A method and system for sequential planning of the dismantling of waste power batteries
By constructing a knowledge graph of power battery assembly and using the D3QN algorithm, the disassembly of power batteries is planned hierarchically, which solves the problems of low disassembly efficiency and poor process continuity in existing technologies, and realizes the generation of efficient and coherent disassembly sequences.
Patent Information
- Application Number
- CN202611279265.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-21
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies for planning the dismantling of used power batteries suffer from several problems: reliance on manual extraction constraints leading to low efficiency; global planning being prone to causing combinatorial explosions; and insufficient consideration of the costs of process switching resulting in low dismantling efficiency.
By constructing a knowledge graph of power battery assembly, the assembly sequence and spatial constraints are extracted, converted into disassembly priority constraints, a disassembly matrix is constructed hierarchically, and the D3QN algorithm is used for independent planning at each level to generate a disassembly sequence.
It realizes the automated conversion of assembly knowledge into disassembly constraints, reduces the dimensions of state space and action space, and outputs an efficient and coherent disassembly sequence, solving the problems of low disassembly efficiency and poor process continuity.
Smart Images

Figure CN123118084A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of battery recycling technology, specifically to a method and system for sequential planning of the dismantling of waste power batteries based on an assembly knowledge graph. Background Technology
[0002] With the rapid development of the new energy vehicle industry, the number of retired power batteries has surged year by year, making the recycling and harmless treatment of used power batteries a key focus of the industry. In the power battery recycling process, dismantling is a prerequisite for realizing the cascade utilization of high-value components or the recycling of materials. A well-planned dismantling sequence can effectively improve the efficiency of dismantling operations and reduce dismantling costs, making it one of the key technologies for the large-scale recycling and treatment of used power batteries.
[0003] Currently, research on dismantling sequence planning for waste products largely focuses on directly utilizing 3D models or relying on manual experience to extract dismantling constraints. However, power batteries have complex internal structures, numerous components, and multiple levels of modules, including thermal management units. Traditional methods, when dealing with such complex assemblies, typically require manual verification of spatial occlusion and assembly connections of components one by one to generate dismantling priority constraints. This approach struggles to fully integrate original product data from multiple sources, such as process documents and bills of materials, resulting in low efficiency in acquiring prior knowledge such as dismantling constraints and hindering the automatic conversion from assembly knowledge to dismantling rules.
[0004] When solving the disassembly sequence after obtaining the disassembly constraints, existing disassembly sequence planning methods typically treat all components within the battery pack as operation objects at the same level for a global search. For power batteries with a large number of components, this hierarchical global planning approach leads to an exponential increase in the state space and action space, easily causing combinatorial explosion and making it difficult for the planning algorithm to converge within a reasonable time. Furthermore, existing planning methods often limit their evaluation of sequence quality to the time consumed by a single disassembly action, rarely considering the costs of tool changes and operational posture adjustments in actual operations. This results in poor process continuity and long periods of ineffective operation time in the planned disassembly sequence during actual execution. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides a method and system for planning the step-by-step dismantling sequence of spent power batteries. This method solves the problems of existing spent power battery dismantling planning methods, such as constraint extraction relying on manual intervention, single-level global planning being prone to combinatorial explosions, and low actual dismantling efficiency due to insufficient consideration of process switching costs.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for sequential planning of the dismantling of waste power batteries, comprising the following steps: Obtain the original assembly data of power battery products; Based on the original assembly data, a knowledge graph of power battery assembly is constructed, the assembly sequence relationship and spatial constraint relationship between each node are extracted, and the assembly sequence relationship and spatial constraint relationship are converted into disassembly priority constraint. Based on the aforementioned disassembly priority constraint, the overall hierarchical disassembly matrix of the battery pack and the internal hierarchical disassembly matrix of each sub-assembly are constructed respectively. The sequential planning problem is modeled as a sequential decision process by decomposing it step by step. The state space, action space and reward function of each decomposition level are defined, and the planning model of each level is established by adopting a step-by-step independent planning strategy. The D3QN algorithm is used to solve the planning model at each level, and the decomposition sequence of each level is output. The decomposition sequences of each level are combined according to the hierarchical expansion relationship to output a complete step-by-step decomposition sequence.
[0007] Preferably, the construction of the power battery assembly knowledge graph specifically includes: Entities are extracted from the original assembly data, including the battery pack as the top-level entity, the sub-assemblies as the middle-level entities, and the base parts as the bottom-level entities. Extract the relationships between entities, including hierarchical composition relationships that connect upper and lower level entities, as well as assembly connection relationships with connection type and assembly order attributes; The extracted entity nodes and relation edges are stored in a graph database to obtain a knowledge graph of power battery assembly.
[0008] Preferably, converting the assembly sequence relationship and the spatial constraint relationship into disassembly priority constraints specifically includes: Based on the timing reversal rule, the order of assembly is reversed to a disassembly priority constraint of disassembling the last assembly first. Based on the spatial release rule, the spatial occlusion relationship between the occluded part and the occluding part is transformed into a disassembly priority constraint that prioritizes the removal of the occluding part. Based on the connection release rule, the connection between parts is converted into a disassembly priority constraint where the connector is removed before the connected part.
[0009] Preferably, the construction of the overall battery pack hierarchical disassembly matrix and the internal hierarchical disassembly matrix of each sub-assembly specifically includes: Based on the disassembly priority constraints, a battery pack overall hierarchical disassembly matrix is constructed. The matrix elements of the battery pack overall hierarchical disassembly matrix represent whether there are disassembly priority constraints between battery pack overall hierarchical objects. Construct a hierarchical disassembly matrix for each sub-assembly. The matrix elements of the hierarchical disassembly matrix for each sub-assembly represent whether there are disassembly priority constraints between the basic parts inside the corresponding sub-assembly. Establish an index mapping relationship between the overall battery pack hierarchical object and the corresponding sub-assembly internal hierarchical disassembly matrix.
[0010] Preferably, the definition of the state space, action space, and reward function for each decomposition level specifically includes: The state space is defined as a multi-dimensional vector containing the demolition state identifier of each object, the object number that performed the demolition action in the previous time step, and the cumulative demolition time. Define the action space as a set of decomposable objects that satisfy the priority constraints of all decompositions in the current state; By combining the time cost corresponding to the standard operation time of the demolition object and the process switching cost between continuous dismantling operations, a reward function for evaluating the quality of action selection is calculated.
[0011] Preferably, the step of establishing planning models at each level using a hierarchical independent planning strategy specifically includes: The overall hierarchy of the battery pack is planned to generate a disassembly sequence for the overall battery pack hierarchy; Determine the type of the current object according to the object order in the overall hierarchical disassembly sequence of the battery pack; When the object is a subassembly, the corresponding subassembly internal hierarchy disassembly matrix is called according to the established index mapping relationship, and the internal hierarchy of the subassembly is planned to generate the corresponding subassembly internal disassembly sequence.
[0012] Preferably, the step of using the D3QN algorithm to solve the planning model at each level specifically includes: For each level, an algorithm model is constructed that includes an evaluation network and a target network, and the action value function is decomposed into a combination of a state value function and an advantage function. Store state experience samples of interaction with the environment in the experience replay pool; The target action value is calculated based on samples taken from the experience replay pool. The evaluation network parameters are updated by minimizing the mean square error loss between the target action value and the estimated action value, and the decomposition sequence of each level is output.
[0013] Preferably, storing state experience samples of interaction with the environment in the experience replay pool specifically includes: Select the current demolition action based on the current state vector; Execute the current demolition action, update the demolition status flag, the object number of the demolition action performed in the previous moment, and the cumulative demolition time, and obtain the next state vector; Calculate the immediate reward using the aforementioned reward function; Data containing the current state vector, the current demolition action, the immediate reward, and the next state vector is stored as an experience sample in the experience replay pool.
[0014] Preferably, the step of combining the decomposition sequences of each level according to the hierarchical expansion relationship specifically includes: The disassembly sequence of the battery pack as a whole is used as the main sequence for traversal; When a basic component object is encountered during iteration, it is directly appended to the end of the output sequence. When a subassembly object is encountered, it is appended to the end of the output sequence. The internal decomposition sequence corresponding to the subassembly is then searched, and all elements in the internal decomposition sequence are appended to the end of the output sequence in their original order to obtain a complete sequence.
[0015] Preferably, after the combination yields the complete sequence, it further includes: Supplement the information of the disassembly objects in each step of the complete hierarchical disassembly sequence with additional annotations; The supplementary annotations include the level identifier of each disassembly step, the type of disassembly tool required, and the standard operating time. Output the complete hierarchical disassembly sequence, including step number, object name, hierarchy, disassembly tool, and standard operation time fields, according to the preset format.
[0016] The waste power battery dismantling sequence planning system includes a knowledge input module, a knowledge extraction module, a matrix construction module, a model construction module, and a sequence solution module. The knowledge input module obtains the original assembly data of power battery products; The knowledge extraction module constructs a power battery assembly knowledge graph based on the original assembly data, extracts the assembly sequence relationship and spatial constraint relationship between each node, and converts the assembly sequence relationship and spatial constraint relationship into disassembly priority constraint. The matrix construction module constructs the overall hierarchical disassembly matrix of the battery pack and the internal hierarchical disassembly matrix of each sub-assembly based on the disassembly priority constraint. The model building module models the hierarchical decomposition of the sequential planning problem as a sequential decision process, defines the state space, action space and reward function of each decomposition level, and establishes the planning model of each level using a hierarchical independent planning strategy; The sequence solving module uses the D3QN algorithm to solve the planning model at each level and outputs the decomposition sequence at each level. The sequence output module combines the decomposition sequences of each level according to the hierarchical expansion relationship and outputs the complete step-by-step decomposition sequence.
[0017] This invention provides a method for sequential planning of the dismantling of spent power batteries. It has the following beneficial effects: This invention constructs an assembly knowledge graph containing the battery pack hierarchy and assembly connection relationships. Based on preset rules such as time reversal, spatial decomposition, and connection decomposition, it automatically converts the extracted assembly sequence and spatial constraints into disassembly priority constraints. This achieves the technical effect of transforming multi-source heterogeneous assembly data into structured disassembly prior knowledge, and solves the technical problem that the extraction of traditional waste power battery disassembly constraints is highly dependent on manual experience and has low extraction efficiency.
[0018] This invention divides the battery pack into overall and sub-assembly levels based on the physical composition characteristics of the power battery. It constructs independent hierarchical disassembly matrices based on disassembly priority constraints and generates a complete disassembly sequence by adopting a step-by-step independent planning strategy. This achieves the technical effect of effectively reducing the dimensions of the state space and action space in each planning stage. It solves the technical problem that the global disassembly sequence planning is prone to combinatorial explosion and low solution efficiency due to the large number of waste battery parts.
[0019] This invention models the dismantling sequence planning as a sequential decision-making process, sets a reward function that comprehensively considers the dismantling time cost and process switching cost, and uses the D3QN algorithm with a dual network and competitive network structure to iteratively solve the model at each level. This achieves the technical effect of outputting a dismantling sequence with a shorter total time and more coherent operation steps while strictly adhering to the dismantling priority constraints. It solves the technical problem that existing dismantling planning methods do not adequately consider the losses caused by tool changes and attitude adjustments, resulting in poor continuity of actual dismantling operations. Attached Figure Description
[0020] Figure 1 This is a system module architecture diagram of the present invention.
[0021] Figure 2 This is a flowchart of the method steps of the present invention. Detailed Implementation
[0022] The technical solutions in 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.
[0023] Please see the appendix Figure 1 This invention provides a sequential planning system for the step-by-step dismantling of waste power batteries based on an assembled knowledge graph, including a knowledge input module, a knowledge extraction module, a matrix construction module, a model construction module, a sequence solving module, and a sequence output module.
[0024] The knowledge input module is used to acquire the raw assembly data of the power battery product. This raw assembly data includes product design drawings, assembly process documents, bill of materials, and technical standards. This raw assembly data describes the composition, assembly connections, and component attribute information of the power battery pack from different dimensions, providing a data foundation for the subsequent construction of a structured knowledge model.
[0025] The knowledge extraction module constructs a knowledge graph for power battery assembly based on the raw assembly data obtained from the knowledge input module. This knowledge graph uses the battery pack as the top-level node, expanding downwards layer by layer into sub-assembly nodes and basic component nodes. Through nodes and relational edges, it structurally expresses the hierarchical composition and assembly connection relationships of the power battery. On the constructed knowledge graph, the knowledge extraction module performs query and traversal operations to extract the assembly sequence relationships and spatial constraints between nodes. Based on preset transformation rules, it converts the assembly relationships into disassembly priority constraints, obtaining the disassembly sequence relationships between various components.
[0026] The matrix construction module is used to construct disassembly matrices at both the overall battery pack level and the internal disassembly matrices of each sub-assembly level, based on the disassembly priority constraints output by the knowledge extraction module. The overall battery pack level disassembly matrix describes the disassembly priority relationships between PACK-level objects, while the internal disassembly matrices of each sub-assembly level describe the disassembly priority relationships between the basic components within that sub-assembly. The values of the matrix elements indicate whether disassembly priority constraints exist between the corresponding objects.
[0027] The model building module models the sequential planning problem of dismantling spent power batteries as a sequential decision-making process. The module defines the state space, action space, and reward function for each dismantling task level. The state space represents the set of dismantled objects in the current dismantling state and the information from the previous dismantling step. The action space represents the set of dismantlable objects that currently satisfy the dismantling priority constraint. The reward function evaluates action selection based on dismantling time and process switching costs. The module employs a hierarchical independent planning strategy, first establishing a planning model for the overall battery pack level, and then establishing planning models for the internal levels of each sub-assembly.
[0028] The sequence solving module is used to solve the hierarchical decomposition sequence planning model established by the model building module using the D3QN algorithm. The module inputs the decomposition matrices of each level and the current decomposition state into the D3QN algorithm model. The agent then selects the next decomposition object from the currently allowed action set. After executing the selected action, the decomposition state is updated, and a reward value is calculated based on the changes in decomposition time and process switching caused by the action execution. This reward value is used to iteratively train the algorithm model. After multiple rounds of training, the algorithm model outputs a decomposition sequence that satisfies the decomposition priority constraint and has a better total decomposition time.
[0029] The sequence output module outputs the planning results generated by the sequence solving module. The output results include the PACK-level disassembly sequence, the disassembly sequence within each sub-assembly, and the complete hierarchical disassembly sequence obtained by combining the above sequences according to their hierarchical expansion relationships. This complete sequence covers all disassembly steps from the overall battery pack level to the level of basic components within each sub-assembly.
[0030] Please see the appendix Figure 2 Based on the above system architecture, this invention provides a method for sequential planning of the dismantling of waste power batteries, including the following steps: Step S1: Obtain the original assembly data of the power battery product through the knowledge input module. The original assembly data includes product design drawings, assembly process documents, bill of materials and technical standards.
[0031] Step S2: Construct a knowledge graph for power battery assembly through the knowledge extraction module, perform query and traversal operations on the assembly knowledge graph, extract the assembly sequence relationship and spatial constraint relationship between each node, and convert the assembly relationship into disassembly priority constraint according to the preset conversion rules.
[0032] Step S3: Based on the disassembly priority constraint, the matrix construction module constructs the disassembly matrix at the overall level of the battery pack and the disassembly matrix at the internal level of each sub-assembly.
[0033] Step S4: The stepwise decomposition of the sequential planning problem is modeled as a sequential decision process through the model building module. The state space, action space and reward function are defined, and the planning model at each level is established using a stepwise independent planning strategy.
[0034] Step S5: The D3QN algorithm is used to solve the planning model at each level through the sequence solving module. After iterative training, the decomposition sequence of each level is output.
[0035] Step S6: The sequence output module combines the decomposition sequences of each level according to the hierarchical expansion relationship to output a complete step-by-step decomposition sequence.
[0036] Step S1 obtains the original assembly data of the power battery product through the knowledge input module, providing a basic data source for the subsequent construction of the assembly knowledge graph and disassembly sequence planning.
[0037] S101, Obtain product design drawings.
[0038] The product design drawings include a 3D assembly model and 2D engineering drawings of the power battery pack. The 3D assembly model records the geometry, spatial position, and assembly fit relationships of each component, while the 2D engineering drawings indicate dimensional tolerances and technical requirements. The overall dimensions of the battery pack, the coordinates of each component's installation position, the spatial arrangement relationships between components, and assembly path information can be extracted from the product design drawings.
[0039] For the analysis of three-dimensional assembly models, those skilled in the art can read the model tree structure and mating constraint information through CAD secondary development interfaces or standard data exchange formats. The specific implementation methods are well-known technologies in this field and will not be elaborated here.
[0040] S102, Obtain assembly process documents.
[0041] The assembly process document specifies the assembly sequence, assembly method, tools used, and process parameters for each component of the power battery pack. The assembly sequence defines the order in which components are installed on the production line. Assembly methods include bolt connections, snap-fit connections, welding, and bonding. Tools used include torque wrenches, screwdrivers, and special fixtures. Process parameters include numerical requirements for tightening torque and pressing force.
[0042] The assembly sequence information and connection type information between components can be extracted from the assembly process documents. The assembly sequence information is an important basis for the subsequent derivation of disassembly priority constraints.
[0043] S103, Obtain Bill of Materials.
[0044] The bill of materials (BOM) records all the components of the power battery pack in a hierarchical table format, including material codes, material names, specifications, quantities, assembly levels, and relationships with higher-level components. The hierarchical relationship in the BOM directly corresponds to the physical assembly level of the battery pack, meaning that the battery pack contains several sub-assemblies, and each sub-assembly contains several basic parts.
[0045] By analyzing the hierarchical structure of the bill of materials, the overall composition framework of the battery pack and the relationships between its components can be clearly defined, providing a basic hierarchical classification basis for constructing an assembly knowledge graph.
[0046] S104, Obtaining technical standards.
[0047] Technical standards include national, industry, and enterprise standards related to power battery packs. These standards specify safe operating requirements for battery pack disassembly, component recycling and classification standards, and disassembly depth requirements.
[0048] Among them, the safety operation requirements include high-voltage power-off procedures, insulation testing methods, and residual charge release standards; the recycling classification standards include classification rules for materials such as battery cells, casings, wire harnesses, and circuit boards; and the dismantling depth requirements specify the hierarchical conditions for stopping dismantling operations.
[0049] This standard information provides a normative basis for defining constraints and objectives in subsequent decomposition sequence planning.
[0050] The raw assembly data obtained in S101 to S104 above describe the product information of the power battery pack from the dimensions of geometric structure, assembly process, material composition, and standard specifications, respectively. The knowledge input module integrates the above multi-source data and then passes it to the knowledge extraction module for further processing.
[0051] Step S2 constructs a knowledge graph of power battery assembly through the knowledge extraction module, and extracts disassembly priority constraints from the graph to realize the automatic conversion from assembly knowledge to disassembly knowledge.
[0052] S201, Constructing a knowledge graph for power battery assembly.
[0053] The knowledge extraction module receives the raw assembly data from the knowledge input module, extracts entities and relationships between them, and constructs a knowledge graph for power battery assembly. The entity types in the knowledge graph include three categories: battery pack, sub-assemblies, and basic parts.
[0054] The battery pack is the top-level entity, representing the complete power battery assembly. Sub-assemblies are the intermediate-level entities, including module units, thermal management units, battery management units, and structural housing units, which are assembled as independent components during the assembly process.
[0055] The basic components are the underlying entities, including bolts, nuts, brackets, connecting plates, busbars, seals, and insulation components, which are independent parts that cannot be further separated.
[0056] Once the disassembly of any basic component is completed, there are no further disassembly steps required for that component.
[0057] The relationship types in knowledge graphs include two categories: hierarchical composition relationships and assembly connection relationships.
[0058] Hierarchical relationships are represented by containment edges, which connect upper-level entities to their directly subordinate next-level entities. For example, a battery pack contains module units, and a module unit contains battery cells and busbars.
[0059] Assembly connections are represented by connection edges. A connection is a physical connection between two entities during the assembly process. The connection attribute further indicates the connection type, which includes threaded connection, snap-fit connection, welded connection, and adhesive connection.
[0060] The directionality of assembly connection is determined by the assembly sequence. That is, for threaded connections, the connected parts are placed first and then the bolts are screwed in during the assembly process, and the direction of the connection side is from the connected parts to the bolts.
[0061] The construction process of a knowledge graph includes three stages: entity extraction, relation extraction, and graph storage. In the entity extraction stage, the knowledge extraction module identifies all entities from the hierarchical structure of the bill of materials and determines the entity type.
[0062] In the relationship extraction stage, the knowledge extraction module extracts the connection relationships between parts from the assembly and mating relationships in the product design drawings, extracts the assembly sequence relationships from the assembly process documents, and transforms the assembly sequence relationships into edge attributes in the knowledge graph.
[0063] In the graph storage stage, the knowledge extraction module uses a graph database to store entity nodes and relation edges, and the storage format is compatible with the graph query language used in subsequent query and traversal operations.
[0064] The selection of graph databases and data import methods can be made by those skilled in the art based on the data scale and query efficiency requirements. These are well-known technologies in the field and will not be elaborated here.
[0065] S202 performs queries and traversals on the assembled knowledge graph.
[0066] The knowledge extraction module performs query and traversal operations on the assembly knowledge graph constructed in step S201 to extract the assembly sequence relationship and spatial constraint relationship between each entity.
[0067] The method for extracting the assembly sequence relationship is as follows: the knowledge extraction module follows the hierarchical composition relationship edge of the knowledge graph, starting from the top-level entity of the battery pack, and traversing down layer by layer to all sub-assembly entities and basic part entities, recording the parent entity and child entity of each entity.
[0068] Simultaneously, the knowledge extraction module queries for entity pairs with connections along the assembly connection edges, and reads the attributes of the connection edges to obtain the connection type and assembly order information. For entity pairs with assembly order dependencies, the knowledge extraction module records the identifiers of the entities installed later and earlier during the assembly process.
[0069] The spatial constraint relationships are extracted as follows: the knowledge extraction module obtains the occupancy boundaries and assembly paths of each entity in 3D space from the analysis results of the product design drawings. For entities A and B, if the assembly path of entity B passes through or contacts the occupied space area of entity A, then entity A constitutes a spatial occlusion relationship with entity B. The knowledge extraction module traverses all entity pairs, determines whether the above spatial occlusion relationship exists for each pair, and records the entity pairs with spatial occlusion and their occlusion directions.
[0070] S203 converts the assembly relationship into a disassembly priority constraint.
[0071] The knowledge extraction module, based on preset conversion rules, converts the assembly sequence and spatial constraints extracted in step S202 into disassembly priority constraints. The disassembly priority constraint is defined as a binary relation, expressed as: ; In the formula, and These are two different part entities; symbols Indicates the priority of disassembly. Indicates parts Must be in the parts Previously demolished, i.e. yes The pre-constraints for disassembly.
[0072] The transformation rules include time reversal rules, space undoing rules, and connection undoing rules.
[0073] The timing reversal rule states that for a pair of entities with a sequential assembly order, if entity A is assembled before entity B, then entity B is disassembled before entity A. Mathematically, this rule is expressed as follows: if the assembly order is A→B, then the disassembly priority constraint is... .
[0074] The principle behind this rule is that parts installed later in the assembly process are located on the outermost layer or in a more easily accessible position in space, so they should be removed first during disassembly.
[0075] The spatial release rule states that for a pair of entities that have an occlusion relationship in space, if entity C occludes the dismantling path of entity D, then entity C must be dismantled before entity D. Mathematically, this rule is expressed as follows: if entity C occludes entity D, then the dismantling priority constraint is... .
[0076] The knowledge extraction module iterates through all spatial occlusion entity pairs recorded in step S202, generating a corresponding dismantling priority constraint for each entity pair. The principle behind this rule is that occluded parts cannot be removed or operated from the battery pack before the occluding parts are removed; therefore, the occluding parts must be removed first.
[0077] The rule for detaching connections states that for pairs of entities that are connected, removing the connector takes precedence over separating the connected components. This rule handles three scenarios: Scenario 1: If entity E and entity F are connected by a thread, and bolt entity G is a connecting part, then the priority constraint for disassembly is... and That is, the bolts must be removed before the two parts are connected; Scenario 2: If entity E and entity F are connected by a snap-fit mechanism and the snap-fit structure is integrated into entity E, then the disassembly / removal order between entity E and entity F is determined by the unlocking direction of the snap-fit. If the unlocking direction is to separate entity F from entity E, then the disassembly priority constraint is... ; Scenario 3: If entity E and entity F are connected by welding, the two entities are considered as a whole when disassembling, and no mutual disassembly priority constraint is generated. The welded whole is entered into the subsequent planning as a disassembly unit.
[0078] After the above three rules are used for conversion, the knowledge extraction module converts all the assembly relationships extracted in step S202 into a set of disassembly priority constraints. This set is stored in the form of multiple binary relations and passed to the matrix construction module for subsequent processing.
[0079] Step S3 receives the set of disassembly priority constraints output by the knowledge extraction module through the matrix construction module, and constructs the overall hierarchical disassembly matrix of the battery pack and the internal hierarchical disassembly matrix of each sub-assembly, transforming the disassembly priority constraints into a structured matrix data structure.
[0080] S301, Determine the hierarchical division of the layered construction.
[0081] The matrix construction module obtains the hierarchical structure information of the assembled knowledge graph from the knowledge extraction module, and performs hierarchical division of all objects to be disassembled according to entity type. The result of the hierarchical division is two levels: the overall battery pack level and the internal level of the sub-assemblies.
[0082] The overall battery pack hierarchy includes entities that are directly subordinate to the top-level entity of the battery pack, specifically including various sub-assemblies and basic components directly installed at the battery pack hierarchy. For a power battery pack that includes module units, thermal management units, battery management units, and structural housing units, the overall battery pack hierarchy consists of the aforementioned four sub-assemblies, as well as basic components directly installed at the battery pack hierarchy, such as the battery pack top cover and battery pack bottom housing.
[0083] The overall battery pack hierarchy describes the first stage of battery pack disassembly, which involves sequentially removing each sub-assembly and external accessory from the battery pack assembly.
[0084] The internal hierarchy of a subassembly contains the basic components within that subassembly. Taking a module unit as an example, its internal hierarchy includes basic components such as the module cover, busbar, battery cells, end plates, straps, and insulating sheets. The internal hierarchy of a subassembly describes the disassembly task of further breaking down the subassembly into its independent components after it has been removed from the battery pack.
[0085] The above hierarchical division is based on the assembly process characteristics of the power battery pack. During the assembly process, sub-assemblies are assembled as independent components at the sub-assembly station, and then installed into the battery pack as a whole. The disassembly process follows the reverse order of the assembly process. The disassembly tasks within each sub-assembly are relatively independent in space and can be planned separately in terms of process.
[0086] S302, construct the overall hierarchical disassembly matrix of the battery pack.
[0087] The matrix construction module denotes the overall hierarchical breakdown matrix of the battery pack as follows: Its mathematical definition is as follows.
[0088] Assume the total number of objects in the battery pack hierarchy is . To sort these objects from 1 to Number them to form a collection of objects . For one The square array whose elements The definition of is: ; In the formula, For matrix The Middle Line 1 Column elements, This is the row index, with values ranging from 1 to... , For column indexes, the value range is 1 to... ; For the number The object, For the number The object; Representation Object Must be in the object Priority constraints on dismantling previously removed items.
[0089] The process of filling M0 in the matrix construction module is as follows: traverse all binary relations in the set of decomposition priority constraints output in step S203, and for each decomposition priority constraint... ,judge and Do they all belong to the collection of objects at the overall battery pack level? If all belong to Then the matrix elements Assign a value of 1; if at least one object does not belong to Then the constraint relationship is not This information is recorded and will be processed during the construction of the hierarchical matrix within the subassembly. After traversal, elements not assigned a value of 1 remain at 0, indicating that there is no direct disassembly priority constraint between the corresponding two objects.
[0090] matrix The rows represent objects that are demolished first, and the columns represent objects that are demolished later. The first column of matrix M0 represents the first column of the first column. The set of column indices corresponding to elements with a value of 1 in a row represents the object. Which objects' dismantling constraints are removed after dismantling? Matrix The The set of row indices corresponding to elements in the column with a value of 1, representing the row indices of the object. The set of all objects that must be demolished before demolition.
[0091] S303, construct the internal hierarchical disassembly matrix of each subassembly.
[0092] The matrix construction module constructs the internal hierarchical disassembly matrix for each sub-assembly in the battery pack. Let the power battery pack contain K sub-assemblies, numbered from 1 to K. For the k-th sub-assembly, its internal hierarchical disassembly matrix is denoted as... .
[0093] Let the number of basic parts inside the k-th sub-assembly be . These basic components from 1 to Number them to form a collection of objects . For one The square array whose elements The definition of is: ; In the formula, For matrix The Middle Line 1 Column elements; For the first A collection of basic component objects within a sub-assembly; and All The objects in; The disassembly priority constraint relationship is output in step S203.
[0094] Matrix building block population The process involves iterating through all binary relations in the set of decomposition priority constraints output in step S203, and for each decomposition priority constraint... ,judge and Do they all belong to If all belong to Then the matrix The corresponding element is assigned a value of 1; otherwise, it is not recorded. In this configuration, the internal disassembly matrix of each subassembly is constructed independently, with no cross-references between the matrices.
[0095] S304, Establish the index association of the hierarchical matrix.
[0096] After constructing the matrices at each level, the matrix construction module establishes an index mapping relationship between the overall battery pack hierarchical object and the internal hierarchical matrices of the sub-assemblies. Specifically, for the set of overall battery pack hierarchical objects... For each object in the array, if the object is a subassembly, record the subassembly number. Establish from The position index of this object is used to decompose the subassembly internal matrix. The mapping relationship is used by the subsequent sequence solving module to determine which internal matrix should be called to plan the next level of disassembly sequence after a certain sub-assembly in the overall battery pack disassembly sequence is disassembled.
[0097] The matrix building module will decompose the completed battery pack into a hierarchical matrix. All subassemblies internal hierarchical decomposition matrix to The aforementioned index mapping relationship is packaged and passed to the model building module for further processing.
[0098] Step S4 receives the hierarchical dismantling matrix and index mapping relationship from the matrix construction module through the model construction module, models the sequential planning problem of dismantling waste power batteries into a sequential decision process, defines the state space, action space and reward function of each level of dismantling task, and determines the independent planning strategy for each level.
[0099] S401 defines the state space.
[0100] For a given decomposition level, the model building module defines its state space as a multi-dimensional vector, where each dimension of the vector collectively describes the current decomposition progress and operational context.
[0101] Let the total number of objects to be disassembled at the current level be... The object number is 1 to At that moment state vector Defined as: ; In the formula, For object Demolition status indicator Representation Object Not yet demolished. Representation Object Demolished; This is the object number from the previous moment where a dismantling action was performed. It describes the continuity of actions within the context. If the current moment is the start time of the dismantling task, then... The value is 0; This represents the cumulative dismantling time from the start of the dismantling task to the current moment, expressed in seconds.
[0102] The dimension of the state vector is Dimension. State space Defined as the set of all possible state vectors. In a given... In the task of disassembling an object, each object has two possible states, and the cumulative disassembly time is... Since the variables are continuous, the state space S is a mixed space of continuous and discrete variables.
[0103] For the disassembly task at the overall battery pack level, the total number of objects... Values That is, the collection of battery pack hierarchical objects. The number of objects in the [number]. For the [number]th [object]. The task of disassembling the internal hierarchy of a sub-assembly, the total number of objects. Values That is, the collection of basic parts inside the sub-assembly. The number of objects in the process. Each level of decomposition task maintains its own state vector independently, and state vectors are not shared across levels.
[0104] S402 defines the action space.
[0105] The model building module defines the decomposition task at time 1. Action space The set of decomposable objects that satisfy the priority constraint of decomposition in the current state.
[0106] The rule for determining decomposable objects from the decomposition matrix is as follows: for objects to be decomposable... Check the disassembly matrix The Middle All elements of the column. Object At any moment The condition for demolition is that, in the first... All objects corresponding to rows in a column with a value of 1 have been demolished. Mathematically, this condition is expressed as follows: for a matrix... If the elements in the middle, Then it must satisfy Check each row in column j one by one. If all of them meet the above conditions, then object j belongs to the set of separable objects at the current moment.
[0107] Action space The formal definition is: ; In the formula, For a moment Action space; The number of the candidate demolition object; For decomposing the matrix The Middle The row index of each column; To decompose the matrix into its first... Line 1 Column elements; For object The sign indicating the demolition status.
[0108] Action space Each element in the list corresponds to a valid demolition action. When When the set is empty, it indicates that all objects have been demolished, and the demolition task at this level has terminated. Action Space The size of the object shrinks dynamically as the dismantling process progresses, and the range of objects that can be dismantled while meeting the priority constraints gradually expands as the number of dismantled objects increases.
[0109] S403, Define the reward function.
[0110] The model building module designs a reward function to evaluate the agent's performance in a given state. Select demolition target The quality of the reward function is determined by two components: the time cost of dismantling and the cost of process switching.
[0111] The time cost of dismantling reflects the object being dismantled. Standard operating time consumed. Setting the demolition target. The standard operating time is This time, obtained from assembly process documents or a standard time database, represents the time required to disassemble the object using standard tools under normal operating conditions. The shorter the disassembly time, the higher the reward value. The disassembly time cost is expressed as... ,in This is the time cost weighting coefficient.
[0112] Process changeover cost reflects the additional time loss incurred between two consecutive disassembly operations due to tool changes or changes in operating posture. Define the process changeover indication function. When the object to be demolished With the demolition object When the required tools are different or the direction of operation changes. When both require the same tools and operate in the same direction, Tool types include torque wrenches, screwdrivers, pry bars, and insulated clamps, with operating directions including forward, reverse, lateral, and vertical. The process changeover cost is expressed as follows: ,in This represents the penalty coefficient for process switching.
[0113] reward function The complete expression is: ; In the formula, For a moment The state vector; For a moment Selected demolition object number; For demolition targets Standard operating time; For process switching indication function, when and The value is 1 when the process types are different, and 0 when they are the same; This is the time cost weighting coefficient, with a value ranging from 0 to 1; This is the process switching penalty coefficient, ranging from 0 to 1, and satisfying the following conditions: .
[0114] when When the task is being broken down, there is no previous action. At this time, The reward function only includes the time cost of disassembly.
[0115] The reward function is always negative; the agent learns the disassembly sequence with the shortest total disassembly time by maximizing the cumulative reward. Weight coefficients. and The value is adjusted based on the emphasis on efficiency and process continuity in the actual dismantling task.
[0116] S404, establish a hierarchical independent planning strategy.
[0117] The model building module determines the hierarchical independent planning strategy, which specifies the planning order and result combination method of each level of decomposition sequence.
[0118] The execution process of a step-by-step independent planning strategy includes three stages.
[0119] Step 1: Calling the overall hierarchical disassembly matrix of the battery pack Based on the state space, action space, and reward function defined in S401 to S403, a decomposition sequence planning is performed on the overall battery pack hierarchy to generate the overall battery pack hierarchy decomposition sequence. . It is an ordered list, where the list elements are the numbers of the battery pack's overall hierarchical objects, and the order in which they are arranged is the disassembly order.
[0120] Step Two, according to The objects in the sequence are checked one by one. If the current object is a basic part, then it is in... The position in the table represents the moment the part is removed during the entire disassembly process, and does not trigger subsequence planning. If the current object is a sub-assembly, the internal disassembly matrix corresponding to the sub-assembly is determined based on the index mapping relationship established in step S304. The same planning method as in step one is used to perform a decomposition sequence planning for the internal hierarchy of the subassembly, generating the internal decomposition sequence of the subassembly. . It is an ordered list, where each element is the number of the basic part inside the subassembly, and the order of arrangement is the internal disassembly order.
[0121] Step 3: After the overall battery pack hierarchical disassembly sequence and the internal disassembly sequences of all sub-assemblies have been generated, the sequences are combined according to their hierarchical expansion relationships. The combination rule is as follows: in the overall battery pack hierarchical disassembly sequence... In this process, the position of each sub-assembly is determined by the moment it is disassembled, and the internal disassembly sequence of the sub-assembly is as follows. The assembly results unfold immediately following the moment the sub-assembly was removed. The resulting complete sequence covers all disassembly steps from the first object removed at the overall battery pack level to the last object removed within the last sub-assembly level.
[0122] The model building module encapsulates the decomposition matrix, state space definition, action space definition, reward function definition, and hierarchical independent planning strategy for each level, forming planning model instances for each level, which are then passed to the sequence solving module for solving.
[0123] Step S5 receives the planning model instances of each level from the model building module through the sequence solving module, and uses the D3QN algorithm to solve the decomposition sequence planning model of each level respectively. After iterative training, the decomposition sequence that satisfies the decomposition priority constraint and has a better total decomposition time is output.
[0124] S501, construct the network structure of the D3QN algorithm.
[0125] The sequence solving module constructs a D3QN algorithm instance for each level of the decomposed sequence planning task. The D3QN algorithm introduces a dual network structure and a competitive network structure based on the DQN algorithm to solve sequence decision problems in a discrete action space.
[0126] The D3QN algorithm consists of two structurally identical but parameterized deep neural networks: an evaluation network and a target network. The evaluation network calculates the action value of each action based on the current state, while the target network calculates the target action value. The initial parameters of both networks are obtained through random initialization, and the parameters of the target network are updated synchronously from the evaluation network at fixed intervals during training.
[0127] Both the evaluation network and the target network have competitive network structures. The competitive network uses the action-value function... Decomposed into state value function and dominance function Combinations: ; In the formula, It is a state vector; For action; To evaluate the parameters of shared convolutional or fully connected layers in a network; State value function The dedicated network layer parameters; For the advantage function The dedicated network layer parameters; ) is in state Next action The estimated value of the action; For state State value; In the state Next action The dominant function value; The size of the action space; It refers to any action in the action space.
[0128] The parameter set for evaluating the network is as follows The parameter set of the target network is denoted as The evaluation network updates its parameters via gradient descent in each training iteration, while the target network's parameters are copied and updated every C steps from the evaluation network's parameters.
[0129] S502, initialize the experience replay pool and training hyperparameters.
[0130] The sequence solving module initializes an experience replay pool of capacity D to store experience samples generated during the interaction between the agent and the decomposition environment. Each experience sample is a quadruple. , For a moment The state vector, For a moment The chosen demolition action, To perform the action The instant reward obtained afterward To perform the action The next state vector is then transitioned to.
[0131] The sequence solving module sets the training hyperparameters, including the total number of training epochs, maximum number of steps per epoch, learning rate, discount factor, batch size, target network update cycle, experience replay pool capacity, initial exploration rate, final exploration rate, and exploration rate decay factor. The exploration rate decays exponentially, with the current exploration rate at the [missing value]. The value for each training round is: ; In the formula, The exploration rate for the current training round; This is the initial exploration rate; To terminate the exploration rate; To explore the rate of decay factor; This is the number of the current training round.
[0132] S503, execute the training process.
[0133] The sequence solving module performs a training process for each level of the decomposed sequence planning task. The training process includes the following iterative steps.
[0134] Step 1: Reset the environment and state. At the beginning of each training round, the sequence solver module removes the state markers from all objects at the current level. All values are set to 0, and the total disassembly time is recorded. Set to 0, previous action Set to 0 to obtain the initial state vector. After the environment is reset, all objects remain in an undemolished state.
[0135] Step two, select the decomposition action. The sequence solving module at each time step... According to the current state and exploration rate ,use Greedy strategy selects action . The rule of a greedy strategy is to use probability... From the action space Randomly select an action, with probability Select the action with the highest action value. The action space At is determined by the set of detachable objects defined in step S402.
[0136] Step 3: Execute actions and state transitions. The sequence solver module executes the selected actions in the environment. Update the disassembly status. The status update operation includes changing the action... The demolition status flag of the corresponding object is set from 0 to 1, and the record is entered. For the new previous action , will the action Standard operating time Added to total dismantling time The next state vector is obtained after the state update. .
[0137] Step 4: Calculate the immediate reward. The sequence solving module calculates the action to be executed based on the reward function defined in step S403. Instant rewards received The calculation method is as follows .
[0138] Step 5: Store the experience samples. The sequence solving module will store the experience samples generated in this interaction. The samples are stored in the experience replay pool. When the number of samples in the experience replay pool reaches its capacity limit D, the new sample overwrites the oldest sample stored.
[0139] Step Six: Sampling and Network Update. After the sequence solving module has reached the batch size B in the experience replay pool, it randomly samples B samples from the experience replay pool for network updates. For the sampled samples... Calculate the value of the target action : ; In the formula, For the sample The value of the target action; For the sample Instant rewards; Discount factor; For the sample The next state vector; For state The corresponding action space; For action space Candidate actions in; The action value function is calculated by the evaluation network; In the state The optimal action is selected by the evaluation network. The action value of the optimal action is calculated by the target network.
[0140] The sequence solving module uses the mean squared error as the loss function to calculate the loss value for B samples: ; In the formula, Mean squared error loss value; This refers to the batch sampling size. For the sample The value of the target action; To evaluate the network's response to samples In state Next action The estimated value of the action.
[0141] The sequence solving module minimizes the loss function using the gradient descent algorithm. Update the parameter set of the evaluation network. .
[0142] Step 7: Update the target network. Every C training iterations completed, the sequence solving module will evaluate the network parameter set. Copy the value to the target network parameter group .
[0143] Step 8, Round Termination Judgment. After each action is executed, the sequence solving module determines the action space. Is it an empty set? If An empty set indicates that all objects at the current level have been demolished, and this training round terminates. Non-empty set and the number of moves in this round has not been reached. Then return to step two to continue selecting actions. If the number of steps in this round reaches [a certain threshold], then [the next action will be selected]. and If it is still not an empty set, then the current round will be forcibly terminated.
[0144] Step nine, round iteration. The sequence solving module repeats steps one through eight, for a total of [number missing]. There are 10 training rounds. After each training round, the cumulative reward value for that round is recorded. Training can be terminated early if the cumulative reward value fluctuates less than a preset convergence threshold for several consecutive rounds.
[0145] S504, Model Inference and Sequence Generation.
[0146] After training, the sequence solving module uses the final evaluation network parameter set. Perform reasoning to generate a breakdown sequence for the corresponding level. The exploration rate during the reasoning process... Setting it to 0 means that at each step, the action with the greatest action value is greedily selected.
[0147] The sequence solving module starts from the initial state. At each step, it calculates the action value of all legal actions based on the current state, selects the legal action with the highest action value, executes the action, and transitions to the next state. This process is repeated until the action space is empty. The action object numbers executed in each step are arranged in sequence to form a decomposition sequence.
[0148] The sequence solving module performs the above training and inference processes on the planning models of the overall battery pack hierarchy and the internal hierarchy of all subassemblies, respectively, to generate the overall battery pack hierarchy disassembly sequence. and the internal disassembly sequence of all subassemblies to The data is then passed to the sequence output module for combined output.
[0149] Step S6 receives the disassembly sequences of each level generated by the sequence solving module through the sequence output module, and combines the overall disassembly sequence of the battery pack with the disassembly sequence of each sub-assembly according to the hierarchical expansion relationship to output a complete disassembly sequence.
[0150] S601, obtain the disassembly sequence of each level.
[0151] The sequence output module receives the following sequence data from the sequence solving module: overall battery pack hierarchical disassembly sequence. and the internal disassembly sequence of all subassemblies to .in, This represents the total number of subassemblies.
[0152] Battery Pack Overall Layer Disassembly Sequence It is an ordered list, where each element is a number representing a hierarchical object of the battery pack, and the number corresponds to the object set in step S302. The object number in the file. The order in which the elements are arranged determines the order of operations in the first stage of battery pack disassembly; the elements at the beginning are disassembled first, and the elements at the end are disassembled later.
[0153] For the Sub-assemblies, their internal disassembly sequence It is an ordered list, where each element is the number of a basic part within the subassembly, and the number corresponds to the object set in step S303. The basic part number in the system. The order in which the elements are arranged is the order in which the subassembly is further decomposed after being separated from the battery pack.
[0154] S602, Determine the sequence combination rules.
[0155] The sequence output module combines the disassembly sequences of each level into a complete, step-by-step disassembly sequence according to the hierarchical expansion rules. The hierarchical expansion rules are based on the overall hierarchical disassembly sequence of the battery pack. The main sequence, in For each subassembly object that appears, immediately following the demodulation time of that subassembly object, the internal disassembly sequence of that subassembly is unfolded. .
[0156] The execution logic of this rule is as follows: The sequence output module iterates through... For each element in the sequence, the following judgment and operation are performed on the currently traversed element: If the current element is a basic part object, the basic part object is directly appended to the end of the output sequence. If the current element is a sub-assembly object, the sub-assembly object is appended to the end of the output sequence, marking that the sub-assembly is completely removed from the battery pack at this moment; subsequently, the sequence output module finds the internal disassembly sequence corresponding to the sub-assembly according to the index mapping relationship established in step S304. ,Will All elements in the are arranged according to The original sequence is appended sequentially to the end of the output sequence. After appending, the sequence output module continues to iterate. The next element in, until All elements have been traversed.
[0157] After the above rule processing, the overall battery pack disassembly operation and the disassembly operation within each sub-assembly are integrated into a complete linear disassembly sequence. Each element in this linear disassembly sequence corresponds to a specific disassembly operation step, and the order of the operation steps is the actual execution order of the disassembly operation.
[0158] S603 generates a complete step-by-step disassembly sequence.
[0159] The sequence output module denotes the complete disassembled sequence obtained after combination as follows: . It is an ordered list whose length is equal to the total number of objects in the battery pack hierarchy plus the sum of the number of basic parts within all subassemblies. Each element contains two fields: object number and object name. The object name is the entity name obtained from the assembled knowledge graph constructed in step S201.
[0160] Sequence output module The disassembly object information for each step is supplemented with annotations. For each disassembly step, the supplementary annotations include the hierarchical identifier of the step, the type of disassembly tool required, and the standard operating time. The hierarchical identifier indicates whether the step belongs to the overall battery pack hierarchical disassembly or the hierarchical disassembly within a certain sub-assembly. The disassembly tool type and standard operating time are obtained from the process data introduced in step S403.
[0161] S604, outputs the disassembly sequence results.
[0162] The sequence output module will completely decompose the sequence step by step. Output according to the preset format. The output format consists of three parts.
[0163] The first part is the overall disassembly sequence of the battery pack. It is presented in two forms: a list of numbers and a list of object names, with each list item corresponding to a disassembly step.
[0164] The second part is the internal disassembly sequence of each sub-assembly. to They are presented as a list of numbers and a list of object names, respectively, with the internal disassembly sequence of each subassembly listed separately.
[0165] The third part is the complete step-by-step decomposition sequence after assembly. The data is presented in a table format with five fields: step number, object name, hierarchy, disassembly tool, and standard operating time. Each row in the table corresponds to a disassembly step, with step numbers starting from 1 and increasing sequentially. The steps are arranged from top to bottom in the order of disassembly. The hierarchy field indicates whether the step belongs to the overall battery pack hierarchy or the internal hierarchy of a specific subassembly. The disassembly tool field indicates the name of the tool used to perform the step. The standard operating time field indicates the standard time required to perform the step.
[0166] The sequence output module outputs the above three parts as the final planning result, providing executable step-by-step dismantling operation guidance for the actual dismantling of waste power batteries.
[0167] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for sequential planning of the dismantling of waste power batteries, characterized in that, Includes the following steps: Obtain the original assembly data of power battery products; Based on the original assembly data, a knowledge graph of power battery assembly is constructed, the assembly sequence relationship and spatial constraint relationship between each node are extracted, and the assembly sequence relationship and spatial constraint relationship are converted into disassembly priority constraint. Based on the aforementioned disassembly priority constraint, the overall hierarchical disassembly matrix of the battery pack and the internal hierarchical disassembly matrix of each sub-assembly are constructed respectively. The sequential planning problem is modeled as a sequential decision process by decomposing it step by step. The state space, action space and reward function of each decomposition level are defined, and the planning model of each level is established by adopting a step-by-step independent planning strategy. The D3QN algorithm is used to solve the planning model at each level and output the decomposition sequence of each level. The decomposition sequences of each level are combined according to the hierarchical expansion relationship to output the complete step-by-step decomposition sequence.
2. The method for sequential planning of the dismantling of waste power batteries according to claim 1, characterized in that, The construction of the power battery assembly knowledge graph specifically includes: Entities are extracted from the original assembly data, including the battery pack as the top-level entity, the sub-assemblies as the middle-level entities, and the base parts as the bottom-level entities. Extract the relationships between entities, including hierarchical composition relationships that connect upper and lower level entities, as well as assembly connection relationships with connection type and assembly order attributes; The extracted entity nodes and relation edges are stored in a graph database to obtain a knowledge graph of power battery assembly.
3. The method for sequential planning of the dismantling of waste power batteries according to claim 1, characterized in that, The process of converting the assembly sequence relationship and the spatial constraint relationship into disassembly priority constraints specifically includes: Based on the timing reversal rule, the order of assembly is reversed to a disassembly priority constraint of disassembling the last assembly first. Based on the spatial release rule, the spatial occlusion relationship between the occluded part and the occluding part is transformed into a disassembly priority constraint that prioritizes the removal of the occluding part. Based on the connection release rule, the connection between parts is converted into a disassembly priority constraint where the connector is removed before the connected part.
4. The method for sequential planning of the dismantling of waste power batteries according to claim 1, characterized in that, The construction of the overall hierarchical disassembly matrix of the battery pack and the internal hierarchical disassembly matrix of each sub-assembly specifically includes: Based on the disassembly priority constraints, a battery pack overall hierarchical disassembly matrix is constructed. The matrix elements of the battery pack overall hierarchical disassembly matrix represent whether there are disassembly priority constraints between battery pack overall hierarchical objects. Construct a hierarchical disassembly matrix for each sub-assembly. The matrix elements of the hierarchical disassembly matrix for each sub-assembly represent whether there are disassembly priority constraints between the basic parts inside the corresponding sub-assembly. Establish an index mapping relationship between the overall battery pack hierarchical object and the corresponding sub-assembly internal hierarchical disassembly matrix.
5. The method for sequential planning of the dismantling of waste power batteries according to claim 1, characterized in that, The definition of the state space, action space, and reward function for each decomposition level specifically includes: The state space is defined as a multi-dimensional vector containing the demolition state identifier of each object, the object number that performed the demolition action in the previous time step, and the cumulative demolition time. Define the action space as a set of decomposable objects that satisfy the priority constraints of all decompositions in the current state; By combining the time cost corresponding to the standard operation time of the demolition object and the process switching cost between continuous dismantling operations, a reward function for evaluating the quality of action selection is calculated.
6. The method for sequential planning of the dismantling of waste power batteries according to claim 1, characterized in that, The method of establishing planning models at each level using a hierarchical independent planning strategy specifically includes: The overall hierarchy of the battery pack is planned to generate a disassembly sequence for the overall battery pack hierarchy; Determine the type of the current object according to the object order in the overall hierarchical disassembly sequence of the battery pack; When the object is a subassembly, the corresponding subassembly internal hierarchy disassembly matrix is called according to the established index mapping relationship, and the internal hierarchy of the subassembly is planned to generate the corresponding subassembly internal disassembly sequence.
7. The method for sequential planning of the dismantling of waste power batteries according to claim 1, characterized in that, The method of using the D3QN algorithm to solve the planning model at each level specifically includes: For each level, an algorithm model is constructed that includes an evaluation network and a target network, and the action value function is decomposed into a combination of a state value function and an advantage function. Store state experience samples of interaction with the environment in the experience replay pool; The target action value is calculated based on samples taken from the experience replay pool. The evaluation network parameters are updated by minimizing the mean square error loss between the target action value and the estimated action value, and the decomposition sequence of each level is output.
8. The method for sequential planning of the dismantling of waste power batteries according to claim 7, characterized in that, The storage of state experience samples of interaction with the environment in the experience replay pool specifically includes: Select the current demolition action based on the current state vector; Execute the current demolition action, update the demolition status flag, the object number of the demolition action performed in the previous moment, and the cumulative demolition time, and obtain the next state vector; Calculate the immediate reward using the aforementioned reward function; Data containing the current state vector, the current demolition action, the immediate reward, and the next state vector is stored as an experience sample in the experience replay pool.
9. The method for sequential planning of the dismantling of waste power batteries according to claim 1, characterized in that, The process of combining the decomposition sequences of each level according to their hierarchical expansion relationship specifically includes: The disassembly sequence of the battery pack as a whole is used as the main sequence for traversal; When a basic component object is encountered during iteration, it is directly appended to the end of the output sequence. When a subassembly object is encountered, the subassembly object is appended to the end of the output sequence. The internal decomposition sequence corresponding to the subassembly is then searched, and all elements in the internal decomposition sequence are appended to the end of the output sequence in their original order to obtain a complete sequence. After the combination yields the complete sequence, it also includes: Supplement the information of the disassembly objects in each step of the complete hierarchical disassembly sequence with additional annotations; The supplementary annotations include the level identifier of each disassembly step, the type of disassembly tools required, and the standard operating time. Output the complete hierarchical disassembly sequence, including step number, object name, hierarchy, disassembly tool, and standard operation time fields, according to the preset format.
10. A sequential planning system for the step-by-step dismantling of used power batteries, characterized in that, It includes a knowledge input module, a knowledge extraction module, a matrix construction module, a model construction module, a sequence solving module, and a sequence output module; The knowledge input module obtains the original assembly data of power battery products; The knowledge extraction module constructs a power battery assembly knowledge graph based on the original assembly data, extracts the assembly sequence relationship and spatial constraint relationship between each node, and converts the assembly sequence relationship and spatial constraint relationship into disassembly priority constraint. The matrix construction module constructs the overall hierarchical disassembly matrix of the battery pack and the internal hierarchical disassembly matrix of each sub-assembly based on the disassembly priority constraint. The model building module models the hierarchical decomposition of the sequential planning problem as a sequential decision process, defines the state space, action space and reward function of each decomposition level, and establishes the planning model of each level using a hierarchical independent planning strategy; The sequence solving module uses the D3QN algorithm to solve the planning model at each level and outputs the decomposition sequence at each level. The sequence output module combines the decomposition sequences of each level according to the hierarchical expansion relationship and outputs the complete step-by-step decomposition sequence.
