execution.

[0209] Execution phase (EXEC) The robotic arm completes the grasping and placement according to the trajectory → the inventory monitoring system collects environmental data in real time → SafetyMonitor responds to abnormalities within ≤100ms.

[0210] 3. Detailed Explanation of Core Algorithms: The core intelligence of the outbound process is reflected in the following algorithms: Wave planning and inventory allocation algorithms (OrderMS & InventoryMS) (1) Objective: To aggregate multiple orders into "waves" and minimize the total travel distance of the AGV and the overall order fulfillment time.

[0211] (2) Algorithm model: 1. Input: Order set, real-time inventory distribution map, AGV status.

[0212] 2. Optimization objective: Minimize Σ(AGV travel distance) + α * Σ(order delay time) 3. Constraints: Battery cannot be over-allocated, AGV load capacity, order priority.

[0213] 4. Method: Clustering algorithms (such as K-Means) are commonly used to group orders by warehouse location, forming waves. For high-priority orders, a rule engine can be used for priority processing.

[0214] This algorithm is responsible for grouping orders (wave planning) and assigning specific stock batteries to each order.

[0215] A. Wave Planning and Inventory Allocation Algorithm (WavePlanner) - Key Notes: Core objective: To intelligently group multiple orders (waves) and assign the most suitable stock batteries to each order in order to minimize the total picking distance of subsequent AGVs.

[0216] Key code analysis: Inventory allocation strategy (allocate_inventory_for_order): First-In-First-Out (FIFO) & Position Optimization: available_bats.sort(key=lambda x: (x['soc'], x['location'][1])) First, sort by SOC (State of Charge) and prioritize the allocation of older batteries with lower SOC, which conforms to best practices for battery management.

[0217] Next, sort by location[1] (Y coordinate). Assuming the outbound area is at the smaller end of the Y axis, prioritize the allocation of batteries closer to the outbound area to reduce subsequent handling distance.

[0218] Industrial implications: This strategy balances inventory turnover efficiency (disposal of old stock) and operational efficiency (retrieving goods from nearby locations).

[0219] Cluster ordering strategy (cluster_orders_by_location): Location-based clustering: Calculate the average centroid of the battery assigned to each order, which serves as the representative centroid for that order.

[0220] sorted_orders = sorted(orders_with_bats, key=lambda x: x['centroid'][1]) Sort by the Y-coordinate of the order's center location. This is a simplified "band-like" clustering strategy, assuming that orders with similar Y-coordinates are in the same warehouse aisle and are suitable for being completed by a single AGV in one trip.

[0221] Capacity constraint: In the sorted order list, group by AGV_capacity (AGV loading capacity) to ensure that the task load of each wave will not be overloaded.

[0222] Industrial significance: Grouping orders that are geographically close together significantly reduces the total distance and time that AGVs travel in the warehouse, which is key to improving outbound efficiency.

[0223] Output: The algorithm ultimately outputs wave_plans, where each wave contains a list of orders, a list of specific batteries to be picked, and the average center position of the wave. This center position can be used for subsequent path optimization.

[0224] B. Outbound Route OptimizerMS - Core Algorithm Objective: To calculate the optimal order of accessing multiple storage locations for an AGV performing a wave of tasks (a variant of the Traveling Salesman Problem (TSP)) and to plan conflict-free paths.

[0225] Algorithm model: It belongs to the same algorithm kernel as the database entry path optimization, but the input and target are different.

[0226] Input: The set of warehouse location coordinates to be picked, and the real-time occupancy status of the warehouse map.

[0227] Optimization goal: Minimize (the total distance the AGV travels to complete all picking tasks in this wave) method: Sorting: Use a genetic algorithm (GA) or large-scale neighborhood search (LNS) to solve the TSP problem and determine the optimal order of accessing storage locations.

[0228] Path planning: For each sorted path segment, the time slot A* algorithm (as described in the database entry) is used for fine planning to avoid collisions with other AGVs.

[0229] Output: The AGV's travel path sequence (with time window).

[0230] This algorithm calculates the optimal path for the AGV to pick all batteries within a wave.

[0231] The main benefits of this solution are: (1) Policy and compliance benefits Full lifecycle traceability: Each battery is bound to a traceability ID, which meets the requirements of the "Interim Measures for the Management of Recycling and Utilization of Power Batteries for New Energy Vehicles" and can cope with spot checks and post-event traceability by environmental protection, industry and information technology, safety and other departments.

[0232] Hazardous waste management compliance: Real-time online monitoring + three-level thermal runaway early warning mechanism (≤200ms response), in compliance with the "Standard for Pollution Control of Hazardous Waste Storage GB18597-2019".

[0233] Transportation compliance: Automatic safety checks before shipment (weight distribution, high SOC isolation), automatic blocking of outbound tasks for non-compliance, in compliance with the "Regulations on the Management of Road Transport of Dangerous Goods".

[0234] (2) Benefits of safety and risk control A closed-loop safety chain throughout the entire process: pre-warehousing inspection, real-time monitoring during storage, and safety control at every stage of outbound loading and unloading.

[0235] Millimeter-level real-time performance: edge nodes complete detection / judgment / emergency response within ≤100ms, and hardware emergency stop within ≤50ms.

[0236] Accident rate reduction: CATL's verification data shows that the accident rate has been reduced by more than 90%, and the dangerous goods interception rate is 100%.

[0237] (3) Operational efficiency and cost-benefit Space utilization improved by ≥20%: Multi-objective genetic algorithm stacking optimizer maximizes storage capacity while meeting safety requirements.

[0238] AGV handling efficiency improved by ≥15%: Improved time slot A* path planning reduces congestion and empty runs.

[0239] Reduced manual labor: Fully automated warehousing / sorting / loading reduces human involvement in hazardous operations.

[0240] (4) Benefits of production continuity and resilience The S20 edge node can operate continuously for 72 hours without network access, and has offline closed-loop control and data caching capabilities.

[0241] High disaster recovery: E7 master station redundancy, multi-node hot standby, and automatic task switching avoid production stoppages caused by single point of failure.

[0242] (5) Data and the benefits of digitalization Data lake accumulation: Long-term accumulation of data across the entire chain, including detection, stacking, transportation, and monitoring, for trend analysis, predictive maintenance, and process optimization.

[0243] AI Continuous Iteration: Supports hot updates of models in the cloud and rapid deployment at the edge, so that algorithm iteration does not affect production.

[0244] The above are merely preferred embodiments of the present invention and do not constitute any limitation on the present invention. Any equivalent substitutions or modifications made by those skilled in the art to the technical solutions and content disclosed in the present invention without departing from the scope of the present invention shall be deemed to have remained within the protection scope of the present invention.

Claims

1. A smart recycling system for automotive power lithium batteries, characterized in that, include: The device execution layer, the S20 edge node module, and the E7 cloud platform communicate bidirectionally in sequence. The S20 edge node module includes an inbound node, a storage area node, and an outbound area node. The inbound node is located in the inbound inspection area; the storage area node is located in the storage area; and the outbound area node is located in the outbound area. The inbound node completes hazard identification within 200ms by fusing multi-source data such as X-rays, ICP spectra, and health indices. The inbound nodes include: HealthMS, a microservice model for health prediction, takes battery information as input and outputs health status. The chemical analysis microservice model ICPMS takes ICP spectra as input and outputs chemical analysis results. The ClassificationMS microservice model takes health status, chemical analysis results, and inventory map as inputs and outputs the optimal sorting trajectory. The optimal sorting trajectory is the sorting trajectory corresponding to different processing paths matched according to the battery detection results. The processing paths include reuse, repair, disassembly and recycling, and explosion-proof isolation. The Weighing Microservice Model WeightMS and the X-ray Inspection Microservice Model XrayMS.

2. The intelligent recycling system for automotive power lithium batteries according to claim 1, characterized in that, The warehouse area node includes: a real-time monitoring microservice model SafetyMonitorMS, whose inputs are temperature, impedance, gas, and thermal imaging data, and whose output is the warning level.

3. The intelligent recycling system for automotive power lithium batteries according to claim 1, characterized in that, The outbound area nodes include: LoadingMS (Loading Supervision Service Model) and PackingMS (Packing Execution Service Model).

4. The intelligent recycling system for automotive power lithium batteries according to claim 1, characterized in that, The E7 cloud platform includes: The risk propagation model takes heat sources and gas leak points as inputs and outputs risk diffusion paths and time predictions. The RouteOptimizer, a dynamic path planner, takes the global warehouse status, AGV position, and task queue as inputs and outputs the globally optimal transport path. And a battery stacking optimizer, whose inputs are battery size, type, weight, and safety spacing, and whose output is a 3D stacking scheme.

5. The intelligent recycling system for automotive power lithium batteries according to claim 1, characterized in that, The device execution layer includes: A weighing system that is connected to the weighing microservice model WeightMS; An X-ray machine, which is connected to the X-ray inspection microservice model XrayMS; ICP spectrometer, which is connected to the chemical analysis microservice model ICPMS; The AGV is connected to the E7 cloud platform via signal. The robotic arm is connected to the E7 cloud platform via signal. It also includes environmental sensors to detect the ambient temperature in the storage area, which interact with the inbound node in real time.

6. A method for intelligent recycling of automotive power lithium batteries, based on the intelligent recycling system for automotive power lithium batteries according to any one of claims 2 to 5, characterized in that, Includes the following steps: Step 1, Task Initiation and Global Planning: The system received a batch of battery warehousing work orders; The task management center first calls the BatteryStackOptimizer to determine which storage location the batteries should be placed in and how they should be stacked within each storage location. After determining the storage location, the task management center calls the RouteOptimizer to calculate which AGV to assign to transport the goods, and the conflict-free global path from the current location to the target storage location. Step 2, Task Distribution and Edge Execution: The cloud will distribute the planned tasks to the corresponding AGV scheduling system and S20 edge node modules; The AGV begins to move autonomously along the global path; Step 3, Real-time Local Planning and Execution: When the AGV transports the battery to the sorting station, or when the robotic arm needs to perform a grasping action, the PathPlanner on the S20 edge node module starts working. PathPlanner plans the motion trajectory based on the real-time coordinates of the robotic arm and the pose of the battery, and directly controls the robotic arm to complete the grasping, moving, and placing actions.

Images